Research on Pointer Defect Detection Based on Deep Learning

Aiming at the characteristics of multiple types, small size and high detection difficulty of the stamping pointer defect of automobile dashboard, a pointer defect detection algorithm with SSD(single shot-multiboxp detector) and RFCN(region-based fully convolutional networks) was designed. First, Basler camera was used on site to collect photos of defective pointers and annotate them, and mobileNet network was used as the feature extraction network of SSD algorithm to classify stamping and no stamping pointers. Then, the image of stamping pointer was clipped, and the defect of clipped pointer image was marked. Finally, the feature extraction network of RFCN was improved based on ResNet network to detect four typical defects. The software system was built under the Ubuntu16.04 system and Tensorflow-gpu deep learning framework for experimental testing. The results show that the algorithm can accurately detect the tiny defects on the pointer surface and effectively improve product quality and production efficiency.