基于RepED-YOLOv8n的太阳能电池缺陷检测算法OA

The existence of small object defects on the surface of solar cells and the variety of defect morphology will cause efficiency loss of the power generation system.In view of this,the thesis proposes a novel solar cell defect detection algorithm RepED-YOLOv8n based on YOLOv8n.Firstly,by deeply integrating the ESE mechanism into the SE layer of MBConv,an EMBC is innovatively constructed to replace the original C2f module,which realizes the improvement of the detection accuracy while maintaining a small number of parameters.Secondly,the YOLOv8n neck network is reconstructed by Efficient RepGFPN;by integrating CSPNet,the efficient layer aggregation network(ELAN),and re-parameterization techniques,and introducing a deformable large kernel attention(DLKA)module,the model can well adapt to capture the complex and subtle defect features on solar cell surfaces,which results in a novel architecture termed the reparameterized generalized feature pyramid network with DLKA(RepGFPN-DLKA).Finally,the