Improved YOLO11s-Based Net-Hole Detection for Deep-Sea Aquaculture Platforms
Abstract
Net-hole detection is critical for the structural safety of deep-sea aquaculture platforms, yet automated inspection faces challenges from extremely small damage regions, complex underwater backgrounds, and variable illumination. This paper proposes three improvements to YOLO11s: embedding a Convolutional Block Attention Module (CBAM) for channel-spatial feature re-weighting, introducing a P2 high-resolution detection head expanding detection scales from three to four, and adopting Focal-CIoU loss to focus training on hard-to-localize samples. A systematic ablation study with five model variants on a 2,116-image dataset reveals that CBAM and P2 each provide independent gains, with their combination (+CBAM+P2) achieving the highest detection accuracy at a parameter increase of only 0.09M (+0.95%); however, Focal-CIoU proves strictly detrimental to this balanced-sample task, exhibiting negative synergy. We recommend +CBAM+P2 for optimal detection and +CBAM for high-localization-precision deployment.