Next Article in Journal
Historic Low Wall Detection via Topographic Parameter Images Derived from Fine-Resolution DEM
Previous Article in Journal
Hybrid Spatial Data Model for Indoor Space: Combined Topology and Grid
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Classification of Hyperspectral Images Using Kernel Fully Constrained Least Squares †

1
The Key Laboratory of Advanced Process Control for Light Industry (Ministry of Education), School of Internet of Things Engineering, Jiangnan University, Wuxi 214122, China
2
School of Computer Science, Nanjing University of Science and Technology, Nanjing 210094, China
*
Author to whom correspondence should be addressed.
This paper is an extended version of our paper published in 2017 IEEE International Geoscience and Remote Sensing Symposium.
ISPRS Int. J. Geo-Inf. 2017, 6(11), 344; https://doi.org/10.3390/ijgi6110344
Submission received: 19 September 2017 / Revised: 26 October 2017 / Accepted: 3 November 2017 / Published: 6 November 2017

Abstract

:
As a widely used classifier, sparse representation classification (SRC) has shown its good performance for hyperspectral image classification. Recent works have highlighted that it is the collaborative representation mechanism under SRC that makes SRC a highly effective technique for classification purposes. If the dimensionality and the discrimination capacity of a test pixel is high, other norms (e.g., 2 -norm) can be used to regularize the coding coefficients, except for the sparsity 1 -norm. In this paper, we show that in the kernel space the nonnegative constraint can also play the same role, and thus suggest the investigation of kernel fully constrained least squares (KFCLS) for hyperspectral image classification. Furthermore, in order to improve the classification performance of KFCLS by incorporating spatial-spectral information, we investigate two kinds of spatial-spectral methods using two regularization strategies: (1) the coefficient-level regularization strategy, and (2) the class-level regularization strategy. Experimental results conducted on four real hyperspectral images demonstrate the effectiveness of the proposed KFCLS, and show which way to incorporate spatial-spectral information efficiently in the regularization framework.

1. Introduction

Sparse representation classification (SRC) has been widely used in many applications, such as pattern recognition [1,2], visual classification [3,4], and hyperspectral image classification [5,6,7,8,9,10,11]. Unlike the common classifiers (e.g., support vector machines (SVMs) [12] and multinomial logistic regression [13]), SRC is not a learning-based classifier, which first represents a test sample as the sparse linear combination of all training samples and then directly assigns a class label to the test sample by evaluating which class leads to the minimum reconstruction error. Although the usage of sparsity prior in the literature often leads to robust classification performance, recent works [14,15] have shown that it is the collaborative representation (CR) mechanism under SRC (i.e., representing a test sample collaboratively with training samples from all classes) that makes SRC a highly effective technique for classification purposes. Moreover, if the dimensionality and the discrimination capacity of a sample is high, other regularization terms such as 2 -norm can play the same role as the sparsity 1 -norm. Several approaches have demonstrated the effectiveness of the classification method using 2 -norm in many applications [14,16], including hyperspectral image classification [17,18]. For the sake of simplicity, the method using 2 -norm is referred to as collaborative representation classification (CRC), and both SRC and CRC are referred to as CR-based classification methods.
Although CR-based classification methods can get good performance, it is difficult to use them to classify data that is not linearly separable. Moreover, for hyperspectral image classification, the discriminability of a pixel is generally low, owing to the presence of redundant spectral bands, although its dimensionality is high. As the pixel-wise classification results reported in [17,18], SRC often produces superior performance compared to CRC. Some approaches have considered using the kernel method that is widely used in SVM classification [12,19] to mitigate these problems [20,21,22], since in the kernel feature space the dimensionality of a sample is very high, and its discriminability is generally enhanced [23]. In hyperspectral image classification, kernel CR (KCR)-based classification has shown an improvement over CR-based classification [21,24,25], and kernel CRC (KCRC) exhibits competitive advantages in terms of classification accuracy and computational cost when compared with kernel SRC (KSRC) [23].
In the development of CR-based classification methods, more attention is paid to the selection of norms. However, for both SRC or CRC they belong to the regularized least squares. That is to say, the improvement brought by the norms can also be achieved by other regularization terms or constraint terms. Among the numerous terms, the nonnegative constraint is an effective one that is very common in many other techniques and applications, such as nonnegative matrix factorization [26] and spectral unmixing [27]. Moreover, the nonnegative constraint can also bring the sparsity of the coding coefficients [28,29]. Accordingly, we consider exploiting kernel nonnegative constrained least squares (KNLS) for hyperspectral image classification. Since in the kernel feature space the dimensionality and the discrimination capacity of a pixel is high, the nonnegative constraint may play a role of CR. Considering that the nonnegative coding coefficient reflects the similarity between a test pixel and the related training pixel, we suggest to provide the posterior probabilistic outputs by enforcing the summation of the nonnegative coding coefficients of each pixel to be one, and thus investigate kernel fully constrained least squares (KFCLS) [30,31] for hyperspectral image classification.
The investigated KFCLS is a pixel-wise classifier that treats hyperspectral data as an unordered list of feature vectors but not as images. In order to handle the coarse classification maps brought by a pixel-wise classifier, previous methods have considered incorporating spatial-contextual information during the classification process [32]. According to the relationship between the pixel-wise classification process and the fusion of spatial-spectral information, these methods can be roughly divided into three categories:
(1)
The first category can be treated as pre-processing methods. These methods usually extract the spatial features first, and then incorporate both the spatial and spectral features into a pixel-wise classifier. For instance, in [33] a composite kernel framework is proposed to combine the spatial and spectral features first and subsequently embed into SVM for classification purposes. In [34], the authors extract multiple types of spatial features from both linear and nonlinear transformations first, and then integrate them via multinomial logistic regression. In [35], a convolutional neural network is utilized to extract deep features from high levels of the image data and the final classification is done by using SRC.
(2)
The second category can be treated as post-processing methods. These methods usually perform a pixel-wise classifier first, and then refine the pixel-wise results by incorporating spatial information. For instance, in [23,36,37,38] the class conditional probability density functions are first estimated using a probabilistic pixel-wise classifier, and then refined by using some regularization models to incorporate the spatial information. In [39,40], the original hyperspectral image is first classified per pixel and simultaneously segmented into several adaptive neighborhoods, and then a decision fusion mechanism is undertaken within the pixel-wise classification results of these neighborhoods. In [41], KSRC is first used to get the coding coefficients of the original hyperspectral image, and then the coding coefficients are refined by incorporating the spatial information for the final classification.
(3)
The last category can be treated as co-processing methods, which jointly integrate the pixel-wise classification process and the fusion of spatial-spectral information. For CR-based classification, the related methods usually assign a neighborhood/window to each pixel, and perform the representation of a test pixel jointly by its neighbouring pixels [21,42,43,44,45]. In addition, there are other methods that consider incorporating the spatial information by appending a spatial-spectral term to the coding model of CR-based classification [11,25,46].
Notably, regularization is an important technique for CR-based classification, since all CR-based classification methods are built by regularization technique. As a widely used technique in mathematical and image processing problems [47], the regularization technique is very suitable for the integration of different prior knowledge owing to its flexibility and availability. This paper considers incorporating the spatial information into KFCLS using regularization technique. For this issue, we consider both the co-processing and post-processing methods, and propose a weighted H1-norm [48] for the description of spatial information. Furthermore, we investigate two regularization strategies to integrate the spatial and spectral information. One is the coefficient-level regularization strategy that incorporates the spatial information by enforcing or refining the coding coefficients, and the other is the class-level regularization strategy that handles the posterior probabilistic outputs.
The remainder of this paper is organized as follows. Section 2 briefly introduces two instantiations of KCR-based classification methods (i.e., KSRC and KCRC). In Section 3, we first present the proposed KFCLS for hyperspectral image classification, and then introduce the co-processing and post-processing methods for KFCLS using two regularization strategies. The effectiveness of the proposed KFCLS and and the suggested way of incorporating spatial-spectral information are demonstrated in Section 4 by conducting experiments on four real hyperspectral images. Finally, Section 5 concludes this paper.

2. KCR-Based Classification

In this section, we briefly review the general model of KCR-based classification and subsequently introduce its two instantiations. Given a hyperspectral image, every pixel in it can be interpreted as an L-dimensional column vector with L being the number of spectral bands. Suppose the given hyperspectral image includes C classes, and there exists a feature mapping function ϕ which maps a test pixel x R L and J training pixels A = [ a 1 , a 2 , , a J ] R L × J to the high-dimensional feature space: x ϕ ( x ) , A Φ ( A ) = [ ϕ ( a 1 ) , ϕ ( a 2 ) , , ϕ ( a J ) ] . For a mapped pixel ϕ ( x ) , KCR-based classification supposes that it can be collaboratively represented as the linear combination of all mapped training pixels; i.e.,
ϕ ( x ) Φ ( A ) s ,
where s R J is an unknown coding coefficient vector of ϕ ( x ) . To recover a coding coefficient vector s from ϕ ( x ) and Φ ( A ) stably, the regularization method is the best choice, and the corresponding optimization problem can be written as follows:
min s 1 2 ϕ ( x ) Φ ( A ) s 2 2 + λ s q ,
where λ > 0 is a regularization parameter, and q = 0 , 1, or 2.
Using different q will lead to different instantiations of KCR-based classification. KSRC and KCRC are two instantiations, where q is respectively set to 1 and 2. The corresponding optimization problems can be respectively written as follows:
KSRC : min s 1 2 ϕ ( x ) Φ ( A ) s 2 2 + λ s 1 and KCRC : min s 1 2 ϕ ( x ) Φ ( A ) s 2 2 + λ 2 s 2 2 .
After solving the above optimization problems, the obtained s is used for the final classification. For KSRC, the class label y of x is determined via the minimal residual between ϕ ( x ) and its approximation from the mapped training pixels of each class, and the classification rule can be written as follows:
dist : y = arg min c = 1 , 2 , , C ϕ ( x ) Φ ( A ) δ c ( s ) 2 2 ,
where δ c ( · ) is the characteristic function that selects coefficients related to the cth class and makes the rest zero. For KCRC, it considers the discriminative information brought by s , and modifies the classification rule as:
y = arg min c = 1 , 2 , , C ϕ ( x ) Φ ( A ) δ c ( s ) 2 2 / δ c ( s ) 2 2 .
Notably, all ϕ mappings used in kernel methods occur in the form of inner products. For every two pixels x i and x j , we can define a kernel function as:
K = ϕ ( x i ) , ϕ ( x j ) ,
where · , · represents the inner product. In this paper, only the radial basis function (RBF) kernel ( K ( x i , x j ) = exp ( γ x i x j 2 2 ) , γ > 0 ) is considered, owing to its simplicity and empirically observed good performance [12,19,21,49]. After defining the kernel function K , the optimization problems of KSRC and KCRC can be rewritten as:
KSRC : min s 1 2 s T Q s s T b + λ s 1 and KCRC : min s 1 2 s T Q s s T b + λ 2 s 2 2 ,
where the constant terms are dropped, Q = Φ ( A ) , Φ ( A ) R J × J is a positive semi-definite matrix with entry Q i j = K ( a i , a j ) and the mapped test vector b = Φ ( A ) , ϕ ( x ) = [ K ( a 1 , x ) , , K ( a J , x ) ] R J . Similarly, the classification rules of KSRC and KCRC can be rewritten as:
y = arg min c = 1 , 2 , , C δ c T ( s ) Q δ c ( s ) 2 δ c T ( s ) b
y = arg min c = 1 , 2 , , C δ c T ( s ) Q δ c ( s ) 2 δ c T ( s ) b + 1 / δ c T ( s ) δ c ( s )
The optimization problem of KSRC is convex but not smooth. For this type of problem, several algorithms proposed in the sparse representation and compressive sensing community can be adopted to solve it [50,51,52]. In this paper, an alternating direction method of multipliers (ADMM) algorithm [53] is adopted owing to its flexibility and availability, and the details can be seen in [49]. As for the optimization problem of KCRC, it is convex and smooth, and an analytical solution can be derived (see [23]).

