Research On Integrity Inspection Algorithm Of Product Trademark Based On Machine Vision

Quality evaluation and control is necessary in the modern production line of industrial products.Comparing with the traditional artificial method,machine vision technology has advantages of preciseness,continuance and high efficiency,and it is significant to apply machine vision technology to the defects inspection,to improve the production efficiency and the competitiveness of enterprises.Aiming at the trademark integrity inspection of industrial products in packaging industry,an efficicent and fast inspection algorithm based on machine vision is proposed in this thesis.In view of the focal points and difficulties in the design of trademark integrity inspection algorithm,the framework of the trademark integrity inspection algorithm and the key technologies of each module are designed,and a trademark integrity inspection algorithm based on shape feature extraction and Support Vector Machine is proposed in this thesis.Key points of the algorithm are as follows:Anobjectextractionalgorithm,basedonimagesegmentationand connected-component labeling,is applied in the process of image preprocessing,in which the source image is binarized by the image segmentation algorithm in the maximum contrast channel,and the connected-component labeling method is modified in the object locating process,to improve the accuracy of object extraction.A fast shape feature extraction algorithm is proposed,which firstly uses the color image segmentation method based on K-means clustering to get the trademark,and in order to accelerate the convergence and improve the efficiency,the initial clustering centers are updated according to the clustering results,then Hu invariant moments of the trademark are calculated,and Principal Component Analysis is used to reduce the irrelevant features and the redundancy,while the description ability is retained.Finally,a trademark integrity recognition algorithm based on Support Vector Machine is designed.The lower dimensional feature descriptors,which has a bad linear separability,is mapped to the higher dimensional feature space by the RBF kernel,and the linear classification of the trademark integrity is realized in the higher dimensional space,and the cross validation method is used,to search for the parameter values with the best performance.After algorithm porting and the DSP-specified optimization,the speed of the trademark integrity inspection algorithm reaches 13 fps with the accuracy more than 90%in the DSP-based embedded system,and the algorithm is illumination invariant within the lux range of about±30%,and has good performance for defects whose size are more than 0.5mm~2.