Attention Guided Multi-Task Learning for Surface defect identification

Surface defect identification is an essential task in the industrial quality control process, which visual checks are conducted on a manufactured product to ensure that it meets standards. Convolutional Neural Network (CNN) based surface method has proven outperform traditional image processing techniques. However, real-world datasets limited size due expensive data generation process and rare occurrence of defects. To address this issue, paper presents for exploiting auxiliary information beyond primary labels improve generalization ability tasks. Considering correlation between pixel level segmentation masks, object bounding boxes global classification labels, we argue jointly learning features related tasks can performance This proposes framework named Defect-Aux-Net, multi-task with attention mechanisms exploit rich additional from goal simultaneously improving robustness accuracy CNN identification. We series experiments proposed framework. The experimental results showed significantly state-of-the-art models while achieving overall 97.1%, Dice score 0.926 mAP 0.762 classification, detection