Research on UAV Ground Target Detection Based on Improved YOLOv7

Aiming at the problems of difficult detection of small targets and high miss rate in ground target detection under UAV high-altitude view, this paper proposes an improved YOLOv7-based UAV ground target detection method. Considering deployment to UAVs, the improved algorithm uses the YOLO-tiny network model as the baseline model to ensure the balance of speed and accuracy, adds the Convolutional Block Attention Module (CBAM) to improve the model's ability to learn important features, and utilizes data augment methods such as mosaic and mixup to effectively improve the detection performance for small targets and effectively reduce the miss detection rate of small targets. The experimental results show that the improved YOLOv7 algorithm has improved in all evaluation indexes, mAP@.5 increased by 3.57% compared with the baseline, Frames Per Second (FPS) decreased by 2 compared with the baseline, but still reaches 69.8, with the ability of real-time detection, and the improved algorithm exhibits good performance in the real test.