A New Architecture of a Complex-Valued Convolutional Neural Network for PolSAR Image Classification
Abstract
:1. Introduction
- (1)
- A novel CV-CNN is introduced in this study, featuring complex-valued inputs, outputs, as well as complex-valued weights and biases. Our nonlinear module treats the input as a complex number, respecting the mathematical significance of complex-valued inputs and extracting the most discriminative features, resulting in improved classification ability. Our new complex-valued methods are used in different deep learning models and achieve better results than real-valued or old complex-valued versions with the same structure.
- (2)
- In this research, a novel complex-valued max pooling technique is presented for the downsampling of feature maps. This method is designed to reduce computational demands, accelerate training and inference, and, importantly, retain the most essential features.
- (3)
- A novel complex-valued activation function is employed to acquire high-dimensional nonlinear features. This new activation maps the amplitude and phase of the features into the high-dimensional complex domain space and can make the model more sparse.
- (4)
- A novel complex-valued cross-entropy is applied in the training process of the new CV-CNN. The complex-valued probability principle [44,45,46,47,48] is employed to reallocate one-hot labels within the dataset. This loss function utilizes the complex-valued labels and outputs to compute the classification loss and train a better model by backpropagation.
2. Materials and Methods
2.1. Two Deep Learning Models for PolSAR Classification
2.2. Inputs of PolSAR Classification
2.3. Complex-Valued Amplitude-Based Max Pooling
2.4. Complex-Valued Nonlinear Activation
2.5. Complex-Valued Cross-Entropy
2.5.1. Complex-Valued Probability and CV_one-hot Label
2.5.2. Complex-Valued Cross-Entropy
2.6. Complex-Valued PolSAR Classification Algorithm
Algorithm 1: Complex-valued convolutional classification algorithm for PolSAR images |
Preprocessing: 1. Construction of complex-valued models for PolSAR image classification with CVA_Max_Pooling and HReLU 2. Assigning CV_one-hot labels to each pixel of the PolSAR dataset 3. Selection of training set from the PolSAR dataset Input: a training set and corresponding labels, learning rate, batch size, and momentum parameter 4. Repeat: 5. Calling CVA_Max_Pooling to obtain the most efficient features 6. Invoking HReLU to map the amplitude and phase of the feature to the nonlinear domain 7. Calling CV_CrossEntropy to compute the loss during training 8. Updating model parameters with loss 9. Until: Meeting the conditions for termination 10. Inferring the class of the entire PolSAR image with the trained model Output: Prediction of the testing set |
3. Experimental Results
3.1. PolSAR Dataset Description
3.1.1. Flevoland Dataset 1
3.1.2. Flevoland Dataset 2
3.1.3. Oberpfaffenhofen Dataset
3.2. Parameterization
3.3. Evaluation Metrics
3.4. Model Parameters
3.5. Analysis of Experimental Results
3.5.1. Flevoland Dataset 1 Results
3.5.2. Flevoland Dataset 2 Results
3.5.3. Oberpfaffenhofen Dataset Results
3.5.4. Computational Complexity of CNN
4. Discussion
4.1. Ablation Experiment 1: Performance of CVA_Max_Pooling
