Dent defects in button cell batteries frequently arise during production and transportation, which not only impair their aesthetic appeal but also pose safety risks. The detection of these defects is particularly challenging due to the highly reflective surfaces of the cells and the interference caused by stamped characters. To tackle these issues, an automatic optical imaging system featuring dark field lighting is developed to capture time-series images. By employing shape template matching, relative position calculation, and affine transformation, the character regions were accurately located. The threshold segmentation method is then applied to both the original and Gaussian-filtered images, excluding the character regions, to identify potential defect areas. Defect pixel areas are determined using a 200-pixel threshold. Through comparative analysis, the number of time-series images is optimized to 7, significantly enhancing defect recognition accuracy. Online testing of 150,911 batteries demonstrated a 97.87% accuracy rate for normal batteries and a 99.05% detection rate for defective ones. The proposed algorithm processes each sample in under 300 ms, satisfying the requirements for real-time industrial detection. This study presents an effective solution for the real-time monitoring of dent defects in button cell batteries, contributing to improved quality control and safety assurance in the battery manufacturing industry.
扫码关注我们
求助内容:
应助结果提醒方式:
