Research On Image Classification Of Ceramic Tiles

With the rapid development of modern science and technology,the method of manual classification of ceramic tiles in traditional industrial manufacturing can no longer meet the requirements of modern manufacturing industry.With the advent of the industrial 4.0 era,the era of traditional manufacturing industry has come to an end,replaced by the development of high-tech.The research of ceramic tile classifier indicates that human beings will be able to avoid duplication of cheap labor and accelerate the pace of modernization.At the same time,image processing is applied in all aspects of life,and plays an indelible role in military remote sensing,bio-image,monitoring and recognition and other fields.Therefore,it is of great significance to a pply image research to tile classification.This paper studies the influence of different features on the classification accuracy of ceramic tiles produced in the same batch.Starting with the conventional ceramic tile feature extraction method,the gray level co-occurrence matrix algorithm in feature extraction is studied.Then,the feature quantity is extracted by the wavelet packet decomposition selection coefficient energy weighted joint vector(DCWC)algorithm and applied to ceramic tile classification.Aiming at the classification of ceramic tile images,this paper studies the principle of SVM model and proposes a new method.Finally,hypersphere support vector machine(HSSVM)online classification algorithm is used to improve the classification accuracy.Specific research is as follows:First,according to the actual production of the same batch of tiles produced in the same production line,the classification problem needed is analyzed.The tiles image preprocessing,Gauss filtering algorithm and improved steering filtering algorithm are compared to analyze the effectiveness of the steering filtering algorithm to enhance the texture of the tiles image.Secondly,the edge and texture parts of the image correspond to the high frequency part of the wavelet transform,and the corresponding algorithm model is constructed.For feature extraction of ceramic tile image,GLCM and feature extraction algorithm based on wavelet packet decomposition are mainly introduced.The accuracy of experimental classification verifies that the coefficients generated by wavelet packet decomposition represent certain information characteristics,and the texture spatial information of ceramic tile image is reflected by the subimages of the wavelet reconstruction domain obtained from the coefficients.Finally,a tile image classification algorithm based on wavelet packet decomposition selection coefficient weighted reconstruction(DCWC)is proposed.Finally,starting from the concept of support vector machine,this paper introduces the maximum interval and how to find the optimal hyperplane,and converts the solution of the objective function into the quadratic programming problem to find the optimal one.Then,the classification algorithm of support vector machine is introduced,inc luding one-to-one,one-to-many classification algorithm.In practical application,support vector machine(C-SVM)is selected to classify ceramic tile images one by one,and grid parameters are optimized to find the optimal penalty coefficient C and the op timal core radius G.And classify the training set and the test set according to the different number and size.Explain the influence of the size of the data set on the classification results of ceramic tile.Considering the real-time property of ceramic tile classification on pipeline,an online learning classification algorithm based on hyperspherical support vector machine is proposed.With the increase of inspection samples in pipeline production,the training set is continuously expanded and the online training adjustment model is established.The experimental results show that the algorithm is feasible to some extent.In this paper,we need to extract high-level features for classification of such fine images.By introducing sparse self-encoding into deep learning method,we propose that the feature vectors composed of energy coefficients of wavelet decomposition are input into sparse self-encoding to extract second-order features,and the augmented vectors are combined with other features in time domain to train and classify hyperspherical support vector machines.The experimental results show that the sparse self-encoder is used to extract the second-order features of the coefficients in the wavelet domain,and the augmented feature vectors composed of color and texture vectors are used to extract the effective feature vectors.The classification accuracy obtained by the combination of the sparse self-encoder and the improved support vector machine online classification algorithm is higher than that of the convolutional neural network classification,and the disadvantage of over-fitting classification is overcome.Compared with K-nearest neighbor classification algorithm,the method proposed in this paper avoids the shortcomings of large amount of calculation and large deviation of unbalanced prediction in each test sample.