A Lightweight Method for Road Damage Detection Based on Improved YOLOv8n

substantial challenges in striking an optimal balance between precision and processing speed. Furthermore, owing to the extensive number of parameters, these models present significant challenges for effective deployment on edge devices with constrained computational resources. In order to tackle these challenges, this paper introduces a lightweight road damage detection model, BSE-YOLO, which is founded on the optimization of the YOLOv8n architecture. Initially, we reformulate the feature fusion network by integrating the concept of BiFPN to minimize both the parameter count and computational overhead. Subsequently, a novel SC2f module is introduced and integrated into the feature fusion network, thereby further minimizing both the parameter count and computational requirements. Additionally, this study presents the SEAHead module, which makes use of limited computational resources to obtain vital information, consequently improving both the efficiency and precision of detection tasks while reducing computational costs. The experimental results indicate that, in comparison to the original model, the BSEYOLO algorithm achieves a 40% reduction in parameters, a decrease of 2.2G in FLOPs, an increase of 3 in FPS, and only a 0.1% decline in mAP@0.5. This model effectively fulfills the accuracy and real-time processing demands for road damage detection tasks on mobile edge devices.