Fabric Defect Detection Based On Fast Active Contour Model And BP-AdaBoost

The rapid development of the manufacturing industry has also brought new challenges for the defect detection of products.Machine vision-based algorithms have gradually replaced human detection.The existing fabric defect detection algorithms has two drawbacks: low efficiency in defect locating and low classification accuracy.In order to solve these problems,a product defect detection algorithm based on fast active contour segmentation model and BP-AdaBoost is proposed.The algorithm is divided into two parts: defect locating and defect classification.The locating of the defect is solved by image segmentation.After summarizing and analyzing the various existing segmentation methods,a fast active contour model is proposed based on the active contour segmentation model.Firstly,three traditional active contour algorithms,C-V model,LBF model and LIF model are analyzed.Since the energy functions of these algorithms are non-convex,they are easy to fall into the local minimum,which is greatly affected by the initial contour position and gray unevenness.Based on these algorithms,the convex optimization technique is introduced to transform the energy function into a convex optimization problem,which solves the problem that the active contour model is sensitive to the initial contour position.At the same time,the solution process is optimized and the algorithm speed is greatly improved.Comparative experiments were carried out on several different types of pictures.The experimental results show that the fast active contour model has high segmentation precision and the speed is significantly better than the traditional active contour model.The average segmentation time is only 0.13 seconds.In view of the practicality of machine learning in image classification,BP neural network is used to complete the classification of defects after analyzing several typical machine learning classification algorithms.The network structure of BP is simple and effective,and has strong versatility.AdaBoost algorithm is applied to further improve the classification performance of BP network.The BP-AdaBoost algorithm is obtained by replacing the weak classifier in AdaBoost algorithm by BP network.The classification comparison experiment is carried out on three kinds of typical databases: MNIST,Fashion-MNIST and TFDS.The results show that the classification accuracy of BP-AdaBoost algorithm on these three databases is significantly better than the SVM and KNN algorithms.And compared with CNN,BP is more in line with the requirements of real-time and accuracy in actual product detection.The typical defect detection algorithm of fabric was designed based on the active contour model and BP-AdaBoost algorithm to locate and classify the fabric defects.For two typical defects of two different types of fabrics: holes and stains,a series of comparative experiments were designed to determine the optimal parameters of the network.On this basis,the algorithm validity experiments of four kinds of defects are carried out.The experimental results show that the algorithm has high classification accuracy and accurate locating performance,and can be effectively applied to fabric defect detection.