基于图像统计建模的工业过程产品质量监测

Image statistical modeling based production-quality inspection in industrial process

在机器视觉非接触式的在线检测技术中,为了有效感知图像局部子块的随机堆积状态,提出一种基于图像空间结构统计建模的工业产品质量智能鉴别方法。从理论分析了自然图像的局部同质碎片颗粒的韦伯分布(WD)过程;采用多尺度全向高斯导数滤波方法获得图像空间结构细节,基于WD统计建模获得图像空间结构在不同高斯观测尺度下的特征描述;最后,基于最小二乘-支持向量机(LS-SVM)原理建立工业产品质量智能分类模型。在大米加工品质自动分类和针织坯布质量智能鉴别中进行的验证性和对比性测试结果表明,所提出的方法因能有效表征为大量局部碎片颗粒随机堆积而成的复杂纹理图像空间结构的视觉感知特性,工业产品质量智能分类准确率高,为实现工业流水线上产品质量的智能分级和高质量产品的自动生产包装奠定了基础。

A method based on statistical modeling of the spatial structures of the images to identify the product-quality is presented. In advance, the image spatial structures are proved to be subject to a typical Weibull distribution （WD） theoretically. A method of multi-scalar and omni-directional Gaussian deriva- tive filtering is adopted to extract the spatial structure details of the monitoring images. Hence, a signifi- cant omni-directional statistics feature vector of the filtering response based on WD model with various observation scales is obtained. At last,a least square support vector machine （LS-SVM） based product- quality classifier is established to identify the production quality automatically. The proposed method is tested in the fields of automatic identification of the rice production-quality and intelligent assessment of the surface quality of the greige cloth in the corresponding assembly lines of industrial scale. Experimen- tal results indicate that the proposed image statistical modeling method can effectively characterize the statistical distribution of the spatial structure of these complex texture images piled with a large number of local debris particles randomly. Hence, the automatic Classification model can achieve high accurate rate. The proposed method lays a foundation for the intelligent perception of the product-quality and the automatic processing of the high quality products in the assembly lines.