Enhancing Yarn Quality in the Cotton Industry: AI- Based Nep Detection for Improved Manufacturing Processes

The cotton industry places great importance on yarn quality, and its biggest obstacle is nep, which stands for little cotton entanglements that happen during the blowing stage of the yarn manufacturing process. Our goal is to identify these neps as early as possible, allowing us to evaluate the quality of the incoming cotton batch and modify manufacturing settings to reduce the number of neps in succeeding batches. In the end, this approach results in fewer neps and better overall yarn quality. Our AI-based nep detection approach involves installing cameras within carding machines and taking images at a rate of three images per second for a while each day. These images are preprocessed and subsequently processed by a Faster- RCNN model. To improve its training performance in comparison to the prior model, the model has undergone numerous iterations of data annotation and modification. We also demonstrate a relationship between the manual nep count performed by manufacturing executives during the same period and the nep count obtained by this approach. The proposed approach shows competitive performance compared to other available approaches that are presented in the literature.