YUCATAN -A New Yarn Quality Prediction Model Using Entropy Optimized Adaptive OTSU Thresholding and Convolutional Neural Networks
With the advent of the unsupervised learning architectures, intelligent manufacturing is becoming the developing direction of textile and clothing industry. Many textile enterprises are exploring the intelligent techniques to improve production efficiency and product quality on each stage of production line in the textile industry. In spinning mills, examining the yarn strength, Yarn thickness and yarn hairiness parameter are considered important for yarn quality prediction. Yarn hairiness or amounts of released fibers from the staple yarn are one of the main aspects to detect the yarn quality. Most existing approaches uses machine learning algorithm for the image acquired during the cotton knitting process in the spinning and weaving. Despite of several advantage of the machine learning models, still it slightly degrades in terms of prediction accuracy and computation cost. In order to enhance the computation performance, a novel yarn quality prediction approach using deep learning technique for all types of images in the spinning mill has proposed which is named as Yarn QUality PrediCtion using Entropy optimized Adaptive ThResholding And CNN (YUCATAN).
Initially Image pre-processing has carried out with noise filtering using improved wiener filter and skewness correction through Hough Transform. Processed image undergo feature reduction and feature extraction process using SIFT technique. Further image segmentation has been carried out employing information entropy optimized adaptive OTSU thresholding model. Feature Selection and feature vector clustering using markov random field. Finally Convolution Neural Network has been employed for selecting the appropriate properties affecting the yarn qualities such as Yarn Evenness, Yarn Strength and Yarn Mass Parameters to generate the quality index. A quality Index has been computed towards effective prediction of the yarn quality in the yarn sequence image. Experimental results of the proposed model on the yarn images outperform the existing Support Vector Machine in terms of computational time, Quality and efficiency.