3. Proposed Approach

3.1. Pixel-Wise Classification via KFCLS

3.1.1. Problem Formulation

Works in [14,15] point out that if the dimensionality and discriminability of a test sample is high, the estimated coefficient vector will be naturally sparse and concentrate on the training samples whose class labels are the same as that of the test sample, regardless of whether the 1 -norm or 2 -norm is used. Since in the kernel feature space the dimensionality of a test sample is very high and its discriminability is enhanced, KCRC can get the same performance as KSRC [23]. Considering these issues, one may wonder whether other constraint terms can play the same role, except the q -norm regularization terms. Notably, in the coefficient vector s , each entry can be treated as the similarity between the corresponding training pixel and the test pixel. If the test pixel is similar to some training pixels, large values will be assigned to the corresponding entries of s ; otherwise, small values (may be negative) will be assigned. It is natural to enforce the similarity to be nonnegative. Moreover, the nonnegative constraint can promote the sparsity of the coefficient vector [28,29]. For this reason, we consider the KNLS problem, which is defined as follows:
KNLS : min s 1 2 s T Q s s T b subject to s 0 J ,
where 0 J R J is a zero vector with all entries being 0, and the symbol ⪰ denotes component-wise inequality; i.e., s 0 J means entry s j 0 for j = 1 , 2 , , J . Since s is nonnegative and reflects the similarity, it can be regraded as a probability distribution if we enforce the summation of its entries to be one [54]. Accordingly, we obtain KFCLS and the corresponding problem can be written as follows:
KFCLS : min s 1 2 s T Q s s T b subject to s 0 J , 1 J T s = 1 ,
where 1 J R J is a vector with all entries being 1. Figure 1 shows a comparison of the coefficient vectors obtained by KNLS, KFCLS, and KSRC. It can be observed that the coefficients of KNLS and KFCLS are almost as sparse as those of KSRC. Although the number of training pixels of each class may be unequal, the summation of the entries of δ c ( s ) can reflect the similarity between the cth class and the test pixel [23]. Figure 2 shows the summation of the entries of each δ c ( s ) . It is apparent that the summation value of the true class label is predominant. Moreover, the outputs of a classifier should be calibrated posterior probabilities to facilitate the subsequent processing, which are very useful in spatial-spectral classification [23,36,37]. With the aforementioned observation and motivation in mind, we have designed a posterior probability in this context as follows:
p ( y = c | x ) = ( T s ) c ,
where ( · ) c denotes the cth entry of a vector, and the summation matrix T R C × J is defined by
T c j = 1 if class ( a j ) = c 0 else , c , j .
With the definition of the posterior probability, the classification rule of KFCLS can be written as follows:
prob : y = arg max c = 1 , 2 , , C p ( y = c | x ) .
As for KNLS, we use the classification rule (4) in this paper. Notably, the classification rule (4) is also suitable for KFCLS.

3.1.2. Optimization Algorithm

In this paper, ADMM is adopted to solve the optimization problems (10) and (11). For KFCLS, the optimization problem (11) can be rewritten as the following equivalent form:
min s 1 2 s T Q s s T b + ι R + J ( s ) + ι 1 ( 1 J T s ) ,
where ι S is the indicator function of the set S (i.e., ι S ( x ) = 0 if x S and ι S ( x ) = if x S ). By introducing a variable v R J , the optimization problem (15) can be rewritten as follows:
min s , v 1 2 s T Q s s T b + ι R + J ( v ) + ι 1 ( 1 J T s ) subject to s = v .
The augmented Lagrangian function of (16) can be written as follows:
L ( s , v , d ) = 1 2 s T Q s s T b + ι R + J ( v ) + ι 1 ( 1 J T s ) + μ 2 s v d 2 2 ,
where μ > 0 is the penalty parameter and d R J is an auxiliary variable. The ADMM iteration procedure can be written as:
s t = arg min s 1 2 s T Q s s T b + ι 1 ( 1 J T s ) + μ 2 s v t 1 d t 1 2 2 v t = arg min v ι R + J ( v ) + μ 2 s t v d t 1 2 2 d t = d t 1 ( s t v t )
where t > 0 is the iteration number. The first step of (18) is to solve the s -subproblem, and the solution can be derived as:
s t = P 1 F 1 ( b + μ ( v t 1 + d t 1 ) ,
where F = Q + μ I with I being the identity matrix, and the projection operator P 1 ( s ) = s F 1 1 J ( 1 J T F 1 1 J ) 1 ( 1 J T s 1 ) . The second step of (18) is to solve the v -subproblem, which is the well-known proximal operator [55]:
v t = max ( s t d t 1 , 0 ) ,
where max ( · , 0 ) is used to set the negative components to zero and keep the nonnegative components unchanged. The last step of (18) is used to update the auxiliary variable. The algorithm of KFCLS is detailed as follows.
  • Input: A training dictionary A R L × J , and a hyperspectral pixel x R L .
  • Select the parameter γ for the RBF kernel and compute the matrix Q and the vector b .
  • Set t = 1 , choose μ , s 1 , v 1 , d 1 .
  • Repeat
  • Compute s t , v t , d t using (18).
  • t = t + 1 .
  • Until some stopping criterion is satisfied.
  • Output: The estimated label of x using (14) or (4).
As for KNLS, its ADMM iteration procedure is almost as same as that of KFCLS, where we do not need the projection operator P 1 in (19).

3.2. Spatial-Spectral Classification

The suggested KFCLS is just a pixel-wise classifier that does not treat hyperspectral data as images but as an unordered list of pixels. In order to incorporate the spatial-spectral information, several methods have been proposed as discussed in Section 1. Among these methods, the regularization strategy is an important one for CR-based classification, since CR-based classification is also a group of regularization methods. In this section, we show the incorporation of spatial-spectral information for KFCLS using both the co-processing and post-processing methods, and consider two regularization strategies to combine the spatial-spectral information.

3.2.1. Problem Formulation

Suppose that a hyperspectral image is composed of a set of I pixels X = [ x 1 , x 2 , , x I ] R L × I . Correspondingly, we get the coefficient matrix S = [ s 1 , s 2 , , s I ] R J × I , the probability matrix P = [ p 1 , p 2 , , p I ] R C × I with ( p i ) c = p ( y = c | x i ) , and the mapped test matrix B = [ b 1 , b 2 , , b I ] R J × I . Then, the unconstrained optimization problem (15) for X can be written as:
min S 1 2 Tr ( S T Q S ) Tr ( S T B ) + ι R + J × I ( S ) + ι 1 I T ( 1 J T S ) ,
where Tr ( · ) denotes the trace of a matrix. In this paper, the spatial relationship between two adjacent pixels x i and x j is modeled by the similarity defined as
W i j = exp ( β | | x ¯ i x ¯ j | | 2 ) + ϵ , β > 0
where ϵ = 10 6 is a small positive constant and x ¯ i and x ¯ j are the pixels of the first three principle components of the hyperspectral image X . For each pixel x i , its neighborhood N i is built by its eight spatially-adjacent neighbors, and W i j is set to 0 if x j does not belong to N i .

3.2.2. Co-Processing Methods

The spatial arrangement of the coefficient matrix S is associated with that of the hyperspectral image X . That is to say, the spatial relationship between every two pixels x i and x j is also suitable for that of the coefficient vectors s i and s j . It is natural to integrate the spatial information of X by enforcing S . If x i is similar to x j (i.e., W i j is relatively large), s i and s j should be close to each other, and vice versa. In this paper, the weighted H1-norm that is convex and smooth is adopted to describe the aforementioned relationship and the joint regularization model (JRM) can be written as follows:
JRM : min S 1 2 Tr ( S T Q S ) Tr ( S T B ) + ι R + J × I ( S ) + ι 1 I T ( 1 J T S ) + λ 4 w S F 2 ,
where λ > 0 , · F denotes the Frobenius norm, and w S F 2 is the weighted H1-norm of S with w s i = { W i j ( s i s j ) | j N i } . We may note that the spatial arrangement of the probability matrix P is also associated with that of the hyperspectral image X . That is to say, we can integrate the spatial information of X by enforcing P (i.e., TS ). In view of this, we propose the following class-level JRM (CJRM):
CJRM : min S 1 2 Tr ( S T Q S ) Tr ( S T B ) + ι R + J × I ( S ) + ι 1 I T ( 1 J T S ) + λ 4 w P F 2 subject to P = TS .
Since the objective solutions of JRM and CJRM are the coefficient matrix S and the columns of which sum to one, both the classification rules (4) and (14) are suitable for the final classification.

3.2.3. Post-Processing Methods

For the post-processing methods, the procedures of the pixel-wise classification and the integration of spatial-spectral information are performed separately. Generally, the incorporation of spatial-spectral information can be done by refining the coefficient matrix S [41] or the probability matrix P [23,36,37,56]. For this issue, we propose the corresponding post-processing regularization model (PRM) and class-level PRM (CPRM) for KFCLS, which are defined as follows:
PRM : min V 1 2 S V F 2 + λ 4 w V F 2
CPRM : min U 1 2 P U F 2 + λ 4 w U F 2
Notably, we can verify that the columns of solutions V and U sum to one with reference to (34). The objective solution of PRM is the refined coefficient matrix V , and thus both the classification rules (4) and (14) are suitable for the final classification; whereas the objective solution of CPRM is the refined probability matrix U , and thus only the classification rule (14) is suitable for classification purposes. Furthermore, we can prove that PRM using the classification rule (14) is equivalent to CPRM using the classification rule (14) with reference to (34). Accordingly, PRM using the classification rule (14) is dropped in the experiments.

