Fabric screening is a long, tedious, and expensive process. Technology has solved this problem by making automated textile testing programs. In this method, we used and evaluated a few in-depth study models with pre-image processing and convolutional neural networks (CNNs) to detect unregulated errors. We also used multispectral images, integrating standard (RGB) and near-infrared (NIR) images to improve our system and increase its accuracy. We propose two systems: a semi-manual system that uses a simple CNN network to work in different patterns and an automated integrated system that uses modern CNN architectures to work across the database without specifying the previous pattern. Images are pre-processed using a limited variable Histogram. Equalization (CLAHE) to improve their features. We concluded that in-depth learning is effective and can be used for the acquisition of complex patterns. The proposed Efficient Net CNN method provided high accuracy of up to 99% approximately. Multispectral photography is very profitable and produces high accuracy.
₹10000 (INR)
NON IEEE -2022