Focus on Complex Background and Multi-Scale Remote Sensing Images Objects Detection

Remote Sensing images are characterized by multi-scale objects and complex background compared with ordinary natural images. Due to the existence of large size objects and small size objects of remote sensing images, and the problem of imbalance between foreground and background information, it has always been a bottleneck to improve the accuracy of object detection in visible remote sensing images. This paper proposes a detection DA-DET, it contains ResNet-DCN module and Augment-FPN module. Compared with other detectors, DA-DET can extract object features better. ResNet-DCN can extract the features of the object and neglect background information, in addition, Augment-FPN can fusion of low-level spatial feature information with high-level semantic feature information. Specifically, the images is fed into the detection, ResNet-DCN can extract more information on object features and Augment-FPN can fully fuse feature information. Verify the validity of the method in the experiments.