Fusion attention hole residual networks for hyperspectral image classification
LUO Jiming, ZHU Tongjun, HUANG Mingming, HUANG Quanzhen, ZHANG Yang, ZHAO Junhao, YANG Lianpu
School of Electrical Information Engineering, Henan Institute of Engineering, Zhengzhou 451191, China
Abstract: Aiming at the high-dimensional characteristics of hyperspectral image data, in order to further improve the accuracy of image classification, a three-dimensional convolutional neural network model is designed for the hyperspectral classification problem. This method is based on 3D convolution and uses a multi-size convolution kernel strategy to extract the characteristic information of hyperspectral images from different scales. The use of an dilated convolution kernel effectively extracts feature information and expands the receptive field of the network. We propose a spatial-spectral attention model block that adaptively focuses on pertinent information, thereby augmenting the feature representation capability of the hyperspectral image space and spectrum. The proposed method was tested on public data sets such as University of Pavia and Indian Pines, and achieved overall classification accuracy of 99.61% and 99.58%, respectively. Compared with the results of SVM, 2D-CNN, 3D-CNN, RES-3D-CNN and other algorithms, the algorithm proposed in this paper is superior to other algorithms in accuracy and classification performance.
Keywords: image classification;hyperspectral image;neural networks;spatial spectral attention;multiscale
2023, 49(9):111-119  收稿日期: 2022-12-7;收到修改稿日期: 2023-2-3
基金项目: 河南省科技攻关项目（212102210022）；河南省自然科学基金青年科学基金项目（212300410127）；河南省科技攻关项目（212102210014）
作者简介: 骆继明（1975-），男，河南光山县人，副教授，硕士，主要从事图像处理、发电机控制设计方面的研究。
参考文献
[1] 杨蕴睿, 郑东文. 一种用于高光谱图像分类的空谱协同编码方法[J]. 中国测试, 2022, 48(12): 162-171.
[2] WASKE B, VAN D, BENEDIKTSSON J A, et al. Sensitivity of support vector machines to random feature selection in classification of hyperspectral data[J]. IEEE Transactions on Geoscience and Remote Sensing, 2010, 48(7): 2880-2889.
[3] MARIO HAUT J, PAOLETTI M, PLAZA J, et al. Cloud implementation of the K-means algorithm for hyperspectral image analysis[J]. Journal of Supercomputing, 2017, 73(1): 514-529.
[4] 吴建, 王红军. 基于卷积神经网络的农作物植株干旱检测[J]. 中国测试, 2022, 48(4): 102-109.
[5] WANG W, DOU S, JIANG Z, et al. A fast dense spectral–spatial convolution network framework for hyperspectral images classification[J]. Remote Sensing, 2018, 10(7): 1068.
[6] LI R, PAN Z, WANG Y, et al. A convolutional neural network with mapping layers for hyperspectral image classification[J]. IEEE Transactions on Geoscience and Remote Sensing, 2020, 58(5): 3136-3147.
[7] ZHU M, JIAO L, LIU F, et al. Residual spectral-spatial attention network for hyperspectral image classification[J]. IEEE Transactions on Geoscience and Remote Sensing, 2021, 59: 449-462.
[8] 杨蕴睿, 郑东文. 一种空谱协同编码方法用于高光谱图像分类[J/OL]. 中国测试, 2023: 1-14. http://kns.cnki.net/kcms/detail/51.1714.TB.20220325.0919.008.html.
[9] LI Y, ZHANG H, SHEN Q. Spectral-spatial classification of hyperspectral imagery with 3D convolutional neural network[J]. Remote Sensing, 2017, 9(1): 67.
[10] RAO M, TANG P, ZHANG Z. A developed siamese CNN with 3D adaptive spatial-spectral pyramid pooling for hyperspectral image classification[J]. Remote Sensing, 2020, 12(12): 1964.
[11]  SEYDGAR M, NAEINI A A, ZHANG M M, et al. 3-D convolution-recurrent networks for spectral-spatial classification of hyperspectral images[J]. Remote Sensing, 2019, 11(7): 883-883.
[12] PANQU W, PENGFEI C, YE Y, et al. Understanding convolution for semantic segmentation[C]//IEEE Winter Conference on Applications of Computer Vision (WACV), 2018.
[13] HE K, ZHANG X, REN S, et al. Spatial pyramid pooling in deep convolutional networks for visual recognition[J]. Pattern Analysis & Machine Intelligence IEEE Transactions on, 2015, 37(9): 1904-1916.
[14] HAO S Y, WANG W, YE Y X, et al. Two-stream deep architecture for hyperspectral image classification[J]. IEEE Transactions on Geoscience & Remote Sensing, 2018, 56(4): 2349-2361.
[15] LIU B, YU X, ZHANG P, et al. Deep 3D convolutional network combined with spatial-spectral features for hyperspectral image classification[J]. Acta Geod. Cartogr. Sin., 2019, 48: 53-63.