Multi-Attention Network for Sewage Treatment Plant Detection

As an important facility for effectively controlling water pollution discharge and recycling waste resources, accurate sewage treatment plant extraction is very protecting quality, function, sustainable development of the environment. However, due to presence rectangular circular facilities in plants, shapes are diverse scales different, resulting poor performance conventional object detection algorithms. This paper proposes a multi-attention network (MANet) plants using remote sensing images. MANet consists three major components: light backbone used obtain multi-scale features, channel spatial attention module that realizes feature representation dimension dimension, scale scale-aware features. The results from extensive experiments performed on dataset suggest our proposed exhibits superior compared with other competing methods. Meanwhile, we well-trained model predict GF-2 data Beijing area. By comparing manually obtained method can achieve accuracy 80.1% while maintaining recall rate at high level (90.4%).