Research on remote sensing target detection algorithm with improved YOLOv3

This is the authors' abstract. We don't add key points for this paper.

The issue of inadequate precision in identifying airplanes in remote sensing images arises from fluctuations in lighting conditions. To tackle this problem, a refined approach for detecting targets in remote sensing using the YOLOv3 algorithm is proposed in this study. Primarily, a CBAM attention module (Convolutional block attention module) is introduced to emphasize crucial regions of the input object, thereby obtaining more significant information. Secondly, the Darknet53, acting as the backbone network in YOLOv3, is substituted with ResNet152 to enhance the network's capability in extracting features. Experimental results reveal that the algorithm achieves 92.75% mAP in the DOTA dataset, addressing the issue of insufficient detection accuracy caused by lighting variations in remote sensing images to a certain extent. This outcome demonstrates the feasibility and efficacy of the algorithm.