A Real-Time UAV Target Detection Algorithm Based on Edge Computing

Small UAV target detection plays an important role in maintaining the security of cities and citizens. targets have characteristics low-altitude flights, slow speeds, miniaturization. Taking these into account, we present a real-time algorithm called Fast-YOLOv4 based on edge computing. By adopting computing platform NVIDIA Jetson Nano, intelligent analysis can be performed video to realize fast targets. However, current iteration edge-embedded has low accuracy poor performance. To solve problems, this paper introduces lightweight networks MobileNetV3, Multiscale-PANet, soft-merge improve YOLOv4, thus obtaining model. The backbone model uses depth-wise separable convolution inverse residual structure simplify network’s its speed. neck adds scale fusion branch feature extraction ability strengthen small-scale detection. Then, predicted boxes filtering function replace traditionally used NMS (non-maximum suppression). Soft-merge model’s by fusing information boxes. Finally, experimental results show that mAP (mean average precision) FPS (frames per second) reach 90.62% 54 f/s, respectively, workstation. In Nano platform, is 2.5 times YOLOv4. This improved performance meets requirements for theoretical significance application value.