DGANet: Dynamic Gradient Adjustment Anchor-Free Object Detection in Optical Remote Sensing Images

Remote sensing image object detection has been studied by many researchers in recent years using deep neural networks. However, optical remote images contain scenes with small and dense objects, resulting a high rate of misrecognition. Firstly, this work we selected layer aggregation network updated deformable convolution layers as the backbone to extract features. The classification objects was based on center-point without non-maximum suppression. Secondly, dynamic gradient adjustment embedded into loss function put forward harmonize quantity imbalance between easy hard examples, well positive negative examples. Furthermore, complete intersection over union (CIoU) objective bounding box regression, which achieves better convergence speed accuracy. Finally, order validate effectiveness precision (DGANet), conducted series experiments public datasets UCAS-AOD LEVIR. comparison demonstrate that DGANet more accurate result images.