A Robust Feature Extraction Method for Heterogenous Remote Sensing Images Based on Feature Transfer.

A robust feature extraction method based on feature transfer for heterogeneous remote sensing images is proposed to address the problem of insufficient generalization ability of traditional edge feature extraction methods applied to heterogeneous remote sensing images. Firstly, the heterogeneous remote sensing images are transferred based on CycleGAN to reduce the impact of heterogeneous remote sensing images on the edge feature extraction results due to the cross-data domain. Secondly, dilated convolution is introduced in the edge feature extraction model DexiNed, which makes the optimized network model pay more attention to the structured features of the image. The experimental results show that the proposed method improves the structured feature extraction accuracy of heterogenous remote sensing images, which has better robust feature extraction effect compared with DexiNed.