A high-dimensional multi-source remote sensing image classification method based on SCA and 3D-CNN fusion network

Abstract

The application discloses a kind of high-dimensional multi-source remote sensing image classification methods based on SCA and 3D-CNN fusion network, steps have: (1) obtain GF-1 and Sentinel-2 remote sensing image data, carry out radiation calibration, atmospheric correction and other image preprocessing work;(2) set training parameter, carry out layer-by-layer greedy pre-training to stack convolutional auto-encoding network, complete the initialization of encoding network;(3) the convolutional auto-encoding network is connected with 3D-CNN network, and classification training is carried out, the overall network is fine-tuned, while using 3D-CNN to add the neighborhood information of central pixel, the overall effect of network is optimized, and the classification precision is improved.The application combines the advantages of auto-encoding network and CNN network, integrates the data dimension reduction process and the classification process, changes the traditional "dimension reduction first, then classification" data processing mode, simplifies the work flow of remote sensing image classification, and achieves high precision in the process of remote sensing image classification, which provides a new idea for remote sensing image classification.

Images