基于图像学习的透明构件缺陷检测方法研究

Research on defect detection method of transparent component based on image learning

光学PC板、玻璃幕墙等透明构件中存在的气泡、缺胶等先天缺陷严重影响着设备的运维保障。因此,提出了一种透明构件缺陷检测方法。首先,基于视觉图像特性分析了透明构件缺陷检测机制;其次,改良了高斯滤波模型,建立了边缘特征检测模型,并确立了缺陷区域分割模型,以降低环境条件的干扰;再次,在进行缺陷区域分级的基础上,建立了依据缺陷特性的强特征参数优选方法,以提高缺陷的精准分类效果;最后,搭建了双通道图像检测试验平台对典型试验对象进行了测试验证,验证了视觉图像学习方法应用于透明构件缺陷检测的有效性和实用性。

Due to the complexity of the production process and service environment,transparent components such as optical PC board and glass curtain wall often generates different degrees of bubbles,scratches,starvation,and other tiny defects.When the structure is in the service environment of cyclic load,it is very easy to induce cracks,bursts,self-explosion,and other functional failures,which seriously affect the operation and maintenance of the equipment.Therefore,a defect detection method for transparent components was proposed.Firstly,based on the analysis of the defect types of transparent components and the different values of their plane gray levels,the basic structure of the defect detection system for transparent components was proposed,and the defect feature recognition process was established.Secondly,when the contrast of the original image is low,the defect edge will be nebulized.Therefore,to improve the image preprocessing effect,the Gaussian filter model was improved.At the same time,based on the analysis of the influence of weak edge performance sensitivity on feature extraction,the edge feature detection model was established,and the defect region segmentation model was further established to reduce the interference of environmental conditions.Thirdly,based on defect region classification,a strong feature parameter optimization method that depends on defect image characteristics was established to improve the accuracy of defect classification.Finally,the dual-channel image detection test platform was built,and the typical test objects were tested and verified.The test results show that all the detected defects by the dual-channel test platform were correctly classified,which is consistent with the engineering practice.For excellent measure,the recognition efficiency of the dual-channel image detection test platform was better than that of the single-channel image detection test platform.Therefore,it is proved that the visual image learning method is effective and practical in the defect detection of transparent components.