Lightweight highway debris detection algorithm based on improved YOLOv8n

To address the issues of poor detection performance, high false positive and missed detection rates, and low real-time capability in highway debris detection algorithms, a streamlined detection algorithm leveraging an enhanced YOLOv8n model is introduced. Firstly, the backbone HGnetv2 network from the Real-Time Detection Transformer model is integrated to optimize the detection functionality of small objects and debris in complex scenarios. Secondly, the improved CSP-PartialConv structure is used to replace C2f in the neck network, reducing parameter volume and computational demand. Additionally, the model's multi-scale feature fusion capability is improved by incorporating a weighted bidirectional feature pyramid network. Finally, the WIoU loss function is utilized for replace CIoU, optimizing the network's bounding box loss, and enhancing the model's convergence rate and accuracy in regression. The trial findings indicate that the improved model presented in this paper improves accuracy by 0.5% over the original model, while also reducing the parameter count by 24.1%, computational cost by 11%, and model size by 22.3%. This demonstrates a notable lightweight effect and improved detection performance, providing effective technical support for highway safety management.