基于改进Yolov5s的混凝土坝裂缝检测研究

Crack Detection of Concrete Dam Based on Improved Yolov5s

大坝作为我国的重要水利工程,其安全性不言而喻。裂缝作为大坝的主要安全隐患,大坝裂缝的智能化检测,对管控大坝风险具有非凡的意义。提出一种基于改进YOLOv5s目标检测算法的大坝裂缝检测方法。首先,在YOLOv5s的主干特征提取网络中引入轻量级GhostNet中的Ghost模块,对YOLOv5s的主干网络进行优化,得到轻量级的模型YOLOv5s-Ghost,以降低模型的复杂度,提高裂缝的检测速度;然后在模型预测输出端融合高效的CA(Coordinate Attention)注意力机制进一步增强裂缝特征提取能力,提高裂缝模型检测的性能。实验结果表明:该方法与现有的YOLOv5s相比,模型大小复杂度降低了44.8%,准确率提升了2.6%,验证了改进方案的有效性,提高了裂缝检测效率。通过引入GhostNet中的Ghost模块和融合CA注意力机制相结合,使得YOLOv5模型的复杂度降低,参数量减少,实现了对裂缝检测速度与精度的提高,增强了网络性能。

As an important hydraulic project in China,the safety of dam is self-evident.Cracks are the main potential safety hazards of dams.Intelligent detection of dam cracks is of great significance for dam risk management and control.In this paper,a dam crack detection method based on improved YOLOv5s target detection algorithm is proposed.Firstly,Ghost module in lightweight GhostNet is introduced into the backbone feature extraction network of YOLOv5s to optimize the backbone network of YOLOv5s,and a lightweight model YOLOv5s-Ghost is obtained to reduce the complexity of the model and improve the speed of crack detection.Then,the effective CA(Coordinate Attention) attention mechanism is fused at the output of model prediction to further enhance the ability of crack feature extraction and improve the performance of crack model detection.The experimental results show that compared with the existing YOLOv5s,the complexity of model size is reduced by 44.8%,which verifies the effectiveness of the improved scheme in this paper and improves the efficiency of crack detection.By introducing Ghost module in GhostNet and combining CA attention mechanism,the complexity of YOLOv5s model is reduced and the number of parameters is reduced,the speed and accuracy of crack detection is improved and the network performance is enhanced.