3.2.4. Optimization Algorithm

In this paper, the optimization problems (23) and (24) are solved by ADMM. For CJRM, the optimization problem (24) can be rewritten as the following formulation by introducing a variable V R J × I :
min S , V , P 1 2 Tr ( S T Q S ) Tr ( S T B ) + ι R + J × I ( V ) + ι 1 I T ( 1 J T S ) + λ 4 w P F 2 subject to P = TS , V = S .
The optimization problem (27) accords with the framework of ADMM, and the corresponding augmented Lagrangian function of (27) can be written as:
L ( S , V , D , P , R ) = 1 2 Tr ( S T Q S ) Tr ( S T B ) + ι R + J × I ( V ) + ι 1 I T ( 1 J T S ) + λ 4 w P F 2 + μ 2 S V D F 2 + μ 2 TS P R F 2 ,
where D R J × I and R R C × I are two auxiliary variables. Then, the optimization problem (27) can be solved by the following ADMM iterations:
S t = arg min S L ( S , V t 1 , D t 1 , P t 1 , R t 1 ) V t = arg min V L ( S t , V , D t 1 , P t 1 , R t 1 ) P t = arg min P L ( S t , V t , D t 1 , P , R t 1 ) D t = D t 1 ( S t V t ) R t = R t 1 ( TS t P t )
Similar to (18), the solutions of the first two subproblems of (29) can be derived as:
S t = P 1 I T ( F 1 ( B + μ ( V t 1 + D t 1 ) + μ T T ( P t 1 + R t 1 ) ) )
V t = max ( S t D t 1 , 0 )
where F = Q + μ I + μ T T T , and the projection operator P 1 I T ( S ) = S F 1 1 J ( 1 J T F 1 1 J ) 1 ( 1 J T S 1 I T ) . The third subproblem of (29) can be written as follows:
min P μ 2 P ( TS t R t 1 ) F 2 + λ 4 w P F 2 .
The optimization problem (32) is a linear system, which can be solved by the Gauss–Seidel method according to [48,51]. In addition, the optimization problem (32) can also be rewritten as the following formulation:
min P μ 2 P ( TS t R t 1 ) F 2 + λ 2 Tr ( P G P T ) ,
where G R I × I can be treated as the graph Laplacian with G i i = j W i j and G i j = W i j ( j i ). The analytical solution of (33) can be derived as:
P = ( TS t R t 1 ) λ μ G + I 1 .
The algorithm of CJRM is detailed as follows.
  • Input: A training dictionary A R L × J , and a hyperspectral data matrix X R L × I .
  • Choose β and compute the weights W i j according to (22).
  • Select the parameter γ for the RBF kernel and compute the matrices Q and B .
  • Set t = 1 , choose μ , λ , S 1 , V 1 , P 1 , D 1 , R 1 .
  • Repeat
  • Compute S t , V t , P t , D t , R t using (29).
  • t = t + 1 .
  • Until some stopping criterion is satisfied.
  • Output: The estimated label of x i using (4) or (14), i = 1 , , I .
As for JRM, the aforementioned procedure is also suitable for the optimization problem (23). By changing the summation matrix T to the identity matrix I , we can obtain the ADMM iterations of JRM. For the two post-processing methods PRM and CRPM, their formulations are as same as that of (32). Thus, they can be solved by the Gauss–Seidel method according to [48,51], and we can get their analytical solutions with reference to (34).

3.3. Discussion

The 2 norm characterization of coding residual (or fidelity term)—i.e., the first term of (2)—is related to the robustness of KCR-based classification to noise, as stated in [14,15]. For the proposed pixel-wise classifiers KNLS and KFCLS, they can be treated as two instantiations of KCR-based classification, where the collaborative representation mechanism is implemented by the nonnegative constraint. Therefore, their performance is almost as same as that of KSRC and KCRC, which is experimentally demonstrated in Section 4 by using four different hyperspectral scenes. Moreover, in KCR-based classification, the accuracy of one class is usually not vulnerable to the number of training samples taken from another class, since all training samples contribute collaboratively (or competitively) to represent a test sample. In Section 4, the experimental results confirm this phenomenon, where the numbers of training samples of some classes are far less than those of the others (see Table 1 and Table 2). That is to say, KNLS and KFCLS are not too sensitive to class imbalance.
Because of the limitations of remote sensing sensors, a hyperspectral image may contain outliers such as noise and missing or corrupted pixels. The proposed spatial-spectral methods can cope with these pixels. Taking CPRM as an example, the optimization problem (26) can be rewritten as follows:
min u i 1 2 u i p i 2 2 + λ 2 j N i W i j u i u j 2 2 , i = 1 , 2 , , I
where u i is the ith column vector of U . The solutions of (35) can be derived as:
u i = p i + λ j N i W i j u j 1 + λ j N i W i j , i = 1 , 2 , , I .
Since N i is a 3 × 3 neighborhood, (36) can be treated as a 3 × 3 adaptive mean filtering, and thus the outliers can be smoothed by using their neighbors.
Notably, deep learning has attracted a lot of attention very recently. In hyperspectral image classification, various deep models (e.g., stacked autoencoder [57] and convolutional neural network [35,58,59,60]) have been proposed with the observation of good performance in terms of accuracy and flexibility. This paper proposes two new instantiations of KCR-based classification and investigates how to efficiently incorporate spatial-spectral information in the regularization framework. Compared with the deep learning-based methods, the proposed methods have limitations in generalization performance owing to the drawbacks of traditional methods, but have advantages in the requirement of training samples and computational cost as mentioned in the existing deep learning approaches [35,57,58,59,60]. In this paper, we do not expect the proposed methods to exceed the deep learning based methods. It is unfair to compare these two different kinds of methods. Moreover, it is beyond the scope of this paper.

4. Experimental Results and Analysis

4.1. Data Collection and Experimental Setup

In the experiments, four hyperspectral remote sensing datasets have been considered to evaluate the performance of the proposed methods.
(1)
The first one is the Indian Pines dataset taken by the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) over northwest Indiana’s Indian Pines test site in 1992. This dataset contains 220 spectral bands within the wavelength range of 0.4–2.5 μ m, and consists of 145 × 145 pixels. Its spectral and spatial resolutions are 10 nm and 17 m, respectively. There are sixteen ground reference classes of interest, ranging from 20 to 2468 pixels in size. Figure 3 shows the false color composite image and the ground reference map. After removing 20 water absorption bands, there are 200 spectral bands remaining in the experiments. We randomly chose about 5% of the labeled pixels for training, and used the rest for testing, as shown in Table 1.
(2)
The second one is the Kennedy Space Center dataset taken by the AVIRIS sensor over the Kennedy Space Center, Florida, in 1996. This dataset contains 224 spectral bands, covering the wavelength range of 0.4–2.5 μ m. Its spectral and spatial resolutions are 10 nm and 18 m, respectively. This image, with a size of 512 × 614 pixels, contains 176 spectral bands after removing water absorption and low signal-to-noise bands. There are thirteen ground reference classes of interest, ranging from 105 to 927 pixels in size. Figure 4 shows the false color composite image and the ground reference map. In the experiments, we randomly chose 5% of the labeled pixels for training, and used the rest for testing, as shown in Table 2.
(3)
The third one—the University of Pavia dataset—is an urban image collected by the Reflective Optics System Imaging Spectrometer (ROSIS) over University of Pavia, Italy. There are 115 spectral bands in this image, covering the wavelength range of 0.43–0.86 μ m. The image consists of 610 × 340 pixels, with a spatial resolution of 1.3 m per pixel. The false color composite image and the map of nine ground reference classes of interest are shown in Figure 5. In the experiments, there are 103 spectral bands remaining after the removal of noisy bands, and we randomly chose 40 pixels per class for training and used the rest for testing, as shown in Table 3.
(4)
The last one—the Center of Pavia dataset—is another urban image collected by the ROSIS sensor over the center of Pavia city. This image consists of 1096 × 492 pixels, with 102 spectral bands. The reference dataset contains nine classes of interest. The false color composite image and the ground reference map are shown in Figure 6. In the experiments, we randomly chose 20 pixels per class for training and used the rest for testing, as shown in Table 4.
Before the following experiments, the original data were scaled in the range [0,1]. Three metrics—the overall accuracy (OA), the average accuracy (AA) and the kappa coefficient of agreement (KA)—were used to assess the classification accuracy levels, and the quantitative values were obtained by averaging ten random runs. All the experiments were performed on a 64-b quad-core CPU 3.60-GHz processor with 16-GB memory.

4.2. Numerical and Visual Comparisons

In this set of experiments, eleven investigated classification methods described above are compared numerically and visually. These methods can be divided into two categories. The first category comprises five pixel-wise KCR-based classification methods: (1) KSRC, (2) KCRC, (3) the proposed KNLS, (4) the proposed KFCLS using the classification rule “dist” (i.e., (4)), and (5) the proposed KFCLS using the classification rule “prob” (i.e., (14)). The second category comprises six spatial-spectral classification methods mentioned in Section 3.2: (1) JRM using “dist”, (2) JRM using “prob”, (3) CJRM using “dist”, (4) CJRM using “prob”, (5) PRM, and 6) CPRM. Their parameter setting is listed in Table 5.
Table 6 summarizes the class-specific and global classification results for the two AVIRIS datasets, where the processing times in seconds are also included for reference. For the pixel-wise classification, the proposed KNLS and KFCLS can achieve competitive results when compared with KSRC and KCRC, and both the classification rules “dist” and “prob” are suitable for KFCLS. For the spatial-spectral classification, it can be observed that all the spatial-spectral methods perform better than the pixel-wise methods. Among the six spatial-spectral methods, CPRM yields the highest global and most of the best class-specific accuracies followed by CJRM-prob. The improvement of the two JRM methods over the pixel-wise methods is not significant when compared with the other four spatial-spectral methods. This is because JRM combines the spatial-spectral information by enforcing the coding coefficients directly, which is too strict and does not consider the variation of training pixels within each class. For CJRM, the classification rule “prob” is better than the classification rule “dist”. Furthermore, considering PRM using “prob” is equivalent to CPRM, we can conclude that the classification rule “prob” is more suitable for the spatial-spectral classification than the classification rule “dist”. As for the computational cost, KCRC is a cheap pixel-wise classifier since its objective function has an analytical solution. CJRM is the cheapest one among the four kinds of spatial-spectral methods, whereas JRM is the most expensive one.
For the two ROSIS datasets, the classification results and processing times are presented in Table 7. From this table, we can obtain almost the same conclusions as Table 6. It is apparent that KNLS and KFCLS are two competitive methods compared with KSRC and KCRC. When using the same regularization strategy and classification rule, the post-processing methods outperform the co-processing methods. For both the co-processing and post-processing methods, it is better to use the class-level regularization strategy.
Figure 7, Figure 8, Figure 9 and Figure 10 show the classification maps corresponding to one of the ten random tests in each case for the AVIRIS Indian Pines dataset, the AVIRIS Kennedy Space Center dataset, the ROSIS University of Pavia dataset, and the ROSIS Center of Pavia dataset, respectively. From these figures, it can be observed that the numerical comparisons are confirmed by inspecting these classification maps. It is evident that the maps of the spatial-spectral methods are smoother that those of the pixel-wise methods, and the maps of CPRM are closest to the ground truth maps.

4.3. Analysis of Parameters

In the first set of experiments, we investigated the impact of the input parameters on KFCLS. Apart from the ADMM parameter μ that is empirically set to 10 4 , KFCLS has only one parameter γ , which is used for the RBF kernel. Figure 11 shows the impact of γ on the four given datasets, where γ is varied from 2 9 to 2 7 . It can be observed that for all four given datasets, there is a wide optimal range for the choice of γ . When γ is small, the classification rule “prob” is more robust than the classification rule “dist” in most cases, and the difference between them is unapparent when γ is large.
In the next set of experiments, we investigated the impact of the input parameters on CJRM, PRM, and CPRM. Notably, JRM is dropped owing to its low accuracy and heavy computational cost. Apart from the parameters μ and γ that are empirically set to the same values as those used in KFCLS, there are two parameters needing to be tuned. One is the balance parameter λ used in (24)–(26), which is varied in the range [ 10 4 , 10 2 ] for CJRM and [ 10 3 , 10 9 ] for PRM and CPRM; and the other is the weight parameter β used in (22), which is varied in the range [ 5 , 500 ] for CJRM and [ 100 , 1000 ] for PRM and CPRM. Figure 12 shows the classification accuracies for CJRM when applied to the four given datasets. It can be observed that the tuning of λ should synchronize with that of β , the optimal parameters of CJRM-dist are almost the same as those of CJRM-prob, and it is not difficult for us to get a good result in all cases, since there is a wide range for us to choose a suboptimal combination of parameters. Figure 13 shows the classification accuracies for PRM and CPRM when applied to the four given datasets. It is evident that the optimal value of λ is 10 6 in all cases. Notably, a small positive constant ϵ = 10 6 is used in (22), and the majority of W i j will be very small if β is relatively large. In order to connect all the spatially adjacent pixels, it is preferable to fix λ = 1 / ϵ . As for the parameter β , we can choose it in a wide range.

