Machine vision problem for fast recognition of surface defects of thermoelectric cooler components based on deep learning method

Abstract

Abstract During thermoelectric coolers (TEC) production, a complex industrial manufacturing process must be experienced, which may cause defects on the surface of the TEC component. To improve the efficiency of TEC component defect inspection, we propose a machine vision technology based on deep learning for surface defect detection. In order to make the deep learning method based on the you only look once (YOLO) model more efficient, first of all, we use a more lightweight network ResNet34 to improve the original network structure. Then, the loss function is improved to complete intersection over union (CIoU) loss. Experiments performed using the proposed model, show an obvious reduction in the number of parameters, the detection speed is as high as 6.5pcs/s, and the detection accuracy is 97.61%. This method lay a good foundation for the further application of deep learning methods in the field of industrial detection. The experimental results verify the feasibility and effectiveness of the model.