PREDICTION OF YARN QUALITY BY DEEP BELIEF NEURAL NETWORK
Abstract
Many textile industries are utilizing the artificial intelligence techniques to enhance the production quality and product performance on each phases of production unit. In spinning mills, examining the yarn parameters such as unevenness, strength and mass are considered important for yarn quality prediction. In order to improve the performance of yarn quality, an innovative quality yarn prediction using Deep Belief Neural Network (DBNN) approach for yarn images are proposed. Initially pre-processing of the images is carried out using deinterlacing. Deinterlacing is a process which introduces the missing lines of encoded image sequence on spatiotemporal domain. It is employed to recognize and initialize weight to the important features and biases. It used for determination of the quality among the pair of units on heterogeneity information of the yarn. Feature extraction and feature selection is carried out by constructing the feature vector using principal component analysis. DBNN is employed for obtaining the latent and optimal features which represents the Yarn Evenness, Yarn Strength and Yarn Mass Parameters. It is considered as yarn quality determining features. The quality of the yarn is determined with respect to quality index of the features explored in deep learning processes. Experimental results obtained from the proposed model on identifying the quality of the yarn images is outperforming in terms of scalability and efficiency on comparing it against traditional deep learning architecture such as Convolution Neural Network.
Keywords:
:: : Yarn Quality, Feature Extraction, Deep Learning, Yarn Evenness, Yarn Hairiness, Deep Learning, Deinterlacing
Full Text:
PDFReferences
Refbacks
- There are currently no refbacks.