A Lightweight Border Patrol Object Detection Network for Edge Devices

Border patrol object detection is an important basis for obtaining information about the border area and analyzing determining mission situation. Patrol Staffing now equipped with medium to close range UAVs portable reconnaissance equipment carry out its tasks. In this paper, we designed a algorithm TP-ODA task in order improve UAV of detection, which mostly performed embedded devices limited computing power frame imbalance problem improved; finally, PDOEM structure neck network optimize feature fusion module algorithm. verify improvement effect dataset BDP constructed. The experiments show that, compared baseline model, improves mAP by 2.9%, reduces GFLOPs 65.19%, model volume 63.83% FPS 8.47%. comparison were then combined requirements tasks, it was concluded that more suitable can better fulfill detection.