Cotton percentage prediction from fabric images using transfer learning

Abstract

The textile industry is a prominent and influential sector worldwide, contributing significantly to the economy. This industry has long relied on manual inspections for determining the percentage of cotton in fabric samples, a critical factor for quality control and compliance. However, this conventional approach takes a significant amount of time and is subject to human error. A quicker and more accurate alternative is required due to the rising demand for objective quality assurance and speedier manufacturing cycles. In this article, we propose a computer vision based solution to this challenge. We leverage convolutional neural networks (CNNs), to analyze microscopic images of fabrics and precisely calculate the percentage of cotton. To prove our model's effectiveness, we also made a comparative analysis with several CNN models. The proposed solution achieves a mean absolute error of 3.56 and a mean squared error of 48.78 in detecting the percentage of cotton from microscopic images. The proposed automated method has the potential to affect the industry by removing the subjectivity of manual inspection and saving critical time and money leading to the goal of sustainability.