基于深度学习的古建筑木结构裂缝复杂纹理轮廓特征提取方法

Complex Texture Contour Feature Extraction Method of Cracks in Timber Structures of Ancient Architecture Based on Deep Learning

目的 运用多种目标检测方法进行古建筑木结构裂缝检测,提出能够满足古建筑木结构裂缝检测速度快、精度高等要求的智能算法。方法 首先优化古建筑木结构裂缝图片数据集;其次分别用YOLO、SSD以及Faster RCNN等模型中比较典型的算法进行古建筑木结构裂缝检测;最后从平均损失函数、精度、召回率、平均精度、每秒传输帧数、推理时间、总运行时间和权重等定量指标进行模型综合性能比较分析。结果 在训练300轮的情况下,YOLO v5s模型在4个模型里表现出了最好的综合性能,模型最轻便、裂缝检测速度最快、损失率最小、准确率最高、识别裂缝的正确率最高。结论 YOLO v5s模型比较适用于古建筑木结构裂缝检测,满足速度快、精度高的需求。

The mainstream target detection methods such as YOLO,SSD and faster RCNN are used to detect cracks in ancient wooden structures,and an intelligent algorithm is proposed to meet the requirements of fast speed and high accuracy in detecting cracks in ancient wooden structures.Firstly,the image dataset of cracks in ancient wooden structures was optimized.Secondly,typical algorithms in YOLO,SSD and faster RCNN models were used to detect cracks in ancient wooden structures.In the end,the comprehensive performance of the model was compared and analyzed from the quantitative indicators of average loss function,accuracy,recall,average accuracy,frames per second,interfence time,total runtime and weight.In the case of 300 rounds of training,YOLO V5 s model showed the best comprehensive performance among the four models.The model was the lightest,the fastest crack detection speed,the smallest loss rate,the highest accuracy rate,and the highest accuracy rate of crack identification.YOLO V5 s model is relatively suitable for crack detection of ancient wooden structures,which meets the needs of fast speed and high accuracy.