Development Of Die-Cutting Product Edge Defect Detection Algorithm Based On Embedded GPU

At present,most of the inspection of the product produced by the die-cutting machine adopts the method of manual sampling to remove the unqualified die-cutting product.However,the manual detection method has the problems of high cost,low detection rate and easy introduction of secondary damage.Therefore,this thesis developed a set of edge defect detection algorithm for die-cutting product based on embedded GPU,which can detect edge defects of die-cutting product with high efficiency and accuracy,and has high engineering application value.Based on machine vision,the image of die-cutting product is obtained through the combination of linear camera and encoder.After the size correction,brightness correction and positioning of the image,a CLAHE-enhanced improved Canny algorithm for edge detection is developed for the edge of the glue area.After erosion and dilation,Zernike moment sub-pixel edge detection is developed for the edge of the none-glue area.For product with angular offset,a second-order moment based on the non-glue edge pixels as the eigenvalues is calculated and corrected.Then,the edge contour of the die-cutting product is segmented,fitted and error calculated.The contour primitives are divided into straight lines and curves by prior information.The linear equation based on least squares is used for straight line fitting,and the least squares based on curve is used for curve fitting.The multiplied fifth-order polynomial fitting determines whether the edge is defective by the relative minimum mean square error of the equation of the standard workpiece corresponding to the edge fitting and the pixel point of the corresponding area of the product to be measured.Finally,in order to improve the performance of the algorithm,for each link of the edge defect algorithm,the embedded GPU Tegra X1 is used to achieve parallel accelerated optimization based on CUDA,and the accuracy and performance of the algorithm are tested.After testing,the edge defect detection algorithm for die-cutting product developed in this thesis can meet the needs of production line functions and performance.The GPU-based edge defect algorithm for die-cutting product is generally controlled within 800ms,and the detection accuracy rate is as high as 95%.