The Application Of Extreme Learning Machine In Textile Image Processing

In modern textile weaving industry,the quality of textiles directly affects the sales,and improves the competitiveness of the industry only by improving the quality of textiles.Generally speaking,the key points that affecting the quality of textiles are color difference and defects.In the traditional textile industry,the detection of color difference and defects mainly rely on workers who with rich experience in the use of human eyes.However,the detection is inefficient due to the harsh environment and visual fatigue of technical workers,and it is strongly dependent on the subjective awareness of technical workers.Therefore,it is very meaningful to solve the problem that traditional detection method relies on manual and improve the detection accuracy by using computer vision.The research work of this paper is based on the machine learning technology to establish the model of color difference classification and defect detection with good stability,and solved the problem of color difference and defect in dyeing and weaving process.The main work and achievements of this paper are summarized as follows:(1)The basic concepts and research status of color difference and defect are briefly introduced,analyze and compare the advantages and disadvantages of various color difference methods and defect detection algorithms.The algorithm that has better classification performance based on Extreme Learning Machine(ELM)is mainly studied,and the working principle and the existing problems of the ELM are analyzed and discussed.The defect detection algorithm that has good performance based on dictionary learning is mainly studied,and the theory basis and exiting problems of dictionary learning method are analyzed and discussed,which provide a theoretical basis for follow work.(2)To develop a color difference classification model for dyed fabric,we proposed a novel method that based on differential evolution algorithm(DE)with dynamic parameters selection to optimize the regularization extreme learning machine(RELM).Firstly,the DE algorithm that has well global searching ability is used to iterative optimize the input weights and hidden bias to solved the problem that traditional ELM generated input weights and hidden bias randomly.Secondly,the experience risk only considered of ELM in the process of calculated the output weights,so the RELM is introduced to prevent illness solution.Finally,the dynamic parameters selection is used to select the optimal classification model.The experimental results show that the model has higher accuracy,good stability compared with DE-ELM method.(3)Textiles are woven by the specific rules of warp and weft yarns and the textured of the textiles surface presented high correlation,so the defect regions can be treated as local anomaly textures.Considering the texture structure and size of defect in textile surface,we proposed a novel defect detection algorithm based on multi-scale dictionary learning,the multi-scale dictionary that we proposed can describe textile image detail more clearly.Due to the intensive computation of original dictionary learning algorithm,so in the process of learning dictionary,the improved KSVD is adopted in this paper,speed up dictionary elements update.And a unique adaptive differential evolution algorithm optimized regularization extreme learning machine is proposed to establish fabric defect detection model.In the phase of training stage,the self-adaptive mutation operator is used to solve the parameter setting problem of differential evolution algorithm.Experimental results show that compared with traditional Gabor filter method,morphological operations and local binary pattern,the proposed method can accurate locate defect areas,and achieve high detection rate.Furthermore,the proposed algorithm also has very good performance for defect detection of pure color fabric.