3D Octave and 2D Vanilla Mixed Convolutional Neural Network for Hyperspectral Image Classification with Limited Samples
Abstract
:1. Introduction
- Aiming at the classification of tensor-type hyperspectral images, we design 3D octave and 2D vanilla mixed convolutions in order to mine potential spatial–spectral features. Specifically, we first decompose the feature maps into different frequency components, and then apply 3D convolutions to accomplish the complementation of inter-frequency characteristics. Finally, the 2D vanilla convolution is attached to fuse along the channel direction of the feature maps, which reduces the output dimension and improves the generalization performance.
- Note that the final feature maps are sent to the classifier along the channel dimension, that is, the information at the same spatial location is discretely distributed in the vector form. Therefore, we propose the homology-shifting operation to aggregate the information of the same spatial location along the channel direction to ensure more compact features. It is commendable that homology shifting can enhance the generalization performance and stability of the model without any computational consumption.
- Extensive experiments are conducted on four HSI benchmark datasets with small-sized labeling samples. The results show that the proposed Oct-MCNN-HS model outperforms other state-of-the-art deep learning-based approaches in terms of both efficacy and efficiency. The proposed model with optimized parameters has been uploaded online at https://github.com/, accessed on ZhengJianwei2/Oct-MCNN-HS, whose source code will be coming soon after the review phase.
2. Related Work
2.1. 3D-2D Mixed Convolutions
2.2. Covariance Pooling
2.3. 2D Octave Convolution
3. Methodology
3.1. 3D Octave and 2D Vanilla Mixed Convolutions
3.2. Homology Shifting
4. Experiments
4.1. Data Description
- (1)
- Indian Pines (IP): The first dataset was captured in 1992 by the AVIRIS sensor over an agricultural area in northwestern Indiana. The image includes pixels and 200 spectral bands (24 channels are eliminated owing to noise) in the wavelength range of 0.4 to 2.5 m. In this scene, there are 16 representative land-cover categories to be classified.
- (2)
- University of Houston (UH): The second dataset was gathered in 2017 by the NCALM instrument over the University of Houston campus and its neighborhood, comprising 50 spectral channels with a spectral coverage ranging from 0.38 to 1.05 m. With pixels, this dataset is composed of 20 representative urban land-cover/land-use types.
- (3)
- University of Pavia (UP): The third dataset was collected in 2002 by the ROSIS sensor from northern Italy. The image contains spatial resolution and 103 (12 bands are removed due to the noise) bands with a spectral coverage ranging from 0.43 to 0.86 m. Approximately 42,776 labeled samples with 9 classes are extracted from the ground-truth image.
- (4)
- Salinas Scene (SA): The fourth dataset was acquired by the AVIRIS sensor over the Salinas Valley, California, containing 204 spectral bands (20 water absorption bands are discarded) and pixels with the spatial resolution of 3.7 m. The image is divided into 16 ground-truth classes with a total of 54,129 labeled pixels.
