More Efficient Text Detection for Chips Based On Loss Weighting Module

Text detection has always been an important branch in the field of deep learning, with applications in tasks such as character detection and recognition for industrial chips. Most deep learning-based text detection algorithms require a large amount of training data, and their detection accuracy remains relatively low for images with complex backgrounds. To address these challenges, we propose a loss-weighting module based on a text detection network to improve the detection accuracy of the network. We name this enhanced network "LWNet" which focuses on handling complex, small-sample chip images and effectively enhances character detection precision. Experimental validation on a real-world chip image dataset shows that the proposed method achieves a detection accuracy of 90.4% for chip character detection tasks. Compared to other character detection methods, LWNet demonstrates significant advantages in both accuracy and robustness, contributing to the automation and intelligent advancement of chip inspection processes.