Key technology for intelligent detection of colour defects in the clothing

Along with the growing apparel industry, there is a growing demand for clothing. As the colour of clothing is the first impression consumers have of their clothes, it has become a technology being explored by clothing companies to effectively detect colour defects in real-time online. Traditional methods of detecting colour defects in clothing are usually performed manually and lack standard and automated tools. In this paper, we build on the YOLOv5 neural network algorithm for real-time online detection of garment colour defects and improve the YOLOv5 algorithm based on the fact that most garment colour defects are detected by small targets. The small target detection capability of YOLOv5 is further improved by introducing an attention mechanism and improving the loss function. The improved algorithm improves by 12.1% in terms of accuracy and 5.9% in terms of recall, which shows that the improved YOLOv5 algorithm is more effective in detecting the small targets commonly found in garment colour defect detection.