4.4. Impact of the Number of Training Pixels

In this set of experiments, we evaluated the eleven classification methods compared in Section 4.2 in an ill-posed scenario, where different numbers of training pixels are considered. The parameters of these methods are fixed to be the same as those used in Section 4.2. For the two AVIRIS datasets, different percentages of the labeled pixels per class, varied in the range [ 1 % , 20 % ] , were randomly chosen for training, where a minimum of two training pixels per class were taken for very small classes. For the two ROSIS datasets, different numbers of training pixels per class were randomly chosen to build the training sets. Specifically, for University of Pavia, the number was varied to be 10, 20, 40, 60, 80, and 100, and for Center of Pavia, the number was varied to be 5, 10, 20, 40, 60, and 80. Table 8 presents the classification results of the compared methods for the two AVIRIS datasets. It can be observed that for all the compared methods, the OAs increase monotonically as the number of training pixels increases. For the pixel-wise classification, there are no significant gaps between the five methods, and the two investigated classification rules are suitable for the proposed KFCLS. For the spatial-spectral classification, CPRM performs the best, and JRM performs the worst.
Table 9 presents the classification results for the two ROSIS datasets. From this table, we can conclude almost the same results as Table 8. It is evident that the OAs increase monotonically as the number of training pixels increases. For the pixel-wise classification, the proposed KNLS and KFCLS get competitive results compared with KSRC and KCRC. For the spatial-spectral classification, CPRM consistently yields better results than the other five methods, and the improvement of JRM over KFCLS is not significant. Among the three methods CJRM-dist, CJRM-prob, and PRM, CJRM-prob outperformed the others in most cases when applied to the University of Pavia dataset, and they obtained almost the same results when applied to the Center of Pavia dataset.

4.5. Comparison to Other Classification Techniques

In this set of experiments, CPRM is compared with three other techniques that can incorporate the spatial-spectral information into KFCLS-prob. For these methods, the free parameters introduced by KFCLS are set to the same values as those used in KFCLS. The first method is the composite kernel technique [33] using the original spectral features and the extended multiattribute profile (EMAP) features [61,62], termed as CKEMAP, where the EMAP features are extracted from the first three principal components of the hyperspectral image and built by the area and standard deviation attributes as reported in [63]. The additional parameters of CKEMAP are chosen using cross-validation. The second method is the pixel-wise KFCLS-prob followed by Markov random fields (MRF) [36,37], where the MRF technique is utilized to incorporate the spatial-spectral information by refining the posterior probabilistic outputs. The free parameters of MRF are chosen using cross-validation. The last method is the pixel-wise KFCLS-prob followed by majority voting within superpixel regions [39], termed as MV, where the superpixel segmentation algorithm and its free parameters are chosen with reference to [64,65]. Moreover, two baseline classifiers KFCLS-prob and SVM are also included for reference. For SVM, the RBF kernel is used and the free parameters are chosen using cross-validation.
Figure 14 shows the classification accuracies of the six compared methods when different numbers of training pixels are used. It can be observed that CKEMAP performs the best for the two AVIRIS datasets and CPRM performs the best for the two ROSIS datasets. Among the three post-processing methods (i.e., MRF, MV, and the proposed CPRM), CPRM outperforms the other methods in most cases.

5. Conclusions

This paper considers using the nonnegative constraint to achieve the collaborative representation mechanism under SRC and CRC in the kernel space, and thereby proposes KNLS for hyperspectral image classification by replacing 1 -norm or 2 -norm with the nonnegative constraint. In order to provide the posterior probabilistic outputs, we propose KFCLS by enforcing the summation of the nonnegative coding coefficients of each pixel to be one, and subsequently introduce two classification rules to determine the class labels. Compared with KSRC and KCRC, KFCLS can get competitive results and its coding coefficients are more meaningful and useful for the subsequent processing steps. Furthermore, in order to incorporate the spatial-spectral information into KFCLS using regularization technique, we investigated the co-processing and post-processing methods by applying coefficient-level and class-level regularization strategies. Experimental results conducted on four real hyperspectral images have demonstrated: (1) the proposed KFCLS can get competitive results comparing with the other pixel-wise classifiers; (2) the proposed classification rule “prob” is effective; (3) the class-level regularization strategy is better than the coefficient-level regularization strategy; and (4) CPRM is an effective and efficient post-processing method and the most efficient method among the investigated four kinds of methods. In future work, we expect that the suggested regularization method can facilitate the development of spectral unmixing.

Acknowledgments

This work was supported by the National Natural Science Foundation of China (Grant numbers 61601201, 61772274, and 61471199), the Natural Science Foundation of Jiangsu Province (Grant numbers BK20160188 and BK20150160) and the Fundamental Research Funds for the Central Universities (Grant number 30917015104 and JUSRP51635B). The authors would like to thank D. Landgrebe from Purdue University for providing the AVIRIS Indian Pines dataset and P. Gamba from the University of Pavia, Italy, for providing the ROSIS University of Pavia dataset and Center of Pavia dataset.

Author Contributions

Jianjun Liu and Zebin Wu wrote the paper. Jianjun Liu, Zebin Wu and Jinlong Yang analyzed the data. Jianjun Liu and Zhiyong Xiao conceived and designed the experiments. Jianjun Liu performed the experiments. Jianjun Liu contributed analysis tools.

