Optimization of autonomous driving target detection in complex scenarios based on improved YOLOv8

Aiming at the problem of poor detection effect of automatic driving target detection technology in complex scenarios, this paper proposes an improved YOLOv8 model for optimizing automatic driving target detection, whose performance is superior to the basic YOLOv8 model in all aspects. In this study, the Multi-Dilated Channel Refiner (AKConv) module was used to optimize the YOLOv8 backbone network, and the multiple expansion channel refiner (MDCR) module was added to its head to enhance the feature extraction capability and enhance the model's perception ability and detection performance of complex scenes. According to the experimental results, the accuracy, recall rate, F1 score, mAP value and other key indicators of the improved YOLOv8 model in the task of automatic driving target detection have been improved compared with the benchmark model. The mAP value of the improved model reaches 92%, 11.9% higher than that of the original model. It effectively improves the problem of poor detection effect of autonomous vehicle in complex scenes and improves the reliability of autonomous driving system.