Multi-Content Complementation Network for Salient Object Detection in Optical Remote Sensing Images

In the computer vision community, great progresses have been achieved in salient object detection from natural scene images (NSI-SOD); by contrast, optical remote sensing (RSI-SOD) remains to be a challenging emerging topic. The unique characteristics of RSIs, such as scales, illuminations and imaging orientations, bring significant differences between NSI-SOD RSI-SOD. this paper, we propose novel Multi-Content Complementation Network (MCCNet) explore complementarity multiple content for Specifically, MCCNet is based on general encoder-decoder architecture, contains key component named Module (MCCM), which bridges encoder decoder. MCCM, consider types features that are critical RSI-SOD, including foreground features, edge background global image-level exploit them highlight regions over various scales RSI through attention mechanism. Besides, comprehensively introduce pixel-level, map-level metric-aware losses training phase. Extensive experiments two popular datasets demonstrate proposed outperforms 23 state-of-the-art methods, both RSI-SOD methods. code results our method available at https://github.com/MathLee/MCCNet.