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Wright, J.; Yang, A.Y.; Ganesh, A.; Sastry, S.S.; Ma, Y. Robust face recognition via sparse representation. IEEE Trans. Pattern Anal. Mach. Intell. 2009, 31, 210–227. [Google Scholar] [CrossRef] [PubMed]
  2. Yang, J.; Zhang, L.; Xu, Y.; Yang, J.Y. Beyond sparsity: The role of L1-optimizer in pattern classification. Pattern Recognit. 2012, 45, 1104–1118. [Google Scholar] [CrossRef]
  3. Mei, X.; Ling, H. Robust visual tracking and vehicle classification via sparse representation. IEEE Trans. Pattern Anal. Mach. Intell. 2011, 33, 2259–2272. [Google Scholar] [PubMed]
  4. Yuan, X.T.; Liu, X.; Yan, S. Visual classification with multitask joint sparse representation. IEEE Trans. Image Process. 2012, 21, 4349–4360. [Google Scholar] [CrossRef] [PubMed]
  5. Chen, Y.; Nasrabadi, N.M.; Tran, T.D. Hyperspectral image classification using dictionary-based sparse representation. IEEE Trans. Geosci. Remote Sens. 2011, 49, 3973–3985. [Google Scholar] [CrossRef]
  6. Song, B.; Li, J.; Dalla Mura, M.; Li, P.; Plaza, A.; Bioucas-Dias, J.M.; Benediktsson, J.A.; Chanussot, J. Remotely sensed image classification using sparse representations of morphological attribute profiles. IEEE Trans. Geosci. Remote Sens. 2014, 52, 5122–5136. [Google Scholar] [CrossRef]
  7. Xue, Z.; Du, P.; Su, H.; Zhou, S. Discriminative sparse representation for hyperspectral image classification: A semi-supervised perspective. Remote Sens. 2017, 9, 386. [Google Scholar] [CrossRef]
  8. Bian, X.; Chen, C.; Xu, Y.; Du, Q. Robust hyperspectral image classification by multi-layer spatial-spectral sparse representations. Remote Sens. 2016, 8, 985. [Google Scholar] [CrossRef]
  9. He, Z.; Li, J.; Liu, L. Tensor block-sparsity based representation for spectral-spatial hyperspectral image classification. Remote Sens. 2016, 8, 636. [Google Scholar] [CrossRef]
  10. Xu, Y.; Wu, Z.; Li, J.; Plaza, A.; Wei, Z. Anomaly detection in hyperspectral images based on low-rank and sparse representation. IEEE Trans. Geosci. Remote Sens. 2016, 54, 1990–2000. [Google Scholar] [CrossRef]
  11. Liu, J.; Xiao, Z.; Chen, Y.; Yang, J. Spatial-spectral graph regularized kernel sparse representation for hyperspectral image classification. ISPRS Int. J. Geo-Inf. 2017, 6, 258. [Google Scholar] [CrossRef]
  12. Melgani, F.; Bruzzone, L. Classification of hyperspectral remote sensing images with support vector machines. IEEE Trans. Geosci. Remote Sens. 2004, 42, 1778–1790. [Google Scholar] [CrossRef]
  13. Li, J.; Bioucas-Dias, J.M.; Plaza, A. Semisupervised hyperspectral image segmentation using multinomial logistic regression with active learning. IEEE Trans. Geosci. Remote Sens. 2010, 48, 4085–4098. [Google Scholar] [CrossRef]
  14. Lei, Z.; Meng, Y. Sparse representation or collaborative representation: Which helps face recognition? In Proceedings of the IEEE International Conference on Computer Vision, Barcelona, Spain, 6–13 November 2011; pp. 471–478. [Google Scholar]
  15. Zhang, L.; Yang, M.; Feng, X.; Ma, Y.; Zhang, D. Collaborative representation based classification for face recognition. arXiv, 2012; arXiv:1204.2358. [Google Scholar]
  16. Yang, M.; Zhang, L.; Zhang, D.; Wang, S. Relaxed collaborative representation for pattern classification. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Providence, RI, USA, 16–21 June 2012; pp. 2224–2231. [Google Scholar]
  17. Li, W.; Du, Q. Joint within-class collaborative representation for hyperspectral image classification. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2014, 7, 2200–2208. [Google Scholar] [CrossRef]
  18. Li, J.; Zhang, H.; Huang, Y.; Zhang, L. Hyperspectral image classification by nonlocal joint collaborative representation with a locally adaptive dictionary. IEEE Trans. Geosci. Remote Sens. 2014, 52, 3707–3719. [Google Scholar] [CrossRef]
  19. Camps-Valls, G.; Bruzzone, L. Kernel-based methods for hyperspectral image classification. IEEE Trans. Geosci. Remote Sens. 2005, 43, 1351–1362. [Google Scholar] [CrossRef]
  20. Gao, S.; Tsang, I.W.H.; Chia, L.T. Kernel sparse representation for image classification and face recognition. In European Conference on Computer Vision; Springer: Berlin/Heidelberg, Germany, 2010; pp. 1–14. [Google Scholar]
  21. Chen, Y.; Nasrabadi, N.M.; Tran, T.D. Hyperspectral image classification via kernel sparse representation. IEEE Trans. Geosci. Remote Sens. 2013, 51, 217–231. [Google Scholar] [CrossRef]
  22. Wang, D.; Lu, H.; Yang, M.H. Kernel collaborative face recognition. Pattern Recognit. 2015, 48, 3025–3037. [Google Scholar] [CrossRef]
  23. Liu, J.; Wu, Z.; Li, J.; Plaza, A.; Yuan, Y. Probabilistic-Kernel collaborative representation for spatial-spectral hyperspectral image classification. IEEE Trans. Geosci. Remote Sens. 2016, 54, 2371–2384. [Google Scholar] [CrossRef]
  24. Li, J.; Zhang, H.; Zhang, L. Column-generation kernel nonlocal joint collaborative representation for hyperspectral image classification. ISPRS J. Photogramm. Remote Sens. 2014, 94, 25–36. [Google Scholar] [CrossRef]
  25. Li, W.; Du, Q.; Xiong, M. Kernel collaborative representation with Tikhonov regularization for hyperspectral image classification. IEEE Geosci. Remote Sens. Lett. 2015, 12, 48–52. [Google Scholar]
  26. Lee, D.D.; Seung, H.S. Algorithms for non-negative matrix factorization. Adv. Neural Inf. Process. Syst. 2000, 13, 556–562. [Google Scholar]
  27. Iordache, M.D.; Bioucas-Dias, J.M.; Plaza, A. Sparse unmixing of hyperspectral data. IEEE Trans. Geosci. Remote Sens. 2011, 49, 2014–2039. [Google Scholar] [CrossRef]
  28. Foucart, S.; Koslicki, D. Sparse recovery by means of nonnegative least squares. IEEE Signal Process. Lett. 2014, 21, 498–502. [Google Scholar] [CrossRef]
  29. Slawski, M.; Hein, M. Sparse recovery by thresholded non-negative least squares. Adv. Neural Inf. Process. Syst. 2011, 24, 1926–1934. [Google Scholar]
  30. Heinz, D.C. Fully constrained least squares linear spectral mixture analysis method for material quantification in hyperspectral imagery. IEEE Trans. Geosci. Remote Sens. 2001, 39, 529–545. [Google Scholar] [CrossRef]
  31. Broadwater, J.; Chellappa, R.; Banerjee, A.; Burlina, P. Kernel fully constrained least squares abundance estimates. In Proceedings of the IEEE International Geoscience and Remote Sensing Symposium, Barcelona, Spain, 23–28 July 2007; pp. 4041–4044. [Google Scholar]
  32. Fauvel, M.; Tarabalka, Y.; Benediktsson, J.A.; Chanussot, J.; Tilton, J.C. Advances in spectral-spatial classification of hyperspectral images. Proc. IEEE 2013, 101, 652–675. [Google Scholar] [CrossRef]
  33. Camps-Valls, G.; Gomez-Chova, L.; Muñoz-Marí, J.; Vila-Francés, J.; Calpe-Maravilla, J. Composite kernels for hyperspectral image classification. IEEE Geosci. Remote Sens. Lett. 2006, 3, 93–97. [Google Scholar] [CrossRef]
  34. Li, J.; Huang, X.; Gamba, P.; Bioucas-Dias, J.M.B.; Zhang, L.; Benediktsson, J.A.; Plaza, A. Multiple feature learning for hyperspectral image classification. IEEE Trans. Geosci. Remote Sens. 2015, 53, 1592–1606. [Google Scholar] [CrossRef]
  35. Liang, H.; Li, Q. Hyperspectral imagery classification using sparse representations of convolutional neural network features. Remote Sens. 2016, 8, 99. [Google Scholar] [CrossRef]
  36. Tarabalka, Y.; Fauvel, M.; Chanussot, J.; Benediktsson, J.A. SVM- and MRF-based method for accurate classification of hyperspectral images. IEEE Geosci. Remote Sens. Lett. 2010, 7, 736–740. [Google Scholar] [CrossRef]
  37. Li, J.; Bioucas-Dias, J.M.; Plaza, A. Hyperspectral image segmentation using a new Bayesian approach with active learning. IEEE Trans. Geosci. Remote Sens. 2011, 49, 3947–3960. [Google Scholar] [CrossRef]
  38. Wu, Z.; Shi, L.; Li, J.; Wang, Q.; Sun, L.; Wei, Z.; Plaza, J.; Plaza, A. GPU parallel implementation of spatially adaptive hyperspectral image classification. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2017, PP, 1–13. [Google Scholar] [CrossRef]
  39. Tarabalka, Y.; Chanussot, J.; Benediktsson, J.A. Segmentation and classification of hyperspectral images using watershed transformation. Pattern Recognit. 2010, 43, 2367–2379. [Google Scholar] [CrossRef]
  40. Priya, T.; Prasad, S.; Wu, H. Superpixels for spatially reinforced bayesian classification of hyperspectral images. IEEE Geosci. Remote Sens. Lett. 2015, 12, 1071–1075. [Google Scholar] [CrossRef]
  41. Liu, J.; Wu, Z.; Sun, L.; Wei, Z.; Xiao, L. Hyperspectral image classification using kernel sparse representation and semilocal spatial graph regularization. IEEE Geosci. Remote Sens. Lett. 2014, 11, 1320–1324. [Google Scholar]
  42. Feng, J.; Cao, Z.; Pi, Y. Polarimetric contextual classification of PolSAR images using sparse representation and superpixels. Remote Sens. 2014, 6, 7158–7181. [Google Scholar] [CrossRef]
  43. Wang, J.; Jiao, L.; Liu, H.; Yang, S.; Liu, F. Hyperspectral image classification by spatial-spectral derivative-aided kernel joint sparse representation. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2015, 8, 2485–2500. [Google Scholar] [CrossRef]
  44. Zhang, S.; Li, S.; Fu, W.; Fang, L. Multiscale superpixel-based sparse representation for hyperspectral image classification. Remote Sens. 2017, 9, 139. [Google Scholar] [CrossRef]
  45. Tong, F.; Tong, H.; Jiang, J.; Zhang, Y. Multiscale union regions adaptive sparse representation for hyperspectral image classification. Remote Sens. 2017, 9, 872. [Google Scholar] [CrossRef]
  46. Yuan, H.; Tang, Y.Y.; Lu, Y.; Yang, L.; Luo, H. Hyperspectral image classification based on regularized sparse representation. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2014, 7, 2174–2182. [Google Scholar] [CrossRef]
  47. Aubert, G.; Kornprobst, P. Mathematical problems in image processing: Partial differential equations and calculus of variations. Appl. Intell. 2006, 40, 291–304. [Google Scholar]
  48. Zhang, X.; Burger, M.; Bresson, X.; Osher, S. Bregmanized nonlocal regularization for deconvolution and sparse reconstruction. SIAM J. Imaging Sci. 2010, 3, 253–276. [Google Scholar] [CrossRef]
  49. Liu, J.; Wu, Z.; Wei, Z.; Xiao, L.; Sun, L. Spatial-spectral kernel sparse representation for hyperspectral image classification. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2013, 6, 2462–2471. [Google Scholar] [CrossRef]
  50. Yang, J.; Zhang, Y. Alternating direction algorithms for 1-Problems in compressive sensing. SIAM J. Sci. Comput. 2009, 33, 250–278. [Google Scholar] [CrossRef]
  51. Goldstein, T.; Osher, S. The split bregman method for L1-Regularized problems. SIAM J. Imaging Sci. 2009, 2, 323–343. [Google Scholar] [CrossRef]
  52. Mairal, J.; Bach, F.; Ponce, J. Sparse modeling for image and vision processing. Found. Trends Comput. Graph. Vis. 2014, 8, 85–283. [Google Scholar] [CrossRef] [Green Version]
  53. Bioucas-Dias, J.M.; Figueiredo, M.A. Alternating direction algorithms for constrained sparse regression: Application to hyperspectral unmixing. In Proceedings of the IEEE Workshop on Hyperspectral Image and Signal Processing: Evolution in Remote Sensing, Reykjavik, Iceland, 14–16 June 2010; pp. 1–4. [Google Scholar]
  54. Wen, Y.W.; Wang, M.; Cao, Z.; Cheng, X.; Ching, W.K.; Vassiliadis, V.S. Sparse solution of nonnegative least squares problems with applications in the construction of probabilistic Boolean networks. Numer. Linear Algebra Appl. 2015, 22, 883–899. [Google Scholar] [CrossRef]
  55. Combettes, P.L.; Wajs, V.R. Signal recovery by proximal forward-backward splitting. SIAM J. Multiscale Model. Simul. 2005, 4, 1168–1200. [Google Scholar] [CrossRef]
  56. Kang, X.; Li, S.; Fang, L.; Li, M.; Benediktsson, J.A. Extended random walker-based classification of hyperspectral images. IEEE Trans. Geosci. Remote Sens. 2015, 53, 144–153. [Google Scholar] [CrossRef]
  57. Chen, Y.; Lin, Z.; Zhao, X.; Wang, G.; Gu, Y. Deep learning-based classification of hyperspectral data. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2014, 7, 2094–2107. [Google Scholar] [CrossRef]
  58. Makantasis, K.; Karantzalos, K.; Doulamis, A.; Doulamis, N. Deep supervised learning for hyperspectral data classification through convolutional neural networks. In Proceedings of the IEEE International Geoscience and Remote Sensing Symposium, Milan, Italy, 26–31 July 2015; pp. 4959–4962. [Google Scholar]
  59. Hu, W.; Huang, Y.; Wei, L.; Zhang, F.; Li, H. Deep convolutional neural networks for hyperspectral image classification. J. Sens. 2015, 2015, 1–12. [Google Scholar] [CrossRef]
  60. Chen, Y.; Jiang, H.; Li, C.; Jia, X.; Ghamisi, P. Deep feature extraction and classification of hyperspectral images based on convolutional neural networks. IEEE Trans. Geosci. Remote Sens. 2016, 54, 6232–6251. [Google Scholar] [CrossRef]
  61. Dalla Mura, M.; Atli Benediktsson, J.; Waske, B.; Bruzzone, L. Extended profiles with morphological attribute filters for the analysis of hyperspectral data. Int. J. Remote Sens. 2010, 31, 5975–5991. [Google Scholar] [CrossRef]
  62. Mura, M.D.; Benediktsson, J.A.; Waske, B.; Bruzzone, L. Morphological attribute profiles for the analysis of very high resolution images. IEEE Trans. Geosci. Remote Sens. 2010, 48, 3747–3762. [Google Scholar] [CrossRef]
  63. Li, J.; Marpu, P.R.; Plaza, A.; Bioucas-Dias, J.M.; Benediktsson, J.A. Generalized composite kernel framework for hyperspectral image classification. IEEE Trans. Geosci. Remote Sens. 2013, 51, 4816–4829. [Google Scholar] [CrossRef]
  64. Fang, L.; Li, S.; Duan, W.; Ren, J.; Benediktsson, J.A. Classification of hyperspectral images by exploiting spectral-spatial information of superpixel via multiple kernels. IEEE Trans. Geosci. Remote Sens. 2015, 53, 6663–6674. [Google Scholar] [CrossRef]
  65. Liu, J.; Xiao, Z.; Xiao, L. Superpixel-guided multiscale kernel collaborative representation for hyperspectral image classification. Remote Sens. Lett. 2016, 7, 975–984. [Google Scholar] [CrossRef]
Figure 1. Estimated coefficients for the pixels in the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) Indian Pines dataset (about 5% training pixels are used, see Section 4), the corresponding class labels are included in the parentheses, and all coefficients are ranked in order of the class labels. (ac): Pixel taken from Class C9, and the coefficients of C9 are in the range (219, 221]. (df): Pixel taken from Class C5, and the coefficients of C5 are in the range (129, 154]. (gi): Pixel taken from Class C11, and the coefficients of C11 are in the range (270, 394]. (a) KNLS (C9). (b) KFCLS (C9). (c) KSRC (C9). (d) KNLS (C5). (e) KFCLS (C5). (f) KSRC (C5). (g) KNLS (C11). (h) KFCLS (C11). (i) KSRC (C11). Notably, alternating direction method of multipliers (ADMM) is used to solve the optimization problems of KNLS and KFLCS, and thus their coefficients are not strictly nonnegative.
Figure 1. Estimated coefficients for the pixels in the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) Indian Pines dataset (about 5% training pixels are used, see Section 4), the corresponding class labels are included in the parentheses, and all coefficients are ranked in order of the class labels. (ac): Pixel taken from Class C9, and the coefficients of C9 are in the range (219, 221]. (df): Pixel taken from Class C5, and the coefficients of C5 are in the range (129, 154]. (gi): Pixel taken from Class C11, and the coefficients of C11 are in the range (270, 394]. (a) KNLS (C9). (b) KFCLS (C9). (c) KSRC (C9). (d) KNLS (C5). (e) KFCLS (C5). (f) KSRC (C5). (g) KNLS (C11). (h) KFCLS (C11). (i) KSRC (C11). Notably, alternating direction method of multipliers (ADMM) is used to solve the optimization problems of KNLS and KFLCS, and thus their coefficients are not strictly nonnegative.
Ijgi 06 00344 g001
Figure 2. Sum of the estimated coefficients for the pixels in the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) Indian Pines dataset (about 5% training pixels are used, see Section 4), the corresponding class labels are included in the parentheses. (ac): Pixel taken from Class C9. (df): Pixel taken from Class C5. (gi): Pixel taken from Class C11. (a) KNLS (C9). (b) KFCLS (C9). (c) KSRC (C9). (d) KNLS (C5). (e) KFCLS (C5). (f) KSRC (C5). (g) KNLS (C11). (h) KFCLS (C11). (i) KSRC (C11).
Figure 2. Sum of the estimated coefficients for the pixels in the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) Indian Pines dataset (about 5% training pixels are used, see Section 4), the corresponding class labels are included in the parentheses. (ac): Pixel taken from Class C9. (df): Pixel taken from Class C5. (gi): Pixel taken from Class C11. (a) KNLS (C9). (b) KFCLS (C9). (c) KSRC (C9). (d) KNLS (C5). (e) KFCLS (C5). (f) KSRC (C5). (g) KNLS (C11). (h) KFCLS (C11). (i) KSRC (C11).
Ijgi 06 00344 g002
Figure 3. AVIRIS Indian Pines dataset. (a) RGB composite image. (b) Ground reference map.
Figure 3. AVIRIS Indian Pines dataset. (a) RGB composite image. (b) Ground reference map.
Ijgi 06 00344 g003
Figure 4. AVIRIS Kennedy Space Center dataset. (a) RGB composite image. (b) Ground reference map.
Figure 4. AVIRIS Kennedy Space Center dataset. (a) RGB composite image. (b) Ground reference map.
Ijgi 06 00344 g004
Figure 5. Reflective Optics System Imaging Spectrometer (ROSIS) University of Pavia dataset. (a) RGB composite image. (b) Ground reference map.
Figure 5. Reflective Optics System Imaging Spectrometer (ROSIS) University of Pavia dataset. (a) RGB composite image. (b) Ground reference map.
Ijgi 06 00344 g005
Figure 6. ROSIS Center of Pavia dataset. (a) RGB composite image. (b) Ground reference map.
Figure 6. ROSIS Center of Pavia dataset. (a) RGB composite image. (b) Ground reference map.
Ijgi 06 00344 g006
Figure 7. Classification maps and overall classification accuracy levels (in parentheses) obtained for the AVIRIS Indian Pines dataset using different classification methods. (a) KSRC (81.50), (b) KCRC (79.52), (c) KNLS (81.94), (d) KFCLS-dist (81.91), (e) KFCLS-prob (81.49), (f) JRM-dist (85.54), (g) JRM-prob (86.27), (h) CJRM-dist (89.64), (i) CJRM-prob (92.42), (j) PRM (91.29), (k) CPRM (92.73).
Figure 7. Classification maps and overall classification accuracy levels (in parentheses) obtained for the AVIRIS Indian Pines dataset using different classification methods. (a) KSRC (81.50), (b) KCRC (79.52), (c) KNLS (81.94), (d) KFCLS-dist (81.91), (e) KFCLS-prob (81.49), (f) JRM-dist (85.54), (g) JRM-prob (86.27), (h) CJRM-dist (89.64), (i) CJRM-prob (92.42), (j) PRM (91.29), (k) CPRM (92.73).
Ijgi 06 00344 g007
Figure 8. Classification maps and overall classification accuracy levels (in parentheses) obtained for the AVIRIS Kennedy Space Center dataset using different classification methods. (a) KSRC (89.86), (b) KCRC (88.99), (c) KNLS (89.90), (d) KFCLS-dist (89.97), (e) KFCLS-prob (90.17), (f) JRM-dist (90.59), (g) JRM-prob (90.57), (h) CJRM-dist (95.04), (i) CJRM-prob (95.63), (j) PRM (95.51), (k) CPRM (95.75).
Figure 8. Classification maps and overall classification accuracy levels (in parentheses) obtained for the AVIRIS Kennedy Space Center dataset using different classification methods. (a) KSRC (89.86), (b) KCRC (88.99), (c) KNLS (89.90), (d) KFCLS-dist (89.97), (e) KFCLS-prob (90.17), (f) JRM-dist (90.59), (g) JRM-prob (90.57), (h) CJRM-dist (95.04), (i) CJRM-prob (95.63), (j) PRM (95.51), (k) CPRM (95.75).
Ijgi 06 00344 g008
Figure 9. Classification maps and overall classification accuracy levels (in parentheses) obtained for the ROSIS University of Pavia dataset using different classification methods. (a) KSRC (83.44), (b) KCRC (82.74), (c) KNLS (84.40), (d) KFCLS-dist (84.43), (e) KFCLS-prob (84.19), (f) JRM-dist (90.96), (g) JRM-prob (91.29), (h) CJRM-dist (97.23), (i) CJRM-prob (97.71), (j) PRM (97.98), (k) CPRM (98.57).
Figure 9. Classification maps and overall classification accuracy levels (in parentheses) obtained for the ROSIS University of Pavia dataset using different classification methods. (a) KSRC (83.44), (b) KCRC (82.74), (c) KNLS (84.40), (d) KFCLS-dist (84.43), (e) KFCLS-prob (84.19), (f) JRM-dist (90.96), (g) JRM-prob (91.29), (h) CJRM-dist (97.23), (i) CJRM-prob (97.71), (j) PRM (97.98), (k) CPRM (98.57).
Ijgi 06 00344 g009
Figure 10. Classification maps and overall classification accuracy levels (in parentheses) obtained for the ROSIS Center of Pavia dataset using different classification methods. (a) KSRC (96.49), (b) KCRC (96.44), (c) KNLS (96.65), (d) KFCLS-dist (96.65), (e) KFCLS-prob (96.65), (f) JRM-dist (97.45), (g) JRM-prob (97.50), (h) CJRM-dist (98.82), (i) CJRM-prob (98.74), (j) PRM (98.84), (k) CPRM (98.93).
Figure 10. Classification maps and overall classification accuracy levels (in parentheses) obtained for the ROSIS Center of Pavia dataset using different classification methods. (a) KSRC (96.49), (b) KCRC (96.44), (c) KNLS (96.65), (d) KFCLS-dist (96.65), (e) KFCLS-prob (96.65), (f) JRM-dist (97.45), (g) JRM-prob (97.50), (h) CJRM-dist (98.82), (i) CJRM-prob (98.74), (j) PRM (98.84), (k) CPRM (98.93).
Ijgi 06 00344 g010
Figure 11. OA as a function of γ for KFCLS when applied to the four given datasets.
Figure 11. OA as a function of γ for KFCLS when applied to the four given datasets.
Ijgi 06 00344 g011
Figure 12. OA with respect to λ and β for CJRM when applied to the four given datasets. (a) CJRM-dist (Indian Pines), (b) CJRM-prob (Indian Pines), (c) CJRM-dist (Kennedy Space Center), (d) CJRM-prob (Kennedy Space Center), (e) CJRM-dist (University of Pavia), (f) CJRM-prob (University of Pavia), (g) CJRM-dist (Center of Pavia), (h) CJRM-prob (Center of Pavia).
Figure 12. OA with respect to λ and β for CJRM when applied to the four given datasets. (a) CJRM-dist (Indian Pines), (b) CJRM-prob (Indian Pines), (c) CJRM-dist (Kennedy Space Center), (d) CJRM-prob (Kennedy Space Center), (e) CJRM-dist (University of Pavia), (f) CJRM-prob (University of Pavia), (g) CJRM-dist (Center of Pavia), (h) CJRM-prob (Center of Pavia).
Ijgi 06 00344 g012
Figure 13. OA with respect to λ and β for PRM and CPRM when applied to the four given datasets. (a) PRM (Indian Pines), (b) CPRM (Indian Pines), (c) PRM (Kennedy Space Center), (d) CPRM (Kennedy Space Center), (e) PRM (University of Pavia), (f) CPRM (University of Pavia), (g) PRM (Center of Pavia), (h) CPRM (Center of Pavia).
Figure 13. OA with respect to λ and β for PRM and CPRM when applied to the four given datasets. (a) PRM (Indian Pines), (b) CPRM (Indian Pines), (c) PRM (Kennedy Space Center), (d) CPRM (Kennedy Space Center), (e) PRM (University of Pavia), (f) CPRM (University of Pavia), (g) PRM (Center of Pavia), (h) CPRM (Center of Pavia).
Ijgi 06 00344 g013
Figure 14. OA as a function of the number of training pixels when applied to the four given datasets. (a) AVIRIS Indian Pines dataset, (b) AVIRIS Kennedy Space Center dataset, (c) ROSIS University of Pavia dataset, (d) ROSIS Center of Pavia dataset.
Figure 14. OA as a function of the number of training pixels when applied to the four given datasets. (a) AVIRIS Indian Pines dataset, (b) AVIRIS Kennedy Space Center dataset, (c) ROSIS University of Pavia dataset, (d) ROSIS Center of Pavia dataset.
Ijgi 06 00344 g014
Table 1. The ground reference classes in the AVIRIS Indian Pines dataset and the number of training and test pixels used in experiments.
Table 1. The ground reference classes in the AVIRIS Indian Pines dataset and the number of training and test pixels used in experiments.
No.Class NameTrainTest
C01Alfalfa351
C02Corn-no till721362
C03Corn-min till42792
C04Corn12222
C05Grass/pasture25472
C06Grass/trees38709
C07Grass/pasture-mowed224
C08Hay-windrowed25464
C09Oats218
C10Soybeans-no till49919
C11Soybeans-min till1242344
C12Soybean-clean till31583
C13Wheat11201
C14Woods651229
C15Bldg-grass-tree drives19361
C16Stone-steel towers590
Total5259841
Table 2. The ground reference classes in the AVIRIS Kennedy Space Center dataset and the number of training and test pixels used in experiments.
Table 2. The ground reference classes in the AVIRIS Kennedy Space Center dataset and the number of training and test pixels used in experiments.
No.Class NameTrainTest
C01Scrub39722
C02Willow swamp13230
C03Cabbage palm hammock13243
C04Cabbage palm/oak hammock13239
C05Slash pine9152
C06Oak/broadleaf hammock12217
C07Hardwood swamp699
C08Graminoid marsh22409
C09Spartina marsh26494
C10Cattail marsh21383
C11Salt marsh21398
C12Mud flats26477
C13Water47880
Total2684943
Table 3. The ground reference classes in the ROSIS University of Pavia dataset and the number of training and test pixels used in experiments.
Table 3. The ground reference classes in the ROSIS University of Pavia dataset and the number of training and test pixels used in experiments.
No.Class NameTrainTest
C1Asphalt406812
C2Meadow4018,646
C3Gravel402167
C4Trees403396
C5Metal sheets401338
C6Bare soil405064
C7Bitumen401316
C8Bricks403838
C9Shadows40986
Total36043,563
Table 4. The ground reference classes in the ROSIS Center of Pavia dataset and the number of training and test pixels used in experiments.
Table 4. The ground reference classes in the ROSIS Center of Pavia dataset and the number of training and test pixels used in experiments.
No.Class NameTrainTest
C1Water2065,258
C2Trees206488
C3Meadow202885
C4Bricks202132
C5Soil206529
C6Asphalt207565
C7Bitumen207267
C8Tile203102
C9Shadows202145
Total180103,371
Table 5. The optimal combination of parameters for the investigated methods when applied to the four given datasets. JRM: joint regularization model; CJRM: class-level JRM; KCRC: kernel collaborative representation classification; KFCLS: kernel fully constrained least squares; KNLS: kernel nonnegative constrained least squares; KSRC: kernel sparse representation classification; PRM: post-processing regularization model; CPRM: class-level PRM.
Table 5. The optimal combination of parameters for the investigated methods when applied to the four given datasets. JRM: joint regularization model; CJRM: class-level JRM; KCRC: kernel collaborative representation classification; KFCLS: kernel fully constrained least squares; KNLS: kernel nonnegative constrained least squares; KSRC: kernel sparse representation classification; PRM: post-processing regularization model; CPRM: class-level PRM.
Pixel-Wise ClassificationSpatial-Spectral Classification
KSRCKCRCKNLSKFCLSJRMCJRMPRMCPRM
μ = 10 3 μ = 10 4 μ = 10 4 μ = 10 3 μ = 10 4 μ = 10 4 μ = 10 4
Indian γ = 2 γ = 2 γ = 2 γ = 2 γ = 2 γ = 2 γ = 2 γ = 2
Pines λ = 10 4 λ = 10 3 λ = 1 λ = 10 2 λ = 10 6 λ = 10 6
β = 100 β = 25 β = 450 β = 450
μ = 10 3 μ = 10 4 μ = 10 4 μ = 10 3 μ = 10 4 μ = 10 4 μ = 10 4
Kennedy γ = 1 8 γ = 1 8 γ = 1 8 γ = 1 8 γ = 1 8 γ = 1 8 γ = 1 8 γ = 1 8
Space Center λ = 10 4 λ = 10 3 λ = 1 λ = 10 2 λ = 10 6 λ = 10 6
β = 100 β = 25 β = 800 β = 800
μ = 10 3 μ = 10 4 μ = 10 4 μ = 10 3 μ = 10 4 μ = 10 4 μ = 10 4
University γ = 1 2 γ = 1 2 γ = 1 2 γ = 1 2 γ = 1 2 γ = 1 2 γ = 1 2 γ = 1 2
of Pavia λ = 10 4 λ = 10 3 λ = 1 λ = 1 λ = 10 6 λ = 10 6
β = 100 β = 100 β = 450 β = 450
μ = 10 3 μ = 10 4 μ = 10 4 μ = 10 3 μ = 10 4 μ = 10 4 μ = 10 4
Center γ = 1 8 γ = 1 8 γ = 1 8 γ = 1 8 γ = 1 8 γ = 1 8 γ = 1 8 γ = 1 8
of Pavia λ = 10 4 λ = 10 3 λ = 1 λ = 1 λ = 10 6 λ = 10 6
β = 100 β = 100 β = 500 β = 500
Table 6. Classification accuracies for the two AVIRIS datasets using different classification methods. For both the pixel-wise classification and spatial-spectral classification, the best results are highlighted in bold, and the second best results are underlined. AA: average accuracy; KA: kappa coefficient of agreement; OA: overall accuracy.
Table 6. Classification accuracies for the two AVIRIS datasets using different classification methods. For both the pixel-wise classification and spatial-spectral classification, the best results are highlighted in bold, and the second best results are underlined. AA: average accuracy; KA: kappa coefficient of agreement; OA: overall accuracy.
Pixel-Wise ClassificationSpatial-Spectral Classification
KSRCKCRCKNLSKFCLSJRMCJRMPRMCPRM
distprobdistprobdistprob
Indian Pines
C01 56.67 43.7355.6955.49 56.47 ̲ 58.2459.6160.5965.88 77.45 ̲ 77.84
C0278.3573.2878.55 78.63 78.56 ̲ 83.3784.0584.0683.66 86.74 ̲ 86.92
C0364.3153.03 64.49 ̲ 64.36 64.53 71.1972.8579.39 89.58 81.10 85.34 ̲
C0452.1241.40 52.88 ̲ 52.79 52.88 56.9860.1868.06 82.61 ̲ 71.71 83.24
C0589.0386.17 89.17 ̲ 89.17 88.9889.8590.0490.7491.61 92.22 ̲ 92.22
C0696.46 96.80 96.46 96.49 ̲ 95.8597.2497.0898.28 99.00 ̲ 98.91 99.48
C07 74.58 54.5874.17 74.17 ̲ 70.4288.3387.9287.0888.33 92.92 90.00 ̲
C0898.75 99.27 98.86 98.86 ̲ 98.4799.1899.0999.3899.44 99.59 ̲ 100
C0957.7836.1157.78 57.78 ̲ 61.67 67.78 76.11 ̲ 62.78 78.33 59.4436.67
C1072.8761.60 73.33 72.96 73.04 ̲ 77.6079.34 88.18 ̲ 90.02 87.6688.07
C1182.42 89.78 82.53 83.16 ̲ 82.4388.7088.3892.76 95.68 ̲ 93.97 96.39
C1276.7665.0478.16 78.37 ̲ 79.50 85.5188.6492.26 97.26 ̲ 94.34 97.58
C1398.7698.86 98.96 ̲ 99.00 98.5199.0099.0599.2599.30 99.50 ̲ 99.80
C1495.28 97.34 95.68 ̲ 95.6695.2997.2797.0598.13 98.62 ̲ 98.28 98.70
C1553.8041.94 53.91 ̲ 53.68 57.40 52.9656.3762.88 72.27 ̲ 71.33 81.30
C16 88.56 79.89 88.44 ̲ 87.3368.78 86.44 ̲ 79.44 86.56 80.8986.2285.00
OA(%)81.3478.99 81.61 ̲ 81.72 81.4685.5386.1589.49 92.26 ̲ 91.00 92.86
AA(%)77.2869.93 77.44 77.37 ̲ 76.4281.2382.2084.40 88.28 86.96 87.41 ̲
