A UAV Object Detection Algorithm Based on Improved YOLOX

In aerial photography, UAV will cause small objects and mutual occlusion between objects, bringing enormous challenges to the current one-stage object detection algorithm. In order to solve the above problems, this paper proposes a YOLOX-TBP one-stage object detection algorithm. Based on YOLOX-l, the algorithm adds a prediction head to detect objects of different scales to improve the detection effect of small objects. At the same time, the partial convolution of the backbone network is replaced by Transformer to improve the global information of the model, and a weighted bi-directional feature pyramid network (BiFPN) is introduced to improve the multi-scale feature fusion effect through the method of self-learning weights. The final experiments show that the MAP of the YOLOX-TBP model tested on the VisDrone-DET 2021 dataset reaches 40.27%, which is 0.84% higher than the previous SOTA method (DBNet).