摘要
为了研究不同缺陷检测方法对漆包线缺陷检测系统及缺陷分类精度的影响,提出了一种单相机全视角缺陷检测的研究方法,以利用机器视觉技术代替传统人工检测提高漆包线整体质量。首先,构建了漆包线全视角缺陷检测系统,对系统所采集到的图像进行预处理,经图像扩增制作了数据集。然后,研究了因漆包线个别缺陷较小导致识别精度过低的问题,设计了HOG特征提取算法提取漆包线图像特征,将处理后的图像放入支持向量机(SVM)中进行缺陷的分类识别。对于YOLOv5s在小目标识别时存在漏检情况,设计了一种以YOLOv5s为基础的优化算法。最后通过3种缺陷检测模型对比得出,支持向量机的平均准确率为83.78%,YOLOv5s目标检测法为96.92%,而优化后的YOLOv5s模型准确率可达到98.37%。实验证明,优化后的YOLOv5s目标检测法能够有效增加实时漆包线表面缺陷的检测精度。所提缺陷检测研究理论对提高特种线缆的表面缺陷检测精度具有重要的指导意义。
In order to study the influence of different defect detection methods on the defect detection system of varnished wires and the accuracy of defect classification,a research method for full view defect detection with a single camera is proposed,to improve the overall quality of varnished wires by using machine vision technology to replace traditional manual inspection.First of all,a full view defect detection system for varnished wire is constructed,the images collected by the system are preprocessed,a dataset is created through image augmentation;then,the issue of low recognition accuracy caused by small individual defects in varnished wire is addressed,HOG feature extraction algorithm is designed to extract image features of varnished wires,the processed image is put into SVM for defect classification and recognition.To solve the problem of missed detection of small object recognition,an optimized algorithm based on YOLOv5s is designed.Finally,the comparison of the three defect detection models shows that,the average accuracy is 83.78%by SVM,is 96.92%by YOLOv5s object detection method,and the accuracy can reach 98.37%by the optimized YOLOv5s model.Experiments proves that,the optimized YOLOv5s object detection method can effectively increase the detection accuracy of real-time surface defects of varnished wires.The proposed defect detection research theory has guiding significance for improving the surface defect detection accuracy of special cables.
作者
夏扬
于正林
Xia Yang;Yu Zhenglin(Jilin Institute of Electronic Information Products Inspection,Changchun 130012,China;Chongqing Research Institute,Changchun University of Science and Technology,Chongqing 401122,China)
出处
《机电工程技术》
2025年第15期115-122,共8页
Mechanical & Electrical Engineering Technology
