Vehicle Detection Under Unmanned Aerial Vehicle Based on Improved YOLOv3

Under the unmanned aerial vehicle (UAV), the ground vehicle target is generally small in size, and the background information is more prominent. Therefore, we propose an improved detection model based on YOLOv3. First, in order to make full use of network feature information, we add connections between the front and back layers of the network to improve the information flow in the deep network. These layers are constructed as a loop and alternately updated structure, which reduces the depth of the network and improve the reuse of features. In addition, on the one hand, considering that the vehicle target is small in the UAV scene, the imbalance between the foreground and the background category is extremely serious. We improve the loss function and reduce the weight of the easily discernible background confidence loss, so that the model pays more attention to the hard samples. On the other hand, the improved IoU index is used as the loss function of regression task to improve the accuracy of small target detection. Experiments on the vehicle dataset of UAV show that the detection accuracy of the proposed algorithm is improved without reducing the rate.