Object Detection in Remote Sensing Images Based on Feature Fusion and Multi-Branch Attention

Object detection in remote-sensing images is an important and challenging task. With the development of deep learning technology, the method based on convolutional neural network has made considerable progress. However, due to the problems of remote-sensing images, such as dense arrangement, arbitrary direction and complex background, traditional detection networks are difficult to use adequately the semantic information in images. We design a novel single-stage detector based on feature fusion and three-branch attention. The feature map extracted by the backbone network is fully fused with the semantic information of different levels through the balanced pyramid structure, and then the critical foreground features are captured through the angle parameters decoupled three-branch attention network to improve the detection performance. Experimental results show that our method achieves better detection performance than many state-of-the-art methods.