4.2. Framework Setting
4.3. Experimental Setup
4.4. Performance Comparison
4.5. Discussion
5. Conclusions
Author Contributions
Funding
Institutional Review Board Statement
Informed Consent Statement
Data Availability Statement
Conflicts of Interest
References
- Kuras, A.; Brell, M.; Rizzi, J.; Burud, I. Hyperspectral and Lidar Data Applied to the Urban Land Cover Machine Learning and Neural-Network-Based Classification: A Review. Remote Sens. 2021, 13, 3393. [Google Scholar] [CrossRef]
- Scafutto, R.D.M.; de Souza Filho, C.R.; de Oliveira, W.J. Hyperspectral remote sensing detection of petroleum hydrocarbons in mixtures with mineral substrates: Implications for onshore exploration and monitoring. ISPRS J. Photogramm. Remote Sens. 2017, 128, 146–157. [Google Scholar] [CrossRef]
- Bai, C.; Zhang, M.; Zhang, J.; Zheng, J.; Chen, S. LSCIDMR: Large-scale Satellite Cloud Image Database for Meteorological Research. IEEE Trans. Cybern. 2021, 1–13. [Google Scholar] [CrossRef]
- Pandey, P.; Payn, K.G.; Lu, Y.; Heine, A.J.; Walker, T.D.; Acosta, J.J.; Young, S. Hyperspectral Imaging Combined with Machine Learning for the Detection of Fusiform Rust Disease Incidence in Loblolly Pine Seedlings. Remote Sens. 2021, 13, 3595. [Google Scholar] [CrossRef]
- Makki, I.; Younes, R.; Francis, C.; Bianchi, T.; Zucchetti, M. A survey of landmine detection using hyperspectral imaging. ISPRS J. Photogramm. Remote Sens. 2017, 124, 40–53. [Google Scholar] [CrossRef]
- Ma, L.; Crawford, M.M.; Tian, J. Local Manifold Learning-Based k -Nearest-Neighbor for Hyperspectral Image Classification. IEEE Trans. Geosci. Remote Sens. 2010, 48, 4099–4109. [Google Scholar] [CrossRef]
- Delalieux, S.; Somers, B.; Haest, B.; Spanhove, T.; Vanden Borre, J.; Mücher, C. Heathland conservation status mapping through integration of hyperspectral mixture analysis and decision tree classifiers. Remote Sens. Environ. 2012, 126, 222–231. [Google Scholar] [CrossRef]
- Yu, X.; Feng, Y.; Gao, Y.; Jia, Y.; Mei, S. Dual-Weighted Kernel Extreme Learning Machine for Hyperspectral Imagery Classification. Remote Sens. 2021, 13, 508. [Google Scholar] [CrossRef]
- 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] [Green Version]
- Zhang, Y.; Cao, G.; Li, X.; Wang, B.; Fu, P. Active Semi-Supervised Random Forest for Hyperspectral Image Classification. Remote Sens. 2019, 11, 2974. [Google Scholar] [CrossRef] [Green Version]
- Han, Y.; Shi, X.; Yang, S.; Zhang, Y.; Hong, Z.; Zhou, R. Hyperspectral Sea Ice Image Classification Based on the Spectral-Spatial-Joint Feature with the PCA Network. Remote Sens. 2021, 13, 2253. [Google Scholar] [CrossRef]
- Wang, J.; Chang, C.I. Independent component analysis-based dimensionality reduction with applications in hyperspectral image analysis. IEEE Trans. Geosci. Remote Sens. 2006, 44, 1586–1600. [Google Scholar] [CrossRef]
- Chen, M.; Wang, Q.; Li, X. Discriminant Analysis with Graph Learning for Hyperspectral Image Classification. Remote Sens. 2018, 10, 836. [Google Scholar] [CrossRef] [Green Version]
- Cui, B.; Cui, J.; Lu, Y.; Guo, N.; Gong, M. A Sparse Representation-Based Sample Pseudo-Labeling Method for Hyperspectral Image Classification. Remote Sens. 2020, 12, 664. [Google Scholar] [CrossRef] [Green Version]
- Cao, X.; Xu, Z.; Meng, D. Spectral-Spatial Hyperspectral Image Classification via Robust Low-Rank Feature Extraction and Markov Random Field. Remote Sens. 2019, 11, 1565. [Google Scholar] [CrossRef] [Green Version]
- Kang, X.; Li, S.; Benediktsson, J.A. Spectral-Spatial Hyperspectral Image Classification With Edge-Preserving Filtering. IEEE Trans. Geosci. Remote Sens. 2014, 52, 2666–2677. [Google Scholar] [CrossRef]
- Paoletti, M.; Haut, J.; Plaza, J.; Plaza, A. Deep learning classifiers for hyperspectral imaging: A review. ISPRS J. Photogramm. Remote Sens. 2019, 158, 279–317. [Google Scholar] [CrossRef]
- Li, S.; Song, W.; Fang, L.; Chen, Y.; Ghamisi, P.; Benediktsson, J.A. Deep Learning for Hyperspectral Image Classification: An Overview. IEEE Trans. Geosci. Remote Sens. 2019, 57, 6690–6709. [Google Scholar] [CrossRef] [Green Version]
- Madani, H.; McIsaac, K. Distance Transform-Based Spectral-Spatial Feature Vector for Hyperspectral Image Classification with Stacked Autoencoder. Remote Sens. 2021, 13, 1732. [Google Scholar] [CrossRef]
- Li, T.; Zhang, J.; Zhang, Y. Classification of hyperspectral image based on deep belief networks. In Proceedings of the 2014 IEEE International Conference on Image Processing (ICIP), Paris, France, 27–30 October 2014; pp. 5132–5136. [Google Scholar] [CrossRef]
- 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] [Green Version]
- Fang, L.; Liu, G.; Li, S.; Ghamisi, P.; Benediktsson, J.A. Hyperspectral Image Classification With Squeeze Multibias Network. IEEE Trans. Geosci. Remote Sens. 2019, 57, 1291–1301. [Google Scholar] [CrossRef]
- Song, W.; Li, S.; Fang, L.; Lu, T. Hyperspectral Image Classification With Deep Feature Fusion Network. IEEE Trans. Geosci. Remote Sens. 2018, 56, 3173–3184. [Google Scholar] [CrossRef]
- Bai, C.; Huang, L.; Pan, X.; Zheng, J.; Chen, S. Optimization of deep convolutional neural network for large scale image retrieval. Neurocomputing 2018, 303, 60–67. [Google Scholar] [CrossRef]
- Xu, Q.; Xiao, Y.; Wang, D.; Luo, B. CSA-MSO3DCNN: Multiscale Octave 3D CNN with Channel and Spatial Attention for Hyperspectral Image Classification. Remote Sens. 2020, 12, 188. [Google Scholar] [CrossRef] [Green Version]
- Zhong, Z.; Li, J.; Luo, Z.; Chapman, M. Spectral-Spatial Residual Network for Hyperspectral Image Classification: A 3-D Deep Learning Framework. IEEE Trans. Geosci. Remote Sens. 2018, 56, 847–858. [Google Scholar] [CrossRef]
- Haut, J.M.; Paoletti, M.E.; Plaza, J.; Li, J.; Plaza, A. Active Learning With Convolutional Neural Networks for Hyperspectral Image Classification Using a New Bayesian Approach. IEEE Trans. Geosci. Remote Sens. 2018, 56, 6440–6461. [Google Scholar] [CrossRef]
- Yang, J.; Zhao, Y.Q.; Chan, J.C.W. Learning and Transferring Deep Joint Spectral-Spatial Features for Hyperspectral Classification. IEEE Trans. Geosci. Remote Sens. 2017, 55, 4729–4742. [Google Scholar] [CrossRef]
- Xu, X.; Li, W.; Ran, Q.; Du, Q.; Gao, L.; Zhang, B. Multisource Remote Sensing Data Classification Based on Convolutional Neural Network. IEEE Trans. Geosci. Remote Sens. 2018, 56, 937–949. [Google Scholar] [CrossRef]
- Zheng, J.; Feng, Y.; Bai, C.; Zhang, J. Hyperspectral Image Classification Using Mixed Convolutions and Covariance Pooling. IEEE Trans. Geosci. Remote Sens. 2021, 59, 522–534. [Google Scholar] [CrossRef]
- Yang, B.; Bender, G.; Le, Q.V.; Ngiam, J. CondConv: Conditionally Parameterized Convolutions for Efficient Inference. arXiv Prepr. 2019, arXiv:1904.04971. [Google Scholar]
- Zhang, Q.; Jiang, Z.; Lu, Q.; Han, J.; Zeng, Z.; Gao, S.H.; Men, A. Split to Be Slim: An Overlooked Redundancy in Vanilla Convolution. arXiv 2020, arXiv:2006.12085. [Google Scholar] [CrossRef]
- Zhang, C.; Wang, J.; Yao, K. Global Random Graph Convolution Network for Hyperspectral Image Classification. Remote Sens. 2021, 13, 2285. [Google Scholar] [CrossRef]
- Pu, S.; Wu, Y.; Sun, X.; Sun, X. Hyperspectral Image Classification with Localized Graph Convolutional Filtering. Remote Sens. 2021, 13, 526. [Google Scholar] [CrossRef]
- Ma, A.; Filippi, A.M.; Wang, Z.; Yin, Z. Hyperspectral Image Classification Using Similarity Measurements-Based Deep Recurrent Neural Networks. Remote Sens. 2019, 11, 194. [Google Scholar] [CrossRef] [Green Version]
- Mei, X.; Pan, E.; Ma, Y.; Dai, X.; Huang, J.; Fan, F.; Du, Q.; Zheng, H.; Ma, J. Spectral-Spatial Attention Networks for Hyperspectral Image Classification. Remote Sens. 2019, 11, 963. [Google Scholar] [CrossRef] [Green Version]
- Seydgar, M.; Alizadeh Naeini, A.; Zhang, M.; Li, W.; Satari, M. 3-D Convolution-Recurrent Networks for Spectral-Spatial Classification of Hyperspectral Images. Remote Sens. 2019, 11, 883. [Google Scholar] [CrossRef] [Green Version]
- Wang, Q.; Wang, J.; Zhou, M.; Li, Q.; Wen, Y.; Chu, J. A 3D attention networks for classification of white blood cells from microscopy hyperspectral images. Opt. Laser Technol. 2021, 139, 106931. [Google Scholar] [CrossRef]
- Hang, R.; Li, Z.; Liu, Q.; Ghamisi, P.; Bhattacharyya, S.S. Hyperspectral Image Classification With Attention-Aided CNNs. IEEE Trans. Geosci. Remote Sens. 2021, 59, 2281–2293. [Google Scholar] [CrossRef]
- Qing, Y.; Liu, W. Hyperspectral Image Classification Based on Multi-Scale Residual Network with Attention Mechanism. Remote Sens. 2021, 13, 335. [Google Scholar] [CrossRef]
- Xue, Z.; Zhang, M.; Liu, Y.; Du, P. Attention-Based Second-Order Pooling Network for Hyperspectral Image Classification. IEEE Trans. Geosci. Remote Sens. 2021, 59, 9600–9615. [Google Scholar] [CrossRef]
- Chen, Y.; Fan, H.; Xu, B.; Yan, Z.; Kalantidis, Y.; Rohrbach, M.; Shuicheng, Y.; Feng, J. Drop an Octave: Reducing Spatial Redundancy in Convolutional Neural Networks With Octave Convolution. In Proceedings of the IEEE/CVF International Conference on Computer Vision, Seoul, Korea, 27 October–2 November 2019. pp. 3435–3444. [CrossRef] [Green Version]
- Roy, S.K.; Krishna, G.; Dubey, S.R.; Chaudhuri, B.B. HybridSN: Exploring 3-D-2-D CNN Feature Hierarchy for Hyperspectral Image Classification. IEEE Geosci. Remote Sens. Lett. 2020, 17, 277–281. [Google Scholar] [CrossRef] [Green Version]
- Zhong, S.; Chang, C.I.; Li, J.; Shang, X.; Chen, S.; Song, M.; Zhang, Y. Class Feature Weighted Hyperspectral Image Classification. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2019, 12, 4728–4745. [Google Scholar] [CrossRef]
MAP | Map | Class Name | Train | Test | Total |
---|---|---|---|---|---|
Background | - | - | - | ||
Alfalfa | 5 | 41 | 46 | ||
Corn-no till | 5 | 1423 | 1428 | ||
Corn-min till | 5 | 825 | 830 | ||
Corn | 5 | 232 | 237 | ||
Grass-pasture | 5 | 478 | 483 | ||
Grass-trees | 5 | 725 | 730 | ||
Grass-pasture-mowed | 5 | 23 | 28 | ||
Background | 5 | 473 | 478 | ||
Oats | 5 | 15 | 20 | ||
Soybean-no till | 5 | 967 | 972 | ||
Soybean-min till | 5 | 2450 | 2455 | ||
Soybean-clean | 5 | 588 | 593 | ||
Wheat | 5 | 200 | 205 | ||
Woods | 5 | 1260 | 1265 | ||
Bldg-grass-trees-drives | 5 | 381 | 386 | ||
Stone-steel-towers | 5 | 88 | 93 | ||
Total samples | 80 | 10,169 | 10,249 |
MAP | Map | Class Name | Train | Test | Total |
---|---|---|---|---|---|
Background | - | - | - | ||
Healthy grass | 101 | 9698 | 9799 | ||
Stressed grass | 102 | 32,400 | 32,502 | ||
Artificial turf | 101 | 583 | 684 | ||
Evergreen trees | 102 | 13,493 | 13,595 | ||
Deciduous trees | 101 | 4920 | 5021 | ||
Bare earth | 101 | 4415 | 4516 | ||
Water | 101 | 165 | 266 | ||
Residential buildings | 102 | 39,670 | 39,772 | ||
Non-residential buildings | 102 | 223,650 | 223,752 | ||
Roads | 102 | 45,764 | 45,866 | ||
Sidewalks | 102 | 33,927 | 34,029 | ||
Crosswalks | 101 | 1417 | 1518 | ||
Major thoroughfares | 102 | 46,246 | 46,348 | ||
Highways | 101 | 9764 | 9865 | ||
Railways | 101 | 6836 | 6937 | ||
Paved parking lots | 102 | 11,398 | 11,500 | ||
Unpaved parking lots | 73 | 73 | 146 | ||
Cars | 101 | 6446 | 6547 | ||
Trains | 101 | 5268 | 5369 | ||
Stadium seats | 101 | 6723 | 6824 | ||
Total samples | 2000 | 502,856 | 504,856 |
MAP | Map | Class Name | Train/Val | Test | Total |
---|---|---|---|---|---|
Background | - | - | - | ||
Asphalt | 7 | 6617 | 6631 | ||
Meadows | 18 | 18,613 | 18,649 | ||
Corn-min till | 2 | 2095 | 2099 | ||
Corn | 3 | 3058 | 3064 | ||
Grass-pasture | 1 | 1343 | 1345 | ||
Grass-trees | 5 | 5019 | 5029 | ||
Grass-pasture-mowed | 1 | 1328 | 1330 | ||
Hay-windrowed | 4 | 3674 | 3682 | ||
Oats | 1 | 945 | 947 | ||
Total samples | 42 | 42,692 | 42,776 |
MAP | Map | Class Name | Train/Val | Test | Total |
---|---|---|---|---|---|
Background | - | - | - | ||
Alfalfa | 2 | 2005 | 2009 | ||
Corn-no till | 4 | 3718 | 3726 | ||
Corn-min till | 2 | 1972 | 1976 | ||
Corn | 1 | 1392 | 1394 | ||
Grass-pasture | 3 | 2672 | 2678 | ||
Grass-trees | 4 | 3951 | 3959 | ||
Grass-pasture-mowed | 4 | 3571 | 3579 | ||
Hay-windrowed | 11 | 11,249 | 11,271 | ||
Oats | 6 | 6191 | 6203 | ||
Soybean-no till | 3 | 3272 | 3278 | ||
Soybean-min till | 1 | 1066 | 1068 | ||
Soybean-clean | 2 | 1923 | 1927 | ||
Wheat | 1 | 914 | 916 | ||
Woods | 1 | 1068 | 1070 | ||
Bldg-grass-trees-drives | 7 | 7254 | 7268 | ||
Stone-steel-towers | 2 | 1803 | 1807 | ||
Total samples | 54 | 54,021 | 54,129 |
Layer (Kernel Size) | Output Shape | Parameters | Connected To | |
---|---|---|---|---|
(1) | Input_Data | (145, 145, 200) | - | - |
(2) | Preprocessing_Layer | (11, 11, 110, 1) | 0 | (1) |
(3) | 3D_Convolution (8, 3, 3, 3) | (11, 11, 110, 8) | 224 | (2) |
(4) | Average_Pooling | (5, 5, 110, 1) | 0 | (2) |
(5) | 3D_Convolution (8, 3, 3, 3) | (5, 5, 110, 8) | 224 | (4) |
(6) | 3D_Convolution (16, 3, 3, 3) | (11, 11, 110, 16) | 3472 | (3) |
(7) | 3D_Convolution (16, 3, 3, 3) | (5, 5, 110, 16) | 3472 | (5) |
(8) | Average_Pooling | (5, 5, 110, 8) | 0 | (3) |
(9) | 3D_Convolution (16, 3, 3, 3) | (5, 5, 110, 16) | 3472 | (8) |
(10) | 3D_Convolution (16, 3, 3, 3) | (5, 5, 110, 16) | 3472 | (5) |
(11) | Up_Sampling | (11, 11, 110, 16) | 0 | (10) |
(12) | Add | (11, 11, 110, 16) | 0 | (6), (11) |
(13) | Add | (5, 5, 110, 16) | 0 | (7), (9) |
(14) | Average_Pooling | (5, 5, 110, 32) | 0 | (12) |
(15) | 3D_Convolution (32, 3, 3, 3) | (5, 5, 110, 32) | 13,856 | (13) |
(16) | 3D_Convolution (32, 3, 3, 3) | (5, 5, 110, 32) | 13,856 | (14) |
(17) | Add | (5, 5, 110, 32) | 0 | (15), (16) |
(18) | Reshape | (5, 5, 3520) | 0 | (17) |
(19) | 2D_Convolution (512, 1, 1) | (5, 5, 512) | 1,802,752 | (18) |
(20) | Homology_Shifting | (80, 80, 2) | 0 | (19) |
(21) | Flatten_Layer | (12,800) | 0 | (20) |
(22) | Fully_Connected_Layer | (256) | 3,277,056 | (21) |
(23) | Fully_Connected_Layer | (128) | 32,896 | (22) |
(24) | Fully_Connected_Layer | (16) | 2064 | (23) |
Total Parameters: | 5,156,816 |
Class. | HybridSN | SSRN | A-SPN | MCNN-CP | Oct-MCNN-HS |
---|---|---|---|---|---|
1 | 100 | 100 | 100 | 100 | 100 |
2 | 55.66 | 36.26 | 68.58 | 69.99 | 70.12 |
3 | 61.45 | 85.58 | 57.69 | 52.00 | 70.17 |
4 | 22.84 | 78.45 | 98.27 | 68.10 | 79.31 |
5 | 65.90 | 71.55 | 86.61 | 80.75 | 87.87 |
6 | 94.90 | 89.93 | 91.58 | 95.72 | 98.48 |
7 | 100 | 100 | 100 | 100 | 100 |
8 | 65.75 | 99.58 | 94.08 | 95.77 | 100 |
9 | 100 | 100 | 100 | 100 | 100 |
10 | 18.38 | 60.39 | 74.97 | 62.77 | 69.98 |
11 | 49.10 | 27.31 | 62.97 | 54.78 | 71.96 |
12 | 11.23 | 59.69 | 63.29 | 65.14 | 83.16 |
13 | 100 | 90.00 | 100 | 100 | 100 |
14 | 68.33 | 83.81 | 100 | 65.63 | 97.62 |
15 | 62.73 | 68.24 | 88.42 | 83.46 | 80.02 |
16 | 98.86 | 100 | 100 | 100 | 100 |
OA | 53.53 ±5.24 | 60.34 ±8.78 | 77.22 ±1.46 | 68.46 ±5.84 | 78.06 ±3.23 |
AA | 66.75 ±6.63 | 74.17 ±5.41 | 87.61 ±0.89 | 80.08 ±3.01 | 86.89 ±2.85 |
Kappa | 47.87 ±5.66 | 56.41 ±9.88 | 74.36 ±1.51 | 64.41 ±6.26 | 75.27 ±3.33 |
Class. | HybridSN | SSRN | A-SPN | MCNN-CP | Oct-MCNN-HS |
---|---|---|---|---|---|
1 | 92.61 | 84.65 | 88.26 | 89.72 | 88.30 |
2 | 77.88 | 88.04 | 84.11 | 79.15 | 77.90 |
3 | 99.83 | 100 | 100 | 100 | 100 |
4 | 92.00 | 97.93 | 96.50 | 95.95 | 96.72 |
5 | 94.37 | 94.67 | 96.80 | 96.38 | 97.05 |
6 | 100 | 100 | 100 | 99.68 | 100 |
7 | 100 | 100 | 100 | 100 | 100 |
8 | 70.18 | 83.70 | 91.05 | 87.08 | 91.35 |
9 | 75.74 | 74.59 | 78.18 | 80.50 | 84.10 |
10 | 49.55 | 47.31 | 40.81 | 59.16 | 52.79 |
11 | 45.55 | 63.31 | 52.06 | 51.61 | 55.00 |
12 | 63.59 | 67.25 | 73.32 | 74.45 | 86.73 |
13 | 60.16 | 61.94 | 60.45 | 65.98 | 74.86 |
14 | 91.83 | 97.54 | 96.51 | 95.43 | 97.96 |
15 | 99.15 | 99.65 | 98.94 | 99.85 | 99.22 |
16 | 94.61 | 97.53 | 95.15 | 87.08 | 95.53 |
17 | 100 | 100 | 100 | 100 | 100 |
18 | 94.45 | 95.07 | 92.02 | 96.88 | 97.38 |
19 | 93.77 | 99.28 | 93.45 | 98.63 | 99.03 |
20 | 99.78 | 99.64 | 100 | 99.08 | 99.94 |
OA | 72.56 ±1.53 | 77.22 ±2.44 | 75.56 ±0.35 | 78.68 ±0.67 | 81.37 ±0.24 |
AA | 84.75 ±1.12 | 87.65 ±1.53 | 86.98 ±0.19 | 87.79 ±0.18 | 88.94 ±0.10 |
Kappa | 66.16 ±1.48 | 69.49 ±2.39 | 69.81 ±0.38 | 71.89 ±0.79 | 74.75 ±0.26 |
Class. | HybridSN | SSRN | A-SPN | MCNN-CP | Oct-MCNN-HS |
---|---|---|---|---|---|
1 | 78.66 | 93.56 | 96.22 | 74.16 | 89.21 |
2 | 94.29 | 92.90 | 98.23 | 99.64 | 99.87 |
3 | 4.92 | 0 | 9.82 | 23.91 | 41.26 |
4 | 69.82 | 70.73 | 69.45 | 82.63 | 87.93 |
5 | 100 | 100 | 100 | 100 | 100 |
6 | 45.77 | 45.35 | 39.45 | 55.79 | 66.45 |
7 | 10.47 | 28.70 | 7.97 | 95.70 | 46.90 |
8 | 33.64 | 79.56 | 30.64 | 38.21 | 62.16 |
9 | 15.98 | 56.08 | 6.37 | 14.29 | 24.13 |
OA | 72.03 ±4.28 | 76.16 ±4.12 | 73.86 ±1.46 | 78.23 ±2.87 | 81.79 ±1.61 |
AA | 51.39 ±8.24 | 61.71 ±5.52 | 51.20 ±1.71 | 62.36 ±3.71 | 63.56 ±2.34 |
Kappa | 61.25 ±4.37 | 68.52 ±4.63 | 63.23 ±1.68 | 70.69 ±3.01 | 73.14 ±1.83 |
Class. | HybridSN | SSRN | A-SPN | MCNN-CP | Oct-MCNN-HS |
---|---|---|---|---|---|
1 | 100 | 93.87 | 94.46 | 97.31 | 100 |
2 | 100 | 100 | 100 | 83.03 | 100 |
3 | 93.51 | 57.15 | 48.07 | 71.86 | 96.96 |
4 | 49.71 | 96.98 | 15.57 | 89.94 | 81.62 |
5 | 98.91 | 78.93 | 99.81 | 88.77 | 99.25 |
6 | 99.37 | 99.42 | 100 | 98.71 | 99.49 |
7 | 100 | 100 | 97.59 | 99.64 | 100 |
8 | 56.18 | 34.08 | 93.40 | 85.02 | 80.98 |
9 | 100 | 100 | 100 | 98.22 | 100 |
10 | 90.56 | 66.01 | 88.36 | 93.37 | 97.65 |
11 | 50.38 | 0 | 99.25 | 99.91 | 100 |
12 | 50.75 | 82.52 | 51.42 | 91.94 | 95.68 |
13 | 17.94 | 93.36 | 86.66 | 46.06 | 96.83 |
14 | 80.15 | 86.99 | 92.79 | 76.14 | 83.71 |
15 | 48.18 | 65.20 | 42.28 | 63.54 | 72.83 |
16 | 82.36 | 74.88 | 98.89 | 98.17 | 98.23 |
OA | 77.75 ±3.31 | 79.91 ±2.08 | 83.57 ±1.22 | 86.41 ±0.89 | 90.63 ±0.47 |
AA | 76.12 ±1.53 | 80.41 ±1.26 | 81.78 ±1.54 | 86.35 ±1.24 | 93.23 ±0.64 |
Kappa | 74.01 ±3.75 | 75.72 ±2.27 | 81.53 ±1.41 | 84.82 ±0.98 | 89.65 ±0.48 |
Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations. |
© 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).
Share and Cite
Feng, Y.; Zheng, J.; Qin, M.; Bai, C.; Zhang, J. 3D Octave and 2D Vanilla Mixed Convolutional Neural Network for Hyperspectral Image Classification with Limited Samples. Remote Sens. 2021, 13, 4407. https://doi.org/10.3390/rs13214407
Feng Y, Zheng J, Qin M, Bai C, Zhang J. 3D Octave and 2D Vanilla Mixed Convolutional Neural Network for Hyperspectral Image Classification with Limited Samples. Remote Sensing. 2021; 13(21):4407. https://doi.org/10.3390/rs13214407
Chicago/Turabian StyleFeng, Yuchao, Jianwei Zheng, Mengjie Qin, Cong Bai, and Jinglin Zhang. 2021. "3D Octave and 2D Vanilla Mixed Convolutional Neural Network for Hyperspectral Image Classification with Limited Samples" Remote Sensing 13, no. 21: 4407. https://doi.org/10.3390/rs13214407