Study On Classification Of Fabric Wrinkling Level Based On Improved Extreme Learning Machine

With the improvement of people’s life quality,the requirements for the wrinkle resistance of clothing fabrics are becoming higher and higher.The classification of traditional fabric wrinkle levels is generally based on human subjective observations.Although the observation results of experienced professionals have great reliability,human fatigue and subjective thought greatly affect the judgment results.Aiming at this problem,a fabric wrinkle classification model based on extreme learning machine was proposed.This model uses texture features and wrinkle levels obtained from fabric images as input and output to train a prediction model,which improved the prediction accuracy and automatic prediction of wrinkle levels.The research object of this thesis is the pure color fabric which is cut and wrinkled.It is the research on the wrinkle resistance of the fabric.The main work and results are as follows:(1)A classification model structure based on extreme learning machine for multi-classification,which can effectively avoid falling into local optimum,is proposed.In order to improve the accuracy of model classification,the sine-cosine algorithm is used to optimize the input weight and hidden layer bias of classification model.To improve the model’s convergence speed,differential evolution is used to initialize the initial population of the sine-cosine algorithm.(2)According to the relevant research and standards of fabric wrinkle classification,this thesis proposed the method of simulating people’s arm wrinkling in actual working environment,collected wrinkle of fabric material,and obtained wrinkle samples with 1 to 5 different levels according to subjective decisions of experts.(3)According to the image processing,this thesis extracted four image features:the wrinkle density,gray level co-occurrence matrix,wavelet parameters standard deviation,and data fusion features of 300 dpi image,which were combined with the level results obtained by expert subjective to establish complete wrinkle level dataset.(4)For the classification result of proposed classification model,11 other classification algorithms were compared in terms of accuracy.Experiment results show that the proposed model can reach 94.17%in classification accuracy,and the minimum classification accuracy of other types of algorithms is only 44.45%,the highest accuracy is 91.67%,and the classification accuracy is significantly improved.In the stability analysis of box-plot,the proposed algorithm in this thesis does not have abnormal points,while other comparison algorithms have abnormal points.In terms of comparison of optimization algorithm convergence speed,the proposed algorithm also has the fastest convergence speed.In terms of significance analysis of Wilcoxon rank sum test,the comparison with the proposed algorithm h is 1,and the proposed algorithm has good significance.