4.2. Ablation Experiment 2: Performance of HReLU
4.3. Ablation Experiment 3: Performance of CV_CrossEntropy
4.4. Comparison with State-of-the-Art Algorithms
5. Conclusions
Author Contributions
Funding
Data Availability Statement
Acknowledgments
Conflicts of Interest
References
- Lee, J.S.; Pottier, E. Polarimetric Radar Imaging: From Basic to Application; CRC Press: Boca Raton, FL, USA, 2011; pp. 1–22. [Google Scholar]
- Hänsch, R.; Hellwich, O. Skipping the real world: Classification of PolSAR images without explicit feature extraction. ISPRS J. Photogramm. Remote Sens. 2018, 140, 122–132. [Google Scholar] [CrossRef]
- Lee, J.S.; Grunes, M.R.; Kwok, R. Classification of multi-look polarimetric SAR imagery based on the complex Wishart distribution. Int. J. Remote Sens. 1994, 15, 2299–2311. [Google Scholar] [CrossRef]
- Lee, J.S.; Grunes, M.R.; Pottier, E.; Ferro-Famil, L. Unsupervised terrain classification preserving polarimetric scattering characteristics. IEEE Trans. Geosci. Remote Sens. 2004, 42, 722–731. [Google Scholar] [CrossRef]
- Dabboor, M.; Collins, M.; Karathanassi, V.; Braun, A. An unsupervised classification approach for polarimetric SAR data based on the Chernoff distance for the complex Wishart distribution. IEEE Trans. Geosci. Remote Sens. 2013, 51, 4200–4213. [Google Scholar] [CrossRef]
- Wu, Y.; Ji, K.; Yu, W.; Su, Y. Region-based classification of polarimetric SAR images using Wishart MRF. IEEE Geosci. Remote Sens. Lett. 2008, 5, 668–672. [Google Scholar] [CrossRef]
- Song, W.; Li, M.; Zhang, P.; Wu, Y.; Tan, X.; An, L. Mixture WGΓ-MRF model for PolSAR image classification. IEEE Trans. Geosci. Remote Sens. 2018, 56, 905–920. [Google Scholar] [CrossRef]
- Arii, M.; van Zyl, J.J.; Kim, Y. Adaptive model-based decomposition of polarimetric SAR covariance matrices. IEEE Trans. Geosci. Remote Sens. 2011, 49, 1104–1113. [Google Scholar] [CrossRef]
- Clound, S.R.; Pottier, E. An entropy based classification scheme for land applications of polarimetric SAR. IEEE Trans. Geosci. Remote Sens. 1997, 35, 68–78. [Google Scholar] [CrossRef]
- Cloude, S.R.; Pottier, E. A review of target decomposition theorems in radar polarimetry. IEEE Trans. Geosci. Remote Sens. 1996, 34, 498–518. [Google Scholar] [CrossRef]
- An, W.; Cui, Y.; Yang, J. Three-Component Model-Based Decomposition for Polarimetric SAR Data. IEEE Trans. Geosci. Remote Sens. 2010, 48, 2732–2739. [Google Scholar] [CrossRef]
- He, C.; Li, S.; Liao, Z.; Liao, M. Texture Classification of PolSAR Data Based on Sparse Coding of Wavelet Polarization Textons. IEEE Trans. Geosci. Remote Sens. 2013, 51, 4576–4590. [Google Scholar] [CrossRef]
- Lardeux, C.; Frison, P.L.; Tison, C.C.; Souyris, J.C.; Stoll, B.; Fruneau, B.; Rudant, J.P. Support vector machine for multifrequency SAR polarimetric data classification. IEEE Trans. Geosci. Remote Sens. 2009, 47, 4143–4152. [Google Scholar] [CrossRef]
- Melgani, F.; Hashemy, B.A.R.A.; Taha, S.M.R. An explicit fuzzy supervised classification method for multispectral remote sensing images. IEEE Trans. Geosci. Remote Sens. 2000, 38, 287–295. [Google Scholar] [CrossRef]
- Ulaby, F.T.; Elachi, C. Radar polaritnetry for geoscience applications. Geocarto Int. 1990, 5, 38. [Google Scholar] [CrossRef]
- Freeman, A.; Durden, S.L. A three-component scattering model for polarimetric SAR data. IEEE Trans. Geosci. Remote Sens. 1998, 36, 963–973. [Google Scholar] [CrossRef]
- Huynen, J.R. Phenomenological Theory of Radar Targets. Available online: http://resolver.tudelft.nl/uuid:e4a140a0-c175-45a7-ad41-29b28361b426 (accessed on 14 April 2022).
- De, S.; Bruzzone, L.; Bhattacharya, A.; Bovolo, F.; Chaudhuri, S. A Novel Technique Based on Deep Learning and a Synthetic Target Database for Classification of Urban Areas in PolSAR Data. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2018, 11, 154–170. [Google Scholar] [CrossRef]
- Zhou, Y.; Wang, H.; Xu, F.; Jin, Y.Q. Polarimetric SAR image classification using deep convolutional neural networks. IEEE Geosci. Remote Sens. Lett. 2016, 13, 1935–1939. [Google Scholar] [CrossRef]
- Bin, H.; Sun, J.; Xu, Z. A Graph-Based Semisupervised Deep Learning Model for PolSAR Image Classification. IEEE Trans. Geosci. Remote Sens. 2019, 57, 2116–2132. [Google Scholar] [CrossRef]
- Li, Y.; Chen, Y.; Liu, G.; Jiao, L. A novel deep fully convolutional network for PolSAR image classification. Remote Sens. 2018, 10, 1984. [Google Scholar] [CrossRef]
- Pham, M.; Lefevre, S. Very high resolution Airborne PolSAR Image Classification using Convolutional Neural Networks. In Proceedings of the 13th European Conference on Synthetic Aperture Radar (EUSAR 2021), Online, 29 March–1 April 2021; pp. 1–4. [Google Scholar]
- Liu, S.; Luo, H.; Shi, Q. Active Ensemble Deep Learning for Polarimetric Synthetic Aperture Radar Image Classification. IEEE Geosci. Remote Sens. Lett. 2021, 18, 1580–1584. [Google Scholar] [CrossRef]
- Cheng, J.; Zhang, F.; Xiang, D.; Yin, Q.; Zhou, Y. PolSAR Image Classification with Multiscale Superpixel-Based Graph Convolutional Network. IEEE Trans. Geosci. Remote Sens. 2022, 60, 1–14. [Google Scholar] [CrossRef]
- Liu, G.; Li, Y.; Jiao, L.; Chen, Y.; Shang, R. Multiobjective evolutionary algorithm assisted stacked autoencoder for PolSAR image classification. Swarm Evol. Comput. 2021, 60, 100794. [Google Scholar] [CrossRef]
- Jing, H.; Wang, Z.; Sun, X.; Xiao, D.; Fu, K. PSRN: Polarimetric Space Reconstruction Network for PolSAR Image Semantic Segmentation. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2021, 14, 10716–10732. [Google Scholar] [CrossRef]
- Nie, W.; Huang, K.; Yang, J.; Li, P. A Deep Reinforcement Learning-Based Framework for PolSAR Imagery Classification. IEEE Trans. Geosci. Remote Sens. 2022, 60, 1–15. [Google Scholar] [CrossRef]
- Yang, C.; Hou, B.; Chanussot, J.; Hu, Y.; Ren, B.; Wang, S.; Jiao, L. N-Cluster Loss and Hard Sample Generative Deep Metric Learning for PolSAR Image Classification. IEEE Trans. Geosci. Remote Sens. 2022, 60, 1–16. [Google Scholar] [CrossRef]
- Ren, B.; Zhao, Y.; Hou, B.; Chanussot, J.; Jiao, L. A Mutual Information-Based Self-Supervised Learning Model for PolSAR Land Cover Classification. IEEE Trans. Geosci. Remote Sens. 2021, 59, 9224–9237. [Google Scholar] [CrossRef]
- Lee, J.S.; Hoppel, K.W.; Mango, S.A.; Miller, A.R. Intensity and phase statistics of multilook polarimetric and interferometric SAR imagery. IEEE Trans. Geosci. Remote Sens. 1994, 32, 1017–1028. [Google Scholar] [CrossRef]
- Ainsworth, T.L.; Kelly, J.P.; Lee, J.S. Classification comparisons between dual-pol, compact polarimetric and quad-pol SAR imagery. ISPRS J. Photogramm. Remote Sens. 2009, 64, 464–471. [Google Scholar] [CrossRef]
- Turkar, V.; Deo, R.; Rao, Y.S.; Mohan, S.; Das, A. Classification accuracy of multi-frequency and multi-polarization SAR images for various land covers. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2012, 5, 936–941. [Google Scholar] [CrossRef]
- Georgiou, G.M.; Koutsougeras, C. Complex domain backpropagation. IEEE Trans. Circuits Syst. II Analog Digital Signal Process. 1992, 39, 330–334. [Google Scholar] [CrossRef]
- Trabelsi, C.; Bilaniuk, O.; Zhang, Y.; Serdyuk, D.; Subramanian, S.; Santos, J.; Mehri, S.; Rostamzadeh, N.; Bengio, Y.; Pal, C. Deep complex networks. In Proceedings of the International Conference on Learning Representations (ICLR2018), Vancouver, BC, Canada, 30 April–3 May 2018. [Google Scholar]
- Zhang, Z.; Wang, H.; Xu, F.; Jin, Y. Complex-Valued Convolutional Neural Network and Its Application in Polarimetric SAR Image Classification. IEEE Trans. Geosci. Remote Sens. 2017, 55, 7177–7188. [Google Scholar] [CrossRef]
- Li, L.; Ma, L.; Jiao, L.; Liu, F.; Sun, Q.; Zhao, J. Complex Contourlet-CNN for polarimetric SAR image classification. Pattern Recognit. 2020, 100, 107110. [Google Scholar] [CrossRef]
- Xiao, D.; Liu, C.; Wang, Q.; Wang, C.; Zhang, X. PolSAR Image Classification Based on Dilated Convolution and Pixel-Refining Parallel Mapping network in the Complex Domain. arXiv 2020, arXiv:1909.10783. [Google Scholar]
- Zhao, J.; Datcu, M.; Zhang, Z.; Xiong, H.; Yu, W. Contrastive-Regulated CNN in the Complex Domain: A Method to Learn Physical Scattering Signatures From Flexible PolSAR Images. IEEE Trans. Geosci. Remote Sens. 2019, 57, 10116–10135. [Google Scholar] [CrossRef]
- Tan, X.; Li, M.; Zhang, P.; Wu, Y.; Song, W. Complex-Valued 3-D Convolutional Neural Network for PolSAR Image Classification. IEEE Geosci. Remote Sens. Lett. 2020, 17, 1022–1026. [Google Scholar] [CrossRef]
- Zhang, P.; Tan, X.; Li, B.; Jiang, Y.; Song, W.; Li, M.; Wu, Y. PolSAR Image Classification Using Hybrid Conditional Random Fields Model Based on Complex-Valued 3-D CNN. IEEE Trans. Aerosp. Electron. Syst. 2021, 57, 1713–1730. [Google Scholar] [CrossRef]
- Qin, X.; Hu, T.; Zou, H.; Yu, W.; Wang, P. Polsar Image Classification via Complex-Valued Convolutional Neural Network Combining Measured Data and Artificial Features. In Proceedings of the 2019 IEEE International Geoscience and Remote Sensing Symposium (IGARSS2019), Yokohama, Japan, 28 July–2 August 2019; pp. 3209–3212. [Google Scholar]
- Fang, Z.; Zhang, G.; Dai, Q.; Xue, B. PolSAR Image Classification Based on Complex-Valued Convolutional Long Short-Term Memory Network. IEEE Geosci. Remote Sens. Lett. 2022, 19, 1–5. [Google Scholar] [CrossRef]
- Tan, X.; Li, M.; Zhang, P.; Wu, Y.; Song, W. Deep Triplet Complex-Valued Network for PolSAR Image Classification. IEEE Trans. Geosci. Remote Sens. 2021, 59, 10179–10196. [Google Scholar] [CrossRef]
- Abdo, A.J. The paradigm of complex probability and the Brownian motion. Syst. Sci. Control Eng. 2015, 3, 478–503. [Google Scholar] [CrossRef]
- Abdo, A.J. The paradigm of complex probability and Chebyshev’s inequality. Syst. Sci. Control Eng. 2016, 4, 99–137. [Google Scholar] [CrossRef]
- Abdo, A.J. The paradigm of complex probability and Claude Shannon’s information theory. Syst. Sci. Control Eng. 2017, 5, 380–425. [Google Scholar] [CrossRef]
- Abdo, A.J. The paradigm of complex probability and Ludwig Boltzmann’s entropy. Syst. Sci. Control Eng. 2018, 6, 108–149. [Google Scholar] [CrossRef]
- Abdo, A.J. The paradigm of complex probability and Monte Carlo methods. Syst. Sci. Control Eng. 2019, 7, 407–451. [Google Scholar] [CrossRef]
- Xie, W.; Ma, G.; Zhao, F.; Liu, H.; Zhang, L. PolSAR image classification via a novel semi-supervised recurrent complex-valued convolution neural network. Neurocomputing 2020, 388, 255–268. [Google Scholar] [CrossRef]
- Shang, R.; Wang, J.; Jiao, L.; Yang, X.; Li, Y. Spatial feature-based convolutional neural network for PolSAR image classification. Appl. Soft Comput. 2022, 123, 108922. [Google Scholar] [CrossRef]
- Hua, W.; Wang, X.; Zhang, C.; Jin, X. Attention-Based Multiscale Sequential Network for PolSAR Image Classification. IEEE Geosci. Remote Sens. Lett. 2022, 19, 1–5. [Google Scholar] [CrossRef]
Module | Dimension | Module | Dimension | ||
---|---|---|---|---|---|
RV-SCNN | RV-Convolution | 3 × 3 × 9 × 8 | CV-SCNN | CV-Convolution | 3 × 3 × 6 × 6 |
RV-Max-Pooling | 2 × 2 | CVA_Max_Pooling | 2 × 2 | ||
ReLU | HReLU | ||||
RV-Convolution | 3 × 3 × 8 × 22 | CV-Convolution | 3 × 3 × 6 × 12 | ||
RV-Max-Pooling | 2 × 2 | CVA_Max_Pooling | 2 × 2 | ||
ReLU | HReLU | ||||
RV-Average-Pooling | CV-Average-Pooling | ||||
RV-Fully-Connection | 22 × 180 | CV-Fully-Connection | 12 × 128 | ||
RV-Fully-Connection | 180 × K | CV-Fully-Connection | 128 × K | ||
RV-SCNN Params | FLevoland 1: 9147; | FLevoland 2: 8966; | Oberpfaffenhofen: 6975 | ||
CV-SCNN Params | FLevoland 1: 9214; | FLevoland 2: 8956; | Oberpfaffenhofen: 6118 |
Module | Dimension | Module | Dimension | ||
---|---|---|---|---|---|
RV-DCNN | RV-Convolution | 3 × 3 × 9 × 18 | CV-DCNN | CV-Convolution | 3 × 3 × 6 × 12 |
RV-Max-Pooling | 2 × 2 | CVA_Max_Pooling | 2 × 2 | ||
ReLU | HReLU | ||||
RV-Convolution | 3 × 3 × 18 × 36 | CV-Convolution | 3 × 3 × 12 × 24 | ||
RV-Max-Pooling | 2 × 2 | CVA_Max_Pooling | 2 × 2 | ||
ReLU | HReLU | ||||
RV-Convolution | 3 × 3 × 36 × 72 | CV-Convolution | 3 × 3 × 24 × 48 | ||
RV-Max-Pooling | 2 × 2 | CVA_Max_Pooling | 2 × 2 | ||
ReLU | HReLU | ||||
RV-Convolution | 3 × 3 × 72 × 114 | CV-Convolution | 3 × 3 × 48 × 96 | ||
RV-Max-Pooling | 2 × 2 | CVA_Max_Pooling | 2 × 2 | ||
ReLU | HReLU | ||||
RV-Average-Pooling | CV-Average-Pooling | ||||
RV-Fully-Connection | 144 × 312 | CV-Fully-Connection | 96 × 256 | ||
RV-Fully-Connection | 312 × K | CV-Fully-Connection | 256 × K | ||
RV-DCNN Params | FLevoland 1: 174,405; | FLevoland 2: 174,092; | Oberpfaffenhofen: 170,649 | ||
CV-DCNN Params | FLevoland 1: 168,254; | FLevoland 2: 167,740; | Oberpfaffenhofen: 162,086 |
Module | Dimension | Module | Dimension | ||
---|---|---|---|---|---|
RV-(FCN, SegNet) | RV-Convolution | 3 × 3 × 9 × 17 | CV-(FCN, SegNet) | CV-Convolution | 3 × 3 × 6 × 12 |
RV-Max-Pooling | 2 × 2 | CVA_Max_Pooling | 2 × 2 | ||
ReLU | HReLU | ||||
RV-Convolution | 3 × 3 × 17 × 34 | CV-Convolution | 3 × 3 × 12 × 24 | ||
RV-Max-Pooling | 2 × 2 | CVA_Max_Pooling | 2 × 2 | ||
ReLU | HReLU | ||||
RV-Convolution | 3 × 3 × 34 × 68 | CV-Convolution | 3 × 3 × 24 × 48 | ||
RV-Max-Pooling | 2 × 2 | CVA_Max_Pooling | 2 × 2 | ||
ReLU | HReLU | ||||
RV-Convolution | 3 × 3 × 68 × 132 | CV-Convolution | 3 × 3 × 48 × 96 | ||
RV-Max-Pooling | 2 × 2 | CVA_Max_Pooling | 2 × 2 | ||
ReLU | HReLU | ||||
Up-sampling | 2 × 2 | Up-sampling | 2 × 2 | ||
RV-Convolution | 3 × 3 × 132 × 68 | CV-Convolution | 3 × 3 × 96 × 48 | ||
ReLU | HReLU | ||||
Up-sampling | 2 × 2 | Up-sampling | 2 × 2 | ||
RV-Convolution | 3 × 3 × 68 × 34 | CV-Convolution | 3 × 3 × 48 × 24 | ||
ReLU | HReLU | ||||
Up-sampling | 2 × 2 | Up-sampling | 2 × 2 | ||
RV-Convolution | 3 × 3 × 34 × 17 | CV-Convolution | 3 × 3 × 24 × 12 | ||
ReLU | HReLU | ||||
Up-sampling | 2 × 2 | Up-sampling | 2 × 2 | ||
RV-Convolution | 3 × 3 × 17 × 9 | CV-Convolution | 3 × 3 × 12 × 6 | ||
ReLU | HReLU | ||||
Up-sampling | 2 × 2 | Up-sampling | 2 × 2 | ||
RV-Convolution | 3 × 3 × 9 × K | CV-Convolution | 3 × 3 × 6 × K | ||
RV-(FCN, SegNet) Params | FLevoland 1: 218,345; | FLevoland 2: 218,262; | Oberpfaffenhofen: 217,349 | ||
CV-(FCN, SegNet) Params | FLevoland 1: 223,080; | FLevoland 2: 222,968; | Oberpfaffenhofen: 221,736 |
RV-SCNN | Old CV-SCNN | New CV-SCNN | RV-DCNN | Old CV-DCNN | New CV-DCNN | RV-FCN | Old CV-FCN | New CV-FCN | RV-SegNet | Old CV-SegNet | New CV-SegNet | |
---|---|---|---|---|---|---|---|---|---|---|---|---|
Stembeans | 99.85 | 99.95 | 99.33 | 99.48 | 99.67 | 99.62 | 99.72 | 99.84 | 99.92 | 99.95 | 99.98 | 100.00 |
Peas | 95.71 | 99.57 | 99.58 | 95.06 | 99.14 | 99.96 | 97.56 | 99.70 | 98.72 | 98.94 | 98.76 | 99.31 |
Forest | 98.73 | 99.65 | 97.12 | 98.11 | 99.46 | 99.52 | 98.65 | 98.70 | 100.00 | 99.26 | 99.48 | 99.92 |
Lucerne | 98.04 | 99.91 | 96.24 | 96.39 | 96.89 | 96.47 | 88.26 | 99.88 | 99.93 | 98.23 | 99.95 | 99.88 |
Wheat | 97.44 | 97.91 | 94.21 | 93.54 | 98.89 | 95.35 | 99.81 | 98.35 | 99.80 | 99.95 | 100.00 | 100.00 |
Beet | 98.38 | 98.60 | 98.52 | 97.84 | 93.64 | 99.79 | 96.16 | 94.78 | 98.92 | 99.70 | 99.20 | 99.43 |
Potaotes | 97.74 | 96.76 | 97.57 | 99.44 | 95.02 | 99.35 | 94.47 | 99.80 | 98.54 | 99.25 | 99.27 | 99.88 |
Bare soil | 99.97 | 94.41 | 93.01 | 100.00 | 74.27 | 98.31 | 87.69 | 92.76 | 95.58 | 100.00 | 99.94 | 100.00 |
Grass | 94.51 | 92.10 | 92.79 | 96.47 | 95.58 | 98.68 | 98.69 | 77.51 | 99.89 | 99.86 | 99.79 | 100.00 |
Rapeseed | 72.03 | 69.68 | 98.72 | 71.44 | 94.12 | 90.59 | 97.48 | 96.42 | 99.35 | 99.53 | 99.92 | 99.91 |
Barley | 66.85 | 45.26 | 96.79 | 78.30 | 99.46 | 96.95 | 77.71 | 99.58 | 96.03 | 96.80 | 99.83 | 99.64 |
Wheat2 | 95.52 | 99.61 | 88.63 | 97.57 | 97.55 | 99.75 | 98.97 | 95.80 | 98.80 | 100.00 | 100.00 | 99.92 |
Wheat3 | 99.92 | 99.45 | 97.94 | 99.90 | 98.22 | 99.97 | 99.66 | 99.65 | 99.97 | 99.97 | 99.35 | 99.92 |
Water | 77.20 | 99.77 | 99.07 | 87.54 | 96.99 | 99.98 | 86.69 | 93.92 | 95.71 | 98.66 | 98.81 | 99.46 |
Buildings | 98.74 | 96.22 | 93.49 | 83.82 | 83.82 | 98.53 | 85.08 | 96.64 | 82.98 | 85.50 | 84.03 | 82.77 |
OA | 92.65 | 93.81 | 96.66 | 93.67 | 96.79 | 98.13 | 95.40 | 97.10 | 98.86 | 99.31 | 99.49 | 99.76 |
AA | 92.71 | 92.59 | 96.20 | 92.99 | 94.85 | 98.19 | 93.77 | 96.22 | 97.61 | 98.37 | 98.55 | 98.67 |
Kappa | 0.9186 | 0.9315 | 0.9634 | 0.9300 | 0.9648 | 0.9795 | 0.9493 | 0.9682 | 0.9875 | 0.9925 | 0.9944 | 0.9974 |
RV-SCNN | Old CV-SCNN | New CV-SCNN | RV-DCNN | Old CV-DCNN | New CV-DCNN | RV-FCN | Old CV-FCN | New CV-FCN | RV-SegNet | Old CV-SegNet | New CV-SegNet | |
---|---|---|---|---|---|---|---|---|---|---|---|---|
Potato | 99.48 | 99.50 | 99.98 | 99.90 | 99.80 | 99.86 | 98.72 | 97.72 | 99.97 | 99.63 | 99.46 | 99.94 |
Fruit | 100.00 | 99.70 | 99.77 | 99.66 | 99.66 | 99.93 | 98.23 | 99.98 | 99.70 | 96.97 | 90.03 | 98.51 |
Oats | 93.62 | 94.98 | 95.62 | 96.41 | 92.32 | 96.41 | 99.93 | 100.00 | 100.00 | 100.00 | 99.93 | 99.78 |
Beet | 94.20 | 99.06 | 98.87 | 92.75 | 98.54 | 98.87 | 94.82 | 95.21 | 97.71 | 94.14 | 95.41 | 99.92 |
Barley | 93.59 | 99.60 | 99.74 | 96.26 | 99.09 | 99.99 | 98.60 | 98.92 | 99.98 | 98.32 | 99.98 | 99.98 |
Onions | 52.77 | 13.29 | 60.00 | 77.75 | 17.89 | 76.24 | 100.00 | 98.08 | 98.73 | 97.18 | 96.71 | 99.39 |
Wheat | 89.50 | 99.80 | 99.71 | 98.54 | 99.76 | 99.95 | 99.91 | 99.45 | 100.00 | 99.83 | 99.78 | 100.00 |
Beans | 0.09 | 94.27 | 82.16 | 11.18 | 82.53 | 98.43 | 84.84 | 92.42 | 95.84 | 87.99 | 97.97 | 99.91 |
peas | 99.72 | 97.69 | 97.22 | 99.91 | 99.95 | 99.44 | 99.95 | 100.00 | 100.00 | 100.00 | 100.00 | 100.00 |
Maize | 89.61 | 89.15 | 91.86 | 96.28 | 81.16 | 74.11 | 94.42 | 98.99 | 100.00 | 92.56 | 97.75 | 94.42 |
Flax | 98.72 | 97.74 | 99.28 | 97.23 | 99.95 | 99.98 | 99.95 | 100.00 | 100.00 | 98.63 | 99.98 | 100.00 |
Rapessed | 97.62 | 99.42 | 99.55 | 99.29 | 99.27 | 99.95 | 99.27 | 99.58 | 99.97 | 99.99 | 99.87 | 99.99 |
Grass | 85.94 | 82.30 | 95.15 | 97.84 | 95.20 | 99.62 | 97.88 | 98.72 | 99.74 | 100.00 | 100.00 | 99.95 |
Lucerne | 87.94 | 92.48 | 98.88 | 98.17 | 88.79 | 99.80 | 99.93 | 100.00 | 99.97 | 100.00 | 100.00 | 100.00 |
OA | 93.31 | 97.22 | 98.57 | 96.95 | 97.39 | 99.17 | 98.66 | 98.72 | 99.73 | 98.78 | 99.06 | 99.86 |
AA | 84.49 | 89.93 | 94.13 | 90.08 | 89.56 | 95.90 | 97.60 | 98.50 | 99.40 | 97.52 | 98.35 | 99.41 |
Kappa | 0.9190 | 0.9668 | 0.9830 | 0.9638 | 0.9689 | 0.9902 | 0.9841 | 0.9849 | 0.9968 | 0.9856 | 0.9889 | 0.9984 |
RV-SCNN | Old CV-SCNN | New CV-SCNN | RV-DCNN | Old CV-DCNN | New CV-DCNN | RV-FCN | Old CV-FCN | New CV-FCN | RV-SegNet | Old CV-SegNet | New CV-SegNet | |
---|---|---|---|---|---|---|---|---|---|---|---|---|
Built-up areas | 79.90 | 87.36 | 92.50 | 79.38 | 82.83 | 89.83 | 98.55 | 97.22 | 99.46 | 96.45 | 96.24 | 98.96 |
Wood land | 97.36 | 98.27 | 98.65 | 98.74 | 99.11 | 99.44 | 99.69 | 99.30 | 99.31 | 98.74 | 99.59 | 99.20 |
Open areas | 96.21 | 96.07 | 96.12 | 98.71 | 99.30 | 99.78 | 97.46 | 99.05 | 99.11 | 98.68 | 99.04 | 99.51 |
OA | 92.35 | 94.31 | 95.69 | 93.88 | 95.14 | 97.22 | 98.15 | 98.64 | 99.23 | 98.13 | 98.45 | 99.31 |
AA | 91.16 | 93.90 | 95.76 | 92.28 | 93.75 | 96.35 | 98.57 | 98.52 | 99.29 | 97.96 | 98.29 | 99.22 |
Kappa | 0.8512 | 0.8941 | 0.9221 | 0.8831 | 0.9094 | 0.9501 | 0.9679 | 0.9763 | 0.9868 | 0.9673 | 0.9729 | 0.9882 |
Dataset | Methods | OA | AA | Kappa |
---|---|---|---|---|
Flevoland Dataset 1 | RMP-CV-SCNN | 95.65 | 95.17 | 0.9522 |
RAP-CV-SCNN | 94.10 | 93.64 | 0.9349 | |
new CV-SCNN | 96.66 | 96.20 | 0.9634 | |
Flevoland Dataset 2 | RMP-CV-SCNN | 97.94 | 93.56 | 0.9756 |
RAP-CV-SCNN | 96.71 | 92.30 | 0.9609 | |
new CV-SCNN | 98.57 | 94.13 | 0.9830 | |
Oberpfaffenhofen Dataset | RMP-CV-SCNN | 94.78 | 94.90 | 0.9043 |
RAP-CV-SCNN | 94.63 | 94.64 | 0.9014 | |
new CV-SCNN | 95.69 | 95.76 | 0.9221 |
Dataset | Methods | OA | AA | Kappa |
---|---|---|---|---|
Flevoland Dataset 1 | CReLU-CV-SCNN | 95.95 | 95.49 | 0.9554 |
ZReLU-CV-SCNN | 95.45 | 95.00 | 0.9499 | |
Mod-CV-SCNN | 95.53 | 95.03 | 0.9508 | |
new CV-SCNN | 96.66 | 96.20 | 0.9634 | |
Flevoland Dataset 2 | CReLU-CV-SCNN | 98.05 | 93.67 | 0.9769 |
ZReLU-CV-SCNN | 97.84 | 93.42 | 0.9744 | |
Mod-CV-SCNN | 97.68 | 93.25 | 0.9725 | |
new CV-SCNN | 98.57 | 94.13 | 0.9830 | |
Oberpfaffenhofen Dataset | CReLU-CV-SCNN | 94.97 | 95.00 | 0.9082 |
ZReLU-CV-SCNN | 94.79 | 94.82 | 0.9046 | |
Mod-CV-SCNN | 93.77 | 93.67 | 0.8844 | |
new CV-SCNN | 95.69 | 95.76 | 0.9221 |
Dataset | Methods | OA | AA | Kappa |
---|---|---|---|---|
Flevoland Dataset 1 | RCE-CV-SCNN | 96.00 | 95.55 | 0.9560 |
new CV-SCNN | 96.66 | 96.20 | 0.9634 | |
Flevoland Dataset 2 | RCE-CV-SCNN | 98.20 | 93.72 | 0.9787 |
new CV-SCNN | 98.57 | 94.13 | 0.9830 | |
Oberpfaffenhofen Dataset | RCE-CV-SCNN | 95.02 | 95.16 | 0.9089 |
new CV-SCNN | 95.69 | 95.76 | 0.9221 |
RCV-CNN [49] | CV-Contourlet-CNN [36] | SF-CNN [50] | AMSE-LSTM [51] | CV-ConvLSTM [42] | New CV-SegNet | |
---|---|---|---|---|---|---|
Stembeans | 98.61 | 99.81 | - | 97.16 | 94.24 | 100.00 |
Peas | 98.56 | 99.86 | 99.62 | 97.62 | 99.97 | 99.31 |
Forest | 97.81 | 98.98 | - | 98.43 | 99.17 | 99.92 |
Lucerne | 98.22 | 99.55 | 99.93 | 97.54 | 98.56 | 99.88 |
Wheat | 94.50 | 99.59 | 99.46 | 98.82 | 97.56 | 100.00 |
Beet | 94.14 | 99.25 | 99.22 | 94.71 | 99.07 | 99.43 |
Potaotes | 98.90 | 99.18 | 99.50 | 96.40 | 98.49 | 99.88 |
Bare soil | 98.05 | 100.00 | 99.72 | 99.43 | 99.67 | 100.00 |
Grass | 89.17 | 99.85 | - | 98.06 | 96.73 | 100.00 |
Rapeseed | 97.07 | 99.00 | 99.88 | 96.03 | 97.68 | 99.91 |
Barley | 98.20 | 99.77 | 99.50 | 99.72 | 100.00 | 99.64 |
Wheat2 | 97.28 | 99.43 | - | 98.50 | 99.88 | 99.92 |
Wheat3 | 98.56 | 99.39 | - | 99.22 | 98.32 | 99.92 |
Water | 99.89 | 99.58 | - | 99.81 | 99.68 | 99.46 |
Buildings | 80.88 | 99.26 | - | 84.90 | 79.41 | 82.77 |
OA | 97.22 | 99.42 | 99.58 | 97.09 | 98.58 | 99.76 |
AA | - | 99.50 | 99.61 | - | 97.32 | 98.67 |
Kappa | 0.8930 | 0.9902 | 0.9950 | 0.9683 | 0.9845 | 0.9974 |
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2023 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
Ren, Y.; Jiang, W.; Liu, Y. A New Architecture of a Complex-Valued Convolutional Neural Network for PolSAR Image Classification. Remote Sens. 2023, 15, 4801. https://doi.org/10.3390/rs15194801
Ren Y, Jiang W, Liu Y. A New Architecture of a Complex-Valued Convolutional Neural Network for PolSAR Image Classification. Remote Sensing. 2023; 15(19):4801. https://doi.org/10.3390/rs15194801
Chicago/Turabian StyleRen, Yihui, Wen Jiang, and Ying Liu. 2023. "A New Architecture of a Complex-Valued Convolutional Neural Network for PolSAR Image Classification" Remote Sensing 15, no. 19: 4801. https://doi.org/10.3390/rs15194801
APA StyleRen, Y., Jiang, W., & Liu, Y. (2023). A New Architecture of a Complex-Valued Convolutional Neural Network for PolSAR Image Classification. Remote Sensing, 15(19), 4801. https://doi.org/10.3390/rs15194801