An object detection algorithm based on the combination of the improved YOLOv8 and Mamba models

In response to the problems of mutual occlusion of road targets, significant differences in target sizes, and blurring of target edges caused by extreme weather conditions such as smog. This paper proposes an object detection network P-Mamba-YOLOv8 based on the combination of the improved YOLOv8 and Mamba models. The Mamba model is introduced to optimize feature extraction, the selective boundary aggregation module to optimize multi-scale feature fusion, and the DyHead-DCNv3 module and the small target P2 detection head to optimize the feature detection stage. The experimental results show that compared with the original YOLOv8 network, the average accuracy mAP50 of the improved network has increased by 2.8% in smoggy weather and by 1.3% in sunny weather.