Research And Implementation Of Micro Parts Detection And Measurement In Industrial Scene

With the proposal of industry 4.0 and the increasing innovation of science and technology,the measurement requirements of domestic machinery manufacturing related enterprises for micro parts are gradually improving.The traditional manual measurement methods have the problems of poor accuracy and low efficiency,which can not meet the current measurement needs.At the same time,precision measuring instruments have high cost,poor antiinterference ability,and can not meet the needs of real-time measurement of enterprises.The measurement technology of micro parts based on machine vision has the characteristics of high speed,high precision,low cost and non-contact.Therefore,the detection and measurement methods of micro parts in industrial scene are studied by using computer vision technology.The specific research and development work is as follows:(1)Aiming at the problems of poor anti-interference ability and small application range of the traditional shape detection algorithm,after in-depth research and analysis,an improved object detection algorithm based on R3 Det network model is proposed to detect the shape position and category in the part image.R3 Det uses Retina Net as the basic network and adds a feature refining module to solve the problem of feature misalignment,which improves the accuracy of classification and regression.The algorithm can still detect correctly when there are defects on the edge of the part,and can detect any shape without designing different detection algorithms according to different shapes.To solve the problem of low detection accuracy of objects with large aspect ratio,a shape discrimination module is added after object detection to screen the prediction frame,which improves the detection accuracy of objects with large aspect ratio.In the actual production process of enterprises,the application scope of the algorithm is expanded and meets the diversified measurement needs of enterprises.(2)Aiming at the slow speed of sub-pixel edge extraction,a fast sub-pixel edge location method based on pixel level contour and Zernike moment is proposed.By comparing the common image preprocessing methods,the median filter and OTSU binarization are used to preprocess the image.Due to the large part image,the speed of sub-pixel extraction of the whole image is slow.Therefore,firstly,the parts are located and trimmed according to the pixel level contour information,and then the sub-pixel edge location method based on Zernike moment is used to improve the edge accuracy,which not only improves the measurement accuracy,but also ensures the measurement speed and meets the needs of realtime measurement of enterprises.(3)Aiming at the problem that the edge of micro parts is damaged,resulting in the deviation of fitting curve,a step threshold least square fitting method is proposed.This method fully considers the causes of errors caused by "outlier" points,removes them and then fits them,so as to obtain a high-precision fitting curve.The sub-pixel edge location method based on Zernike moment is combined with the step threshold least square fitting method to complete the accurate measurement of the size information of micro parts.Finally,combining the improved object detection method based on R3 Det network,fast subpixel edge location method based on pixel level contour and Zernike moment and step threshold least square fitting method,a micro part measurement software is designed and implemented,which can automatically measure micro parts.The processing time of a single image is less than 1s and the measurement accuracy is about 0.003 mm.The usability and flexibility of the software are verified by testing each function of the software.The software has been applied in the actual production of relevant enterprises in Guilin.