Segmentation of Ship Satellite Images with Transfer Learning and Attention Mechanism
Lin Li, Meng Joo Er, Qianying Li, Yani Zhang · 2022
As an integrated part of national marine supervision system, ship monitoring plays a critical role in ensuing efficient exploitation of marine resources and safety of territorial sea. In this paper semantic segmentation networks are trained using high-resolution satellite ship data to facilitate more detailed ship inspection. Using the UNet network, the network performance is optimized by improving the structure of the convolutional an attention mechanism is incorporated to enhance useful features and suppress useless features so as to improve segmentation accuracy. The encoder module obtains high-level semantic information of Synthetic Aperture Radar (SAR) remote sensing of ship images layer by layer, and the decoder module recovers spatial information step by step. Experimental results show that this method can effectively enhance accuracy of image segmentation and retain more image details.