基于轮廓度量与AI模型的工业缺陷检测算法研究

Research on industrial defect detection algorithm based on contour measurement and AI model

为了解决工业产品表面缺陷难检测的问题,针对当前检测技术现场应用痛点,从方案架构和检测技术两个层面进行研发,设计了基于轮廓度量和AI模型的工业缺陷检测算法。首先,结合图像分析算子和AI模型,建立一套完整检测方案架构。然后,耦合形状度、填充度和边数等特征,定义一种代表缺陷不规则度的轮廓度量因子,结合阈值分割,完成灰阶算子的设计;利用SIFT特征点检测机制和ICP校准方法,完成缺陷图像的匹配;建立传统视觉算子,并基于AI分割模型、分类模型与小样本训练,构建工业缺陷检测模型,准确定位到缺陷边缘区域。实验测试结果显示,相较于目前主流技术与方案,所提算法方案有利于“保检出,降误判”的应用诉求,为工业缺陷检测设备奠定核心算法基础,并确保落地。

In order to solve the problem that it is difficult to detect the surface defects of industrial products, research and development are carried out from the two levels of scheme architecture and detection technology, aiming at the pain points of the current on-site application of detection technology, the industrial defect detection algorithm based on contour measurement and AI model is designed. Firstly, combine image analysis operator and AI model to establish a complete detection scheme architecture. Then, by coupling shape degree, filling degree and edge number,the contour measurement factor representing the irregularity degree of defects is designed. Combined with threshold segmentation, the gray-scale operator is designed;the matching operator is designed by using SIFT feature point detection and ICP calibration. Finally, based on the depth vision segmentation model, classification model and small sample training, the industrial defect detection model is constructed to distinguish the defect from the normal area of the product. The experimental results show that, compared with the current mainstream technology and scheme, the algorithm scheme in this paper is conducive to the application demand of "ensure detection, reduce misjudgment",and lay the core algorithm foundation for industrial defect detection equipment.