Research on Road Object Detection Method Based on Lightweight YOLOv8s

This study introduces SLL-YOLO, a lightweight road object detection framework that strategically enhances YOLOv8s through three synergistic architectural optimizations while maintaining detection accuracy. The model first integrates the StarBlock module from StarNet to reconfigure the original C2f structure into a C2f-Star variant, effectively reducing computational complexity through cross-channel feature reusability while enhancing multi-scale discriminative capability. Building upon this backbone refinement, a shared-convolution detection head is implemented to unify feature processing across hierarchical prediction layers, eliminating redundant parameterization while improving spatial-semantic coherence. The architecture is further streamlined through layer-adaptive magnitude-based pruning (LAMP), which systematically removes non-critical weights identified via gradient-weighted channel importance metrics, thereby substantially optimizing deployment efficiency without compromising essential detection pathways. These coordinated modifications establish an optimal balance between operational speed and recognition precision, specifically tailored for real-time automotive vision systems.