基于YOLOv8n改进的混凝土裂缝识别研究

Research on Concrete Crack Identification Based on Improved YOLOv8n

摘要: 混凝土裂缝是评估结构安全性的关键指标，但传统人工检测方法存在效率低、主观性强等问题。研究旨在解决基于YOLOv8n的裂缝识别模型存在的对小裂缝敏感度不足的问题。通过引入注意力机制、优化多尺度特征融合网络及改进损失函数等方法，对YOLOv8n模型进行针对性改进。消融实验表明，改进的YOLOv8n-BiFPN-CA-FocalLoss模型相比基线，mAP50提升11.89%，召回率与精确率分别提升11.78%和12.68%，同时保持95.14 FPS的实时推理速度，各模块协同增益，有效提升了混凝土裂缝检测性能。

Abstract: Concrete cracks are critical indicators for evaluating structural safety. However, traditional manual detection methods suffer from problems such as low efficiency and strong subjectivity. This study aims to address the insufficient sensitivity to small cracks in the YOLOv8nbased crack detection model. The YOLOv8n model is specifically improved by introducing an attention mechanism, optimizing the multi-scale feature fusion network, and modifying the loss function. Ablation experiments show that the proposed improved model, YOLOv8n-BiFPN-CA-FocalLoss, achieves a 11.89% increase in mAP50 compared with the baseline model, while recall and precision are improved by 11.78% and 12.68%, respectively. Meanwhile, it maintains a real-time inference speed of 95.14 FPS. The synergistic effect of each module effectively enhances the performance of concrete crack detection.