An end-to-end Lithium Battery Defect Detection Method Based on Detection Transformer

The DETR model is often affected by noise information such as complex backgrounds in the application of defect detection tasks, resulting in detection of some targets is ignored. In this paper, AIA DETR model is proposed by adding AIA (attention in attention) module into transformer encoder part, which makes the model pay more attention to correct defect information. Rather than the noise information on the image, so as to improve the detection ability of lithium battery surface defects. Experiments show that AIA DETR model can well detect the defect target of lithium battery, effectively reduce the missed detection problem, and reach 81.9% AP in the lithium battery defect data set