Marine Aquaculture Information Extraction from Optical Remote Sensing Images via MDOAU2-net

Abstract

China has the largest aquaculture area in the world and is still expanding. Extracting the area of marine aquaculture can prevent the overexploitation of marine aquaculture and protect the marine environment. MDOAU-net has an excellent performance in marine aquaculture extraction of SAR images which drives researchers to explore the performance of MDOAU-net in optical remote sensing images. Unlike SAR images, optical remote sensing images needn't consider speckles noises problem. To suit optical remote sensing images, a new method named MDOAU2-net is proposed to accurately extract marine aquaculture areas, which could keep the discriminative character and filter fake objects with similar features. It follows the structure of the U-net and is contained by a multi-scale block and some offset convolution blocks. In experiments, using the images shot by GF-2 satellite as data and compared to other five networks to verify the validity of MDOAU2-net in optical remote sensing images of marine aquaculture extraction.