Introducing AI in garment fault detection using YOLOv5 to reduce bottleneck

Abstract

In order to reduce manpower and bottleneck in the inspection system of industrial garments, we explored the application of the YOLOv5 model using our very own dataset of defective clothing pieces. The economy of Bangladesh heavily depends on the garment industry. However, in this day and age of advanced technology, it is getting harder to have efficient manpower in the garment industry. Motivated to solve this problem, we decided to devise ways to explore AI implementations, particularly in the Bangladeshi garment industry system. Since there is no existing and efficient defective garment dataset specifically for our desired research work so we created our own dataset. This dataset has a total of 2,525 images and 7 different classes. By thoroughly analyzing our data from pre-processing to its performance after the application in the YOLOv5 model, we have tried to create a useful dataset. The models have achieved a good mean average precision across all 7 classes. Our research has only scratched a small surface of an area of interest where advanced AI and machine learning technologies can bring a lot more advancement.