Improved vehicle target detection algorithm based on YOLOx-tiny for lightweight remote sensing images

In applications such as traffic monitoring and autonomous driving, target detection algorithms need to complete target detection and identification within a short period of time, which requires very high real-time requirements for target detection; while current vehicle detection algorithms have certain challenges to be deployed on resource-constrained environmental devices due to high parameter counts and complex models. In order to solve these problems, this paper proposes a lightweight model, M-YOLOx, which is based on YOLOx-tiny with overall optimization and improvement of the model structure. First, PConv convolution is introduced into the backbone network to reduce the number of parameters and computation of the model; second, a new SPP layer is designed for feature processing, which is used to help the model better focus on the key regions and thus improve the detection performance; finally, in order to reduce the number of parameters and computation, a new feature processing module (DWCA) is designed to classify and regress the feature map. By conducting experiments on DIOR and MS-COCO 2017 datasets, the results show that the proposed M-YOLOx model reduces the amount of parameters by 37.5% and computation by about 51.5% compared with the original model YOLOx-tiny, and comparing with other algorithms, the complexity of the M-YOLOx network is the smallest, which reflects the good performance of the algorithm to a certain extent.