KA(%)78.6675.69 78.98 ̲ 79.09 78.8083.4484.1788.00 91.17 ̲ 89.72 91.84
Time(s) 9.28 ̲ 0.97 15.8919.5019.1947.2646.95 14.48 ̲ 14.28 20.0819.28
Kennedy Space Center
C0195.69 97.09 95.80 95.82 ̲ 95.7997.5897.6299.47 99.92 ̲ 99.40 99.96
C0285.0985.04 85.22 ̲ 85.35 84.8785.7485.78 93.35 93.09 ̲ 89.9691.09
C03 90.62 ̲ 91.48 90.0890.0890.2997.1697.2098.1198.19 98.19 ̲ 98.93
C04 46.78 41.8846.1546.40 46.65 ̲ 50.9250.6352.8054.60 65.10 64.27 ̲
C0561.6459.1461.64 61.71 ̲ 62.30 70.6670.99 82.89 82.50 ̲ 79.8780.53
C06 45.58 34.1044.7044.61 45.16 ̲ 36.2736.36 62.30 ̲ 65.58 61.8061.38
C0783.54 84.34 ̲ 83.9483.84 84.44 90.1090.4097.58 98.59 96.46 97.88 ̲
C08 89.29 88.1988.9088.88 89.07 ̲ 93.8193.9497.87 99.07 ̲ 98.34 99.27
C0996.01 96.66 96.32 ̲ 96.3096.2897.9697.96 98.34 ̲ 98.30 98.34 98.30
C1095.3094.15 95.30 ̲ 95.33 93.2496.6696.5898.64 99.27 ̲ 98.80 99.48
C1194.30 95.00 94.6094.60 94.92 ̲ 95.3895.1597.09 97.94 ̲ 97.49 99.45
C12 88.05 ̲ 87.4287.3487.25 88.28 87.2587.4892.01 94.03 ̲ 93.52 94.30
C13 99.84 99.75 ̲ 99.6999.7299.3499.8999.90 100 100 100 100
OA(%) 88.45 87.7688.32 88.33 ̲ 88.2989.9790.0093.56 94.26 ̲ 94.08 94.60
AA(%) 82.44 81.1082.2882.30 82.36 ̲ 84.5784.6190.04 90.85 ̲ 90.56 91.14
KA(%) 87.12 86.3186.96 86.98 ̲ 86.9488.8088.8392.81 93.59 ̲ 93.39 93.97
Time(s) 65.67 ̲ 6.06 117.8142.0140.3965.5963.1 113.4 112 . 2 ̲ 156.1155.3
Table 7. Classification accuracies for the two ROSIS datasets using different classification methods. For both the pixel-wise classification and spatial-spectral classification, the best results are highlighted in bold, and the second best results are underlined.
Table 7. Classification accuracies for the two ROSIS datasets using different classification methods. For both the pixel-wise classification and spatial-spectral classification, the best results are highlighted in bold, and the second best results are underlined.
Pixel-Wise ClassificationSpatial-Spectral Classification
KSRCKCRCKNLSKFCLSJRMCJRMPRMCPRM
distprobdistprobdistprob
University of Pavia
C173.2572.7874.77 74.77 ̲ 75.22 80.0380.2994.73 96.19 ̲ 95.49 98.16
C283.2482.6183.55 83.57 ̲ 83.67 87.7088.3296.64 98.09 ̲ 96.55 98.12
C378.6877.64 78.80 78.77 ̲ 78.5284.7284.6386.7787.08 92.32 ̲ 92.51
C491.65 93.75 92.68 ̲ 92.6191.23 95.00 ̲ 94.7094.8794.86 95.40 93.37
C5 99.42 99.2399.33 99.34 ̲ 98.2799.2498.67 99.28 ̲ 98.3698.94 99.90
C684.77 85.64 85.55 ̲ 85.4884.7293.0493.1798.78 99.55 ̲ 99.12 99.91
C7 92.65 ̲ 94.17 92.5492.3692.2996.7596.7899.59 99.87 99.51 99.72 ̲
C877.99 80.03 78.69 ̲ 78.5778.3787.9387.8295.88 97.30 96.44 ̲ 96.14
C9 99.27 ̲ 99.2299.25 99.28 98.8099.2998.9299.37 99.43 ̲ 99.23 99.54
OA(%)82.9683.05 83.57 83.55 ̲ 83.3988.4588.7196.13 97.19 ̲ 96.60 97.65
AA(%)86.77 87.23 ̲ 87.24 87.2086.7991.5291.4896.2196.75 97.00 ̲ 97.48
KA(%)78.2178.37 78.97 78.94 ̲ 78.7285.1685.4794.94 96.32 ̲ 95.57 96.92
Time(s) 59.59 ̲ 5.74 101.7119.1117.2344.1342.1 75.84 ̲ 74.86 124.5119.0
Center of Pavia
C199.67 99.73 99.6999.69 99.70 ̲ 99.8699.90 100 100 100 100
C291.09 91.33 91.16 91.18 ̲ 91.0791.7191.67 97.66 ̲ 97.77 95.8696.62
C3 89.71 ̲ 90.43 89.6089.5989.25 90.93 ̲ 90.1083.5881.81 91.21 88.72
C4 89.40 88.8488.99 89.00 ̲ 88.7290.8590.7899.3899.09 99.39 ̲ 99.52
C587.95 89.43 ̲ 89.1689.07 89.82 93.2193.39 96.46 ̲ 97.08 95.5496.23
C696.84 97.62 96.9396.94 97.01 ̲ 97.9598.0199.5599.50 99.70 ̲ 99.83
C786.2985.52 86.57 ̲ 86.57 86.2388.7888.64 94.53 94.46 ̲ 93.2593.44
C898.8398.76 98.93 ̲ 98.94 98.8399.2299.1596.9496.66 99.52 ̲ 99.75
C9 99.98 99.9799.98 99.98 ̲ 99.97 100 99.99 ̲ 95.9695.2094.0493.69
OA(%)96.74 96.89 96.8496.84 96.85 ̲ 97.5697.5698.5698.52 98.57 ̲ 98.61
AA(%)93.31 93.51 93.44 ̲ 93.4493.4094.7294.6396.0195.73 96.50 96.42 ̲
KA(%)94.40 94.66 94.5894.57 94.60 ̲ 95.8095.8097.5297.45 97.53 ̲ 97.61
Time(s) 75.93 ̲ 5.81 91.89109.3106.9462.3460.6 70.98 ̲ 69.86 116.8110.7
Table 8. OA as a function of the number of training pixels per class for different classification methods when applied to the two AVIRIS datasets. The standard deviation (in parentheses) of the ten random tests is also reported in each case. For both the pixel-wise classification and spatial-spectral classification, the best results are highlighted in bold, and the second-best results are underlined.
Table 8. OA as a function of the number of training pixels per class for different classification methods when applied to the two AVIRIS datasets. The standard deviation (in parentheses) of the ten random tests is also reported in each case. For both the pixel-wise classification and spatial-spectral classification, the best results are highlighted in bold, and the second-best results are underlined.
Pixel-Wise ClassificationSpatial-Spectral Classification
KSRCKCRCKNLSKFCLSJRMCJRMPRMCPRM
distprobdistprobdistprob
Indian Pines
1% 66.47 66.12 66.39 ̲ 66.3665.8369.9670.5372.2874.98 76.50 ̲ 79.11
(1.83)(1.83)(1.78)(1.80)(1.95)(2.18)(2.04)(2.36)(2.73)(2.82)(2.80)
3%77.5975.67 77.90 77.90 ̲ 77.5582.1782.8985.49 89.42 ̲ 87.49 89.74
(1.12)(1.10)(1.21)(1.00)(1.00)(0.93)(0.81)(0.69)(0.71)(0.59)(0.72)
5%81.3478.99 81.61 ̲ 81.72 81.4685.5386.1589.50 92.25 ̲ 91.00 92.86
(0.52)(0.75)(0.56)(0.42)(0.57)(0.36)(0.33)(0.81)(1.34)(0.79)(1.18)
10%85.9283.25 86.22 ̲ 86.23 85.9390.4090.9893.2793.85 94.37 ̲ 95.75
(0.59)(0.63)(0.57)(0.58)(0.51)(0.64)(0.62)(0.77)(0.84)(0.74)(0.76)
15%88.3885.98 88.69 88.68 ̲ 88.4092.4792.8895.54 96.75 ̲ 95.98 96.82
(0.28)(0.20)(0.30)(0.31)(0.41)(0.60)(0.58)(0.48)(0.37)(0.53)(0.31)
20%89.4186.81 89.70 ̲ 89.70 89.4193.1793.5096.05 96.34 ̲ 96.29 97.10
(0.41)(0.46)(0.43)(0.41)(0.36)(0.44)(0.49)(0.26)(0.45)(0.28)(0.21)
Kennedy Space Center
1% 80.26 80.24 ̲ 80.1680.1079.9782.3782.4984.7185.19 85.94 ̲ 87.02
(2.03)(2.18)(2.09)(2.09)(2.30)(2.32)(2.43)(2.54)(2.76)(2.40)(2.67)
3%86.6186.04 86.72 86.69 ̲ 86.6988.3188.5891.7992.29 92.47 ̲ 92.86
(0.89)(1.09)(0.78)(0.78)(0.97)(1.50)(1.53)(1.56)(1.80)(1.69)(1.71)
5% 88.45 87.7688.32 88.33 ̲ 88.2989.9790.0093.56 94.26 ̲ 94.08 94.60
(0.93)(0.85)(1.02)(1.02)(0.97)(1.52)(1.56)(2.08)(2.17)(1.83)(1.86)
10%91.0089.8891.12 91.13 ̲ 91.13 92.9393.0796.7097.16 97.90 ̲ 98.34
(0.57)(0.33)(0.50)(0.51)(0.53)(0.57)(0.56)(1.42)(1.48)(0.60)(0.69)
15%91.9890.63 92.15 ̲ 92.15 92.1294.3194.4098.28 98.60 ̲ 98.49 98.86
(0.53)(0.41)(0.45)(0.44)(0.44)(0.56)(0.57)(0.59)(0.56)(0.47)(0.48)
20%92.7291.21 92.87 92.84 92.87 ̲ 95.0995.1899.23 99.45 99.18 99.42 ̲
(0.40)(0.44)(0.52)(0.50)(0.47)(0.62)(0.58)(0.33)(0.30)(0.26)(0.24)
Table 9. OA as a function of the number of training pixels per class for different classification methods when applied to the two ROSIS datasets. The standard deviation (in parentheses) of the ten random tests is also reported in each case. For both the pixel-wise classification and spatial-spectral classification, the best results are highlighted in bold, and the second-best results are underlined.
Table 9. OA as a function of the number of training pixels per class for different classification methods when applied to the two ROSIS datasets. The standard deviation (in parentheses) of the ten random tests is also reported in each case. For both the pixel-wise classification and spatial-spectral classification, the best results are highlighted in bold, and the second-best results are underlined.
Pixel-Wise ClassificationSpatial-Spectral Classification
KSRCKCRCKNLSKFCLSJRMCJRMPRMCPRM
distprobdistprobdistprob
University of Pavia
1071.5072.33 72.44 72.36 ̲ 71.5276.5776.6284.38 85.62 ̲ 85.59 87.85
(4.17)(4.21)(4.39)(4.48)(4.56)(5.41)(5.39)(5.48)(5.96)(6.91)(5.69)
2077.5077.81 78.14 78.13 ̲ 77.7883.1183.2191.30 92.45 ̲ 91.97 93.44
(2.33)(2.14)(2.24)(2.25)(2.24)(2.92)(2.90)(2.71)(3.61)(3.44)(3.49)
4082.9683.05 83.57 83.55 ̲ 83.3988.4588.7196.13 97.19 ̲ 96.60 97.65
(1.05)(1.14)(1.17)(1.19)(1.30)(1.84)(1.92)(1.64)(1.47)(2.07)(1.80)
6085.4285.43 85.78 85.78 ̲ 85.6390.8090.9396.63 97.38 ̲ 97.16 98.27
(1.26)(1.43)(1.23)(1.25)(1.15)(1.39)(1.40)(1.02)(1.03)(0.76)(0.64)
8086.3886.26 86.78 86.77 ̲ 86.6891.3391.3797.80 98.30 ̲ 98.17 98.74
(0.75)(0.97)(0.91)(0.91)(0.99)(0.79)(0.86)(0.83)(0.74)(0.62)(0.63)
10087.7087.54 88.05 88.04 ̲ 87.9092.3492.4198.1298.40 98.59 ̲ 99.14
(0.82)(0.89)(0.68)(0.68)(0.61)(0.90)(0.89)(0.56)(0.49)(0.70)(0.24)
Center of Pavia
594.4694.71 94.82 ̲ 94.79 94.93 95.9796.1496.43 96.63 ̲ 96.60 96.71
(0.99)(0.95)(0.74)(0.75)(0.76)(0.74)(0.69)(0.74)(0.71)(0.83)(0.81)
1095.50 96.03 95.8995.88 95.95 ̲ 96.8696.9397.5097.49 97.52 ̲ 97.53
(0.53)(0.41)(0.46)(0.47)(0.38)(0.47)(0.45)(0.31)(0.32)(0.43)(0.41)
2096.74 96.89 96.8496.84 96.85 ̲ 97.5697.5698.5698.52 98.57 ̲ 98.61
(0.42)(0.43)(0.38)(0.38)(0.41)(0.48)(0.51)(0.44)(0.47)(0.49)(0.48)
4097.5697.55 97.58 ̲ 97.58 97.5798.1198.1198.7798.25 99.06 ̲ 99.10
(0.21)(0.16)(0.22)(0.22)(0.24)(0.24)(0.24)(0.26)(0.39)(0.24)(0.26)
6097.9197.80 97.93 ̲ 97.94 97.9298.4298.4199.0798.99 99.33 ̲ 99.36
(0.19)(0.20)(0.21)(0.21)(0.20)(0.25)(0.25)(0.30)(0.32)(0.30)(0.29)
8098.1097.91 98.12 ̲ 98.12 98.1198.5598.5499.2299.11 99.49 ̲ 99.52
(0.12)(0.17)(0.11)(0.11)(0.10)(0.17)(0.17)(0.17)(0.21)(0.13)(0.16)

Share and Cite

MDPI and ACS Style

Liu, J.; Wu, Z.; Xiao, Z.; Yang, J. Classification of Hyperspectral Images Using Kernel Fully Constrained Least Squares. ISPRS Int. J. Geo-Inf. 2017, 6, 344. https://doi.org/10.3390/ijgi6110344

AMA Style

Liu J, Wu Z, Xiao Z, Yang J. Classification of Hyperspectral Images Using Kernel Fully Constrained Least Squares. ISPRS International Journal of Geo-Information. 2017; 6(11):344. https://doi.org/10.3390/ijgi6110344

Chicago/Turabian Style

Liu, Jianjun, Zebin Wu, Zhiyong Xiao, and Jinlong Yang. 2017. "Classification of Hyperspectral Images Using Kernel Fully Constrained Least Squares" ISPRS International Journal of Geo-Information 6, no. 11: 344. https://doi.org/10.3390/ijgi6110344

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop