An Improved YOLOv8-based Precise Detection Model for Underwater Waste using SPPF_LSKA and C2F_MLCA

Underwater waste detection is of great significance for marine ecological protection and the safety of underwater facilities. However, due to the complex underwater environment and the optical properties of water, traditional detection methods and the original YOLOv8 model suffer from low accuracy and poor robustness. This study proposes an improved YOLOv8 model for precise underwater waste detection using SPPF_LSKA and C2F_MLCA. The SPPF_LSKA module enhances the feature expression ability through the large-separable-kernel attention mechanism, effectively capturing features of underwater waste at different scales. The C2F_MLCA module combines multi-scale local and global attention mechanisms further to improve the model's adaptability to complex underwater environments. The model was trained for 200 epochs. The experimental results show that the improved model achieves significant performance improvements on the underwater waste detection dataset. Compared with the original YOLOv8 model, the mean average precision at an IoU threshold of 0.5 (metrics/mAP50(B)) is increased by approximately 4.94% (from 0.94274 to 0.98936), the mean average precision across IoU thresholds from 0.5 to 0.95 (metrics/mAP50-95(B)) is improved by about 7.25% (from 0.7806 to 0.83717), and the recall (metrics/recall(B)) is enhanced by around 4.53% (from 0.93084 to 0.97302). In conclusion, the improved YOLOv8 model based on SPPF_LSKA and C2F_MLCA demonstrates excellent performance in underwater waste detection, providing an effective solution for waste detection in underwater environments and showing broad application prospects and practical significance.