Development Of Die-cutting Duct Size Measurement And Surface Defect Detection Algorithm Based On Embedded GPU

The size measurement and defect detection of the current die-cutting product are mainly manual sampling tests to determine whether the product is qualified.This method is labor-intensive,and may introduce external damage during manual measurement and human eye detection.And more importantly,it is impossible to achieve full inspection,failing to adapt to increasingly strict quality requirements.Therefore,this thesis researches and develops a machine vision-based size measurement and defect detection algorithm,which can perform high-speed and high-precision detection of all the die-cutting product.Based on machine vision,the image is obtained by the combination of the line camera and the encoder.After the brightness correction and contour location,the image is enhanced by the combination of detail enhancement and brightness pull-up.SUSAN combined with the Canny operator is used for the edge of the non-glue zone.The differential and binary edge detection algorithm is used for the glue region after the gray value expansion and corrosion operation.And then the Hough space parameters of the line and the arc will be calculated,and then segments the edge points.Finally,the fitting method of the straight line and the arc combined with the coarse error and the least squares is used to obtain the size measurement results.Aiming at the defects of die-cutting product,this thesis proposes an improved algorithm based on the differential image method to combine the statistical features of single pixel points with the average value of image grayscale to obtain dynamic thresholds,which can achieve defects such as dirt,scratches and damage of die-cut product.In order to improve the performance of the algorithm,this thesis uses embedded GPU to realize parallel optimization of the algorithms for each part of size measurement and defect detection,and analyzes the running time of each algorithm module.After testing,the die-cutting product detection algorithm can meet the function and performance requirements of the actual die-cutting production line.The detection performance of the die-cutting product detection algorithm on the online die-cutting product is about 9 meters per minute.The detection accuracy of defect is over 97%.