Abstract
The utilization of Convolutional Neural Networks (CNNs) in hyperspectral image (HSI) classification has become commonplace. However, traditional CNNs cannot fully extract the features of HSI and are prone to gradient vanishing when the network layer is deepened. We suggest a 2D–3D hybrid convolution and pre-activated residual networks-based HSI classification (HSIC) approach to tackle these problems. Firstly, the joint spatial–spectral features of HSI are extracted by a two-layer 3D convolution. Secondly, combining the advantages of 2D and 3D convolution to construct a spatial–spectral feature extraction module based on pre-activated residual networks, which can accelerate the convergence speed of the model while enhancing the capability of advanced spatial semantic feature extraction of HSI. Then, multiple residual modules are connected to take advantage of the different forms of features extracted by each convolutional layer, while multi-feature fusion is performed between blocks to achieve feature complementarity. Finally, a long-distance residual connection is introduced to fuse the shallow and deep features effectively, which further strengthens the expression ability of features. The results of the experiments conducted on three HSIs show that the overall classification accuracy of the model reaches 99.56%, 99.45% and 99.43%, respectively, when 10%, 1% and 1% of samples are randomly selected for training in each ground object class. Compared with other related CNN-based HSI classification models, our model can obtain higher classification accuracy. Consequently, the suggested method is capable of achieving feature reuse and obtaining deep high-level spatial–spectral features with superior discriminative and robustness, and its classification performance is superior to that of existing state-of-the-art methods.
Similar content being viewed by others
Data availability
The data that support the findings of this study are openly available in http://www.ehu.eus/ccwintco/index.php?title=Hyperspectral_Remote_Sensing_Scenes
References
Zhao, C., Wang, M., Feng, S.: A sparse and spectral smooth regularized low-rank tensor decomposition method for hyperspectral target detection. Int. J. Remote Sens. 43(12), 4608–4629 (2022)
Gao, H., Wang, M., Sun, X., Cao, X., et al.: Unsupervised dimensionality reduction of medical hyperspectral imagery in tensor space. Comput. Methods Progr. Biomed. 240, 107724 (2023)
Liu, G., Wang, L., Liu, D.: Hyperspectral image classification based on a least square bias constraint additional empirical risk minimization nonparallel support vector machine. Remote Sens. 14(17), 4263 (2022)
Wang, H., Celik, T.: Sparse representation-based hyperspectral image classification. Sign. Image Video Process. 12(5), 1009–1017 (2018)
Tan, X., Xue, Z., Yu, X., Sun, Y., et al.: Hyperspectral image classification with deep 3D capsule network and Markov random field. IET Image Process. 16(1), 79–91 (2022)
Yang, L., Chen, J., Zhang, R., Yang, S., et al.: Precise crop classification of UAV hyperspectral imagery using kernel tensor slice sparse coding based classifier. Neurocomputing 551, 126487 (2023)
Hu, W., Huang, Y., Wei, L., Zhang, F., et al.: Deep convolutional neural networks for hyperspectral image classification. J. Sensors 2015, 258619 (2015)
Zhao, W., Du, S.: Learning multiscale and deep representations for classifying remotely sensed imagery. ISPRS J. Photogramm. Remote Sens. 113, 155–165 (2016)
Li, Y., Zhang, H., Shen, Q.: Spectral–spatial classification of hyperspectral imagery with 3D convolutional neural network. Remote Sens. 9(1), 67 (2017)
Zheng, J., Feng, Y., Bai, C., Zhang, J.: Hyperspectral image classification using mixed convolutions and covariance pooling. IEEE Trans. Geosci. Remote Sens. 59(1), 522–534 (2021)
Fırat, H., Asker, M.E., Hanbay, D.: Classification of hyperspectral remote sensing images using different dimension reduction methods with 3D/2D CNN. Remote Sens. Appl.: Soc. Environ. 25, 100694 (2022)
Liu, Z., Mao, X., Huang, J., Gan, M., et al.: Stratified attention dense network for image super-resolution. Sign. Image Video Process. 16(3), 715–722 (2022)
Shi, C., Liao, D., Zhang, T., Wang, L.: Hyperspectral image classification based on 3D coordination attention mechanism network. Remote Sens. 14(3), 608 (2022)
He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for image recognition. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 770–778; 2016.
Qing, Y., Liu, W.: Hyperspectral image classification based on multi-scale residual network with attention mechanism. Remote Sens. 13(3), 335 (2021)
He, Z., Shi, Q., Liu, K., Cao, J., et al.: Object-oriented mangrove species classification using hyperspectral data and 3-D siamese residual network. IEEE Geosci. Remote Sens. Lett. 17(12), 2150–2154 (2020)
Cao, F., Guo, W.: Deep hybrid dilated residual networks for hyperspectral image classification. Neurocomputing 384, 170–181 (2020)
Dang, L., Pang, P., Lee, J.: Depth-Wise separable convolution neural network with residual connection for hyperspectral image classification. Remote Sens. 12(20), 3408 (2020)
He, S., Jing, H., Xue, H.: Spectral-spatial multiscale residual network for hyperspectral image classification. Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. 43, 389–395 (2022)
Lei, R., Zhang, C., Zhang, X., Huang, J., et al.: Multiscale feature aggregation capsule neural network for hyperspectral remote sensing image classification. Remote Sens. 14(7), 1652 (2022)
He, K., Zhang, X., Ren, S., Sun, J.: Identity mappings in deep residual networks. In Computer Vision–ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, October 11–14, 2016, Proceedings, Part IV 14, pp. 630–645: Springer, 2016
Gao, H., Yang, Y., Yao, D., Li, C.: Hyperspectral image classification with pre-activation residual attention network. IEEE Access 7, 176587–176599 (2019)
Huan, H., Li, P., Zou, N., Wang, C., et al.: End-to-End super-resolution for remote-sensing images using an improved multi-scale residual network. Remote Sens. 13(4), 666 (2021)
Wang, X., Xu, H., Yuan, L., Dai, W., et al.: A remote-sensing scene-image classification method based on deep multiple-instance learning with a residual dense attention convnet. Remote Sens. 14(20), 5095 (2022)
Funding
This work was supported by Zhejiang Provincial Education Department General Research Project (No. Y202248546), Public Welfare Applied Research Project of Huzhou (No. 2023GZ29), Natural Science Foundation of Huzhou (No. 2023YZ55) and Zhejiang Provincial College Student Innovation and Entrepreneurship Training Program Project (No. S202310347089).
Author information
Authors and Affiliations
Contributions
HL conceptualized and designed the algorithm, contributed to algorithm improvements, and critically revised the manuscript for important intellectual content. YS built the model, verified and analyzed it experimentally, prepared the original manuscript draft. HZ assisted with manuscript writing and revisions, supervised the project, provided strategic direction in algorithm development and testing, and conducted a thorough review and final approval of the manuscript prior to submission. ML visualized experimental results. All authors reviewed the manuscript.
Corresponding author
Ethics declarations
Conflict of interest
The authors declare no conflict of interest.
Ethical approval
Not applicable.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Lv, H., Sun, Y., Zhang, H. et al. Hybrid 2D–3D convolution and pre-activated residual networks for hyperspectral image classification. SIViP 18, 3815–3827 (2024). https://doi.org/10.1007/s11760-024-03044-0
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11760-024-03044-0