GF-3 PolSAR Marine Aquaculture Recognition Based on Complex Convolutional Neural Networks

摘要

Marine floating raft aquaculture is widely distributed along the coast in China. Polarimetric synthetic aperture radar (PoISAR) images can distinguish marine aquaculture targets from sea water background, but optical satellite remote sensing images cannot detect these effectively and completely. In this paper, considering the complex character of PoISAR data, a complex-value convolutional neural network is utilized for marine aquaculture recognition, which makes the most of phase information implicit in original complex data to improve detection accuracy. Experiments on actual GF-3 PoISAR images substantiate the effectiveness of the proposed approach.