Acta Geodaetica et Cartographica Sinica ›› 2023, Vol. 52 ›› Issue (5): 798-807.doi: 10.11947/j.AGCS.2023.20220163

• Photogrammetry andRemote Sensing • Previous Articles     Next Articles

Dual-channel parallel hybrid convolutional neural networks based classification method for high-resolution remote sensing image

GU Xiaohu1, LI Zhengjun2, MIAO Jianhao1, LI Xinghua1,3, SHEN Huanfeng4   

  1. 1. School of Remote Sensing and Information Engineering, Wuhan University, Wuhan 430079, China;
    2. CCCC Second Highway Consultants Co., Ltd., Wuhan 430056, China;
    3. Key Laboratory of Urban Land Resources Monitoring and Simulation, Ministry of Natural Resources, Shenzhen 518000, China;
    4. School of Resource and Environmental Sciences, Wuhan University, Wuhan 430079, China
  • Received:2022-03-02 Revised:2022-10-12 Published:2023-05-27
  • Supported by:
    The National Natural Science Foundation of China (No. 42171302);The National Key Research and Development Program of China (No. 2019YFB2102904);The Open Fund of Key Laboratory of Urban Land Resources Monitoring and Simulation, Ministry of Natural Resources (No. KF-2021-06-003)

Abstract: High spatial resolution remote sensing images have rich spatial detail information and multi-spectral information. Previous studies have shown that two-dimensional convolutional neural networks (CNN) are suitable for extracting spatial information, while three-dimensional CNN are more suitable for extracting spectral information. In order to make better use of spatio-spectral information, this paper innovatively proposes a dual-channel parallel hybrid convolutional neural networks (DPHCNN), which fully combines the advantages of two-dimensional and three-dimensional CNN in spatio-spectral information extraction. Simultaneously, the hybrid attention mechanism and multi-scale convolution are introduced to enhance the extraction ability of spatial detail features to achieve accurate classification of high-resolution images. In the experiment, the GF-2 image dataset was used for verification. Compared with state-of-the-art deep learning classification methods, the DPHCNN method proposed in this paper not only has the highest classification accuracy and better classification efficiency but maintains the highest robustness in multi-temporal images classification, which has more advantages in comprehensive evaluation.

Key words: hybrid convolutional neural network, high-resolution remote sensing images, multi-scale convolution, hybrid attention mechanism, image classification

CLC Number: