一种基于改进YOLO网络的混凝土裂缝识别方法研究OA

In order to improve the accuracy and real-time performance of concrete crack detection,according to the problems of uneven sample distribution and difficulty in identifying small cracks in traditional methods,we proposed an improved YOLO deep neural network.To solve the issue of uneven sample distribution,we adopted data augmentation strategies and introduced the SlideLoss function,which could effectively alle-viate training difficulties caused by class imbalance.Meanwhile,by integrating the LSKNet attention mechanism,we significantly enhanced the model's ability to capture features of tiny cracks.In addition,the original backbone network of YOLOv8 was replaced with the lightweight MobileNetV3,which could improve the accuracy and efficiency of crack identification while reducing computational load and memory usage.Com-pared with the original YOLOv8 network,the improved network achieves increases of 4.2%,3.8%,and 5.5%in recall,precision,and mAP@0.5,respectively.Experimental results show that the proposed improved model achieves high detection accuracy and significantly accelerated running speed in concrete crack identification tasks,providing an efficient and practical crack detection technology for practical engineering applications.