Lightweight Object Detection Method Based on Improved YOLOv8

Although the existing object detection model has good performance, it is not suitable for deployment and operation on edge devices such as smart cars due to its many parameters, large amount of computation and complex model. In order to improve the detection accuracy of smart cars in the distribution process, a lightweight object detection method based on improved YOLOv8 is proposed, and partial convolution is replaced by AKConv convolution. Reduce the number of parameters in the model; The original loss function CloU is replaced by WloUv3 to reduce the influence of low-quality target samples on the results. To further reduce the size of the model, the detection Head is replaced with Adaptive Head. Experiments show that compared with YOLOv8n in VOC data set, mAP50, Recall and Precision of the proposed target detection method increase by 1 %, 1 % and 2.8% respectively, and the amount of computation and parameter number decrease by 13.6% and 12.9% respectively. Compared with YOLOv8n, the new model has the characteristics of fewer parameters and higher detection accuracy, which has greater advantages for deployment on edge devices such as intelligent vehicles.