Fabric Weave Pattern Recognition and Classification by Machine Learning

The fabric pattern recognition and subsequently the classification is an imperative task in textiles. Currently, this is done manually, therefore, the need of the requirement is to develop a system that could recognize and classify the fabric weave patterns for ease of inspection and storage. The classification of woven fabrics in today's textile industry is generally manual, requiring significant human effort and a long time. Automatic and effective approaches for woven fabric classification are desperately required with the rapid development of computer vision. This paper proposes an automated and real-time classification technique to analyze three woven fabrics: plain, twill, and satin weave. To achieve the objective, ResNet pre-trained Convolutional Neural Network architecture is used for classification. To obtain texture characteristics, the gray-level co-occurrence matrix and Gabor wavelet, are included in the technique. To eliminate redundancy and maximize main component feature vectors, Principal component analysis is then used to select feature vectors. The experimental result shows that with quicker training speed, the Deep CNN classifier can reliably and efficiently identify woven fabrics. Deep Convolutional Neural Network provides the best accuracy 96.15%.