Fabric Finesse: Harnessing YOLOv8 for Enhanced Textile Detection

Nowadays, it is not merely a goal but necessary to ensure impeccable fabric quality in the textile industry. Failing to ensure its quality, can lead the industry to greater financial losses, disrupt the supply chain, and harm a company’s reputation. Traditionally, manual inspection is done by humans which is prone to human error, fatigue, and inefficiency, and leads to low accuracy and missed defects. To overcome these challenges, an advanced framework for defective fabric detection is introduced in this study. The modern world has enriched us by providing sophisticated precision and automation of fabric detection methods which offer a revolutionary alternative and have become paramount to ensuring product quality and optimizing operational efficiency, higher accuracy speed, and consistency. We have used a comprehensive image dataset and leveraged a novel texture-based feature extraction algorithm. To effectively identify the repetitive textures, we went with our hybrid approach that utilizes autocorrelation for fast feature extraction. By applying this method, YOLOv8 enhances its accuracy and efficiency and optimizes real-time defect detection with minimal latency. Applying those techniques, the models that are robust to variations in fabric orientation can be ensured, along with lighting conditions and dimensionality reduction via Principal component analysis (PCA). We have used a dataset consisting of 2697 images. The methods we have used are quite impressive as they bring an accuracy rate of 85.77%. The YOLOv8 model exhibits strong performance metrics with an F1 score of 83.45% which indicates an effective balance between precision and recall. It achieves a recall rate of 84.32% while maintaining a precision of 82.88% which reflects a low rate of false positives.