Research on vehicle detection algorithm based on drone image

To improve the accuracy of vehicle detection, this paper proposes an intelligent model to meet the real-time needs of vehicle detection based on the current common drone shooting methods. To reduce the computational parameters in the feature extraction layer, this paper uses depth separable convolution as the feature extractor. Compared with the conventional convolution kernel that processes the spatial and channel features of the feature map at the same time, the depth separable convolution kernels use depth convolution and point-by-point convolution to extract and learn the spatial and channel features of the image respectively. In addition, aiming at the problems of existing deep learning object detection models such as high hardware requirements and unstable cloud server detection performance, this paper proposes a lightweight end-to-end vehicle detection model that can be deployed on drone embedded hardware terminals and perform real-time inference. Finally, the reliability of the proposed model is verified by experimental research, thus providing a reliable technical reference for subsequent vehicle detection.