【摘要】 针对桥梁高强度螺栓松动检测工作量大、目标小、异常多且难以获取等问题，该文提出一种半监督深度学习模型，即使少量负样本情况下也可得到螺栓松动检测模型，解决了模型训练样本不平衡的问题。YOLOv5-CT模型对螺栓目标检测的精度达98.33%。通过对螺栓数据进行预处理，提高Ganomaly模型对螺栓图像的重构能力。当隐空间向量值为100时，模型的S_AUC最高，具有最佳判别性能。在模型测试阶段，将异常分数阈值设置为0.295，计算模型对高强度螺栓异常松动检测的精度可达到85%以上，实现螺栓的自动识别和检测。更多还原

【Abstract】 High-strength bolt loosening detection of bridges faces problems such as heavy workload, small targets, many anomalies, and difficult collection. Therefore, this paper proposed a semi-supervised deep learning model, which could obtain the bolt loosening detection model even with a small number of negative samples and solve the problem of unbalanced model training samples. The accuracy of the YOLOv5-CT model for bolt target detection reached 98.33%. By preprocessing bolt data, the reconstruction ability of bolt images by the Ganomaly model was improved. When the hidden space vector value was 100, the model had the highest S_AUC and the best discriminant performance. In the model test stage, the threshold of abnormal fraction was set to 0.295, and the accuracy of the calculation model for abnormal loosening detection of high-strength bolts could reach more than 85%. As a result, the automatic identification and detection of bolts were realized.更多还原