Deep learning technology provides technical means for hyperspectral image classification due to its unique advantages in deep mining of features. However, in the pixel-level feature classification of hyperspectral images, the number of deep learning layers is limited due to the influence of the sample input size, and the depth features in the hyperspectral images cannot be fully mined. The classification of hyperspectral image based on feature fusion of residual network is proposed in this paper. First, the principal component analysis (PCA) method is used to extract the first principal component in the original hyperspectral image, and the residual network is used to effectively extract the spatial spectrum features of the ground objects; then the feature map is expanded by the deconvolution algorithm, and after deconvolution, features of different dimensions are fused with multi-scale features to fully mine the depth feature information in the hyperspectral image, thus further improving the classification accuracy of the hyperspectral image. The ground feature classification experiment was conducted on the two areas of Taihu Lake in Jiangsu and Chaohu Lake in Anhui captured by the “Zhuhai-1” satellite. The results show that, compared with other methods, this method can effectively solve the problem of insufficient depth feature extraction in hyperspectral image classification, thus showing better classification performance.
韩彦岭, 崔鹏霞, 杨树瑚, 刘业锟, 王静, 张云. 基于残差网络特征融合的高光谱图像分类[J]. 国土资源遥感, 2021, 33(2): 11-19.
HAN Yanling, CUI Pengxia, YANG Shuhu, LIU Yekun, WANG Jing, ZHANG Yun. Classification of hyperspectral image based on feature fusion of residual network. Remote Sensing for Land & Resources, 2021, 33(2): 11-19.
Li B X, Zhou B, He X, et al. Status and prospects of target classification methods for hyperspectral images[J]. Laser and Infrared, 2020, 50(3):259-265.
Lu G J, Chen L F. Remote sensing image scene classification based on deep convolutional neural network[J]. Journal of Taiyuan Teachers College (Natural Science Edition), 2019, 18(1):57-62.
Wang Y Y, Luo L K, Zhou Z G. Remote sensing object image classification based on deconvolution high-level features[J]. Computer Engineering and Applications, 2020, 56(11):200-206.
Yu P X, Zhou X, Liu S H, et al. Feature extraction and recognition of remote sensing image of eroded gully in black soil region of northeast China[J]. Journal of Remote Sensing, 2018, 22(4):611-620.
[6]
He K M, Zhang X Y, Ren S Q, et al. Deep residual learning for image recognition[C]// 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016, 6:770-778.
Zhang Y Z, Xu M M, Wang X H, et al. Classification of hyperspectral features by hierarchical fusion of residual networks[J]. Spectroscopy and Spectral Analysis, 2019, 39(11):3501-3507.
Zhou J Y, Zhao Y M. Overview of the application of convolutional neural networks in image classification and target detection[J]. Computer Engineering and Applications, 2017, 53(13):34-41.
Pei J Y, Liu Y H. Remote sensing target detection algorithm based on multi-scale receptive field amplification fusion[J]. Command Control and Simulation, 2020, 42(1):34-39.
[11]
Nair V, Hinton G E. Rectified linear units improve restricted boltzmann machines[Z]. In ICML. 2010.
[12]
Chen P H, Lin C J, Schölkopf, B. A tutorial on v-support vector machines[J]. Applied Stochastic Models in Business and Industry, 2005, 21(2):111-136.
doi: 10.1002/asmb.v21:2
[13]
Krizhevsky A, Sutskever I, Hinton G. Imagenet classification with deep convolutional neural networks[J]. Advances in Neural Information Processing Systems, 2012, 25(2):1-9.
[14]
Su H, Wang Y, Xiao J, et al. Improving MODIS sea ice detectability using gray level co-occurrence matrix texture analysis method:A case study in the Bohai Sea[J]. ISPRS Journal of Photogrammetry & Remote Sensing, 2013, 85(85):13-20.