Improved YOLOv3 algorithm for small object detection in remote sensing images

Aiming at the low detection performance of YOLOv3 algorithm in remote sensing image small target detection, an improved YOLOv3 remote sensing image small target detection algorithm is proposed. Firstly, the problem of insufficient feature extraction of small targets is solved by reconstructing the residual block of darket-53. Secondly, a shallow feature enhancement module is proposed, which expands the receptive field of the shallow feature map and enriches the small target feature information of the shallow network by using the multi-branch cavity convolution structure and attention mechanism. Finally, the EIOU loss function is used to replace the target location loss function in the original algorithm, which makes the model converge quickly and reduces the rate of missing detection. The experimental results show that the improved algorithm has a good detection effect on VisDrone dataset. mAP@0.5 saw a 3.3 percent increase compared to YOLOv3. This shows that the improved algorithm can detect small targets in remote sensing images more effectively.