| The quantity statistics and quality inspection of warp-knitted jacquard vamps is an important link in the shoemaking process.At present,the factory mainly uses manual inspection methods to detect the presence of quality problems on the vamps through human vision,manually mark the vamps with quality problems and count the number of qualified vamps.This detection method has disadvantages such as slow detection speed,low detection accuracy,and high labor intensity,which reduces the market competitiveness of enterprises.Using machine vision technology to replace traditional manual inspection is an important way to improve vamp inspection efficiency and accuracy.This subject takes warp-knitted jacquard vamp fabric as the research object,uses machine vision technology to design suitable vamp quantity statistics and quality inspection algorithms,and automatically completes the vamp detection.First of all,combined with the on-site investigation of the factory,understand the factory’s manual inspection process of vamps,analyze the shortcomings of manual inspection methods,and summarize the main quality problems of unqualified vamps.Analyze the research status of machine vision technology and vamp quality inspection methods at home and abroad,and determine the research direction of the algorithm in view of the defects and deficiencies of the current methods.Design and build a machine vision-based warp-knitted jacquard vamp detection system,complete the hardware selection of industrial cameras,lenses,light sources,image capture cards,encoders,etc.Then,the recognition algorithm of warp-knitted jacquard vamp template is studied.Establish a standard template library of warp-knitted jacquard vamps containing 100 different types and styles.Principal component analysis(PCA)method is used to generate the PCA model of the standard vamp template library;solve the minimum Euclidean distance between the PCA model and the vamp feature vector to be tested,and select the vamp template corresponding to the vamp cloth to be tested;introduce singular values decompose to improve the calculation speed and scope of application of the PCA model.Verification of the proposed vamp template recognition algorithm based on principal component analysis shows that the algorithm can accurately identify the corresponding vamp template of the warp-knitted jacquard vamp fabric to be tested.Next,design two sets of warp-knitted jacquard vamp quantity statistics and quality inspection algorithms.The first set of algorithms is based on the normalized crosscorrelation matching algorithm.According to the characteristics of the pictures collected by the industrial line array camera and the algorithm’s requirements for the upper photos,the two upper pictures to be detected are sequentially stitched;the warpknitted jacquard upper is used Image pyramid,reduce image size and ensure image clarity;set an appropriate similarity threshold,detect vamps with quality problems and mark them with crosses,and count the number of qualified vamps.The second set of algorithms is based on convolutional neural network,uses watershed segmentation algorithm to segment the image of vamp cloth morphologically,extracts the target area containing vamp information,removes the background texture information of cloth vamp cloth;improves the Match Net matching network model,design The vamp feature extraction network and the similarity measurement network improve the detection accuracy of the model;the Ada Delta optimization algorithm are adopted to improve the training speed of the model.Finally,the validity and reliability of the two sets of warp-knitted jacquard vamp quantity statistics and quality inspection algorithms are verified through experiments.For three different types of warp-knitted jacquard vamp fabrics,two sets of proposed algorithms are used to perform quantitative statistics and quality inspection of warp-knitted jacquard vamps to verify the accuracy and real-time performance of the algorithm.Through the visualization and quantitative analysis of the detection results,the dete ction accuracy of the algorithm based on normalized cross-correlation matching can reach 96.7%,the detection accuracy of the algorithm based on improved Match Net can reach97.7%,and the detection speed of the vamp fabric detection system can reach0.8 m/s,which is 3～4 times higher than the traditional manual detection method. |
