Robust and high-precision defect detection approach for small-diameter optical lenses using light projection imaging and dual attention networks

Small-diameter optical lenses are extensively utilized in consumer electronics, medical imaging devices, and advanced driver-assistance systems, where high manufacturing precision and yield are of critical importance. However, current defect detection techniques frequently encounter limitations in terms of either insufficient accuracy or low operational efficiency, particularly when addressing common defects such as bubbles, scratches, and edge chipping. To address these challenges, we propose an innovative inspection system that combines a light projection technique for defect imaging with a dual attention network—integrating both spatial and channel attention mechanisms—for robust and efficient defect segmentation. In contrast to the conventional back-illumination method, the proposed light projection approach substantially reduces the adverse effects of nonuniform illumination, thereby improving image quality and enhancing detection accuracy for small-diameter lenses. Experimental results indicate that the proposed dual attention network surpasses the traditional U-Net model, achieving an approximate pixel accuracy improvement of 90.5%. Furthermore, the optimized network achieves a detection speed of over 10 frames per second, making it highly suitable for high-throughput industrial inspection environments. This approach not only provides an effective solution for optical lens quality control but also demonstrates strong adaptability to larger-diameter lens systems, highlighting its broad applicability and potential for widespread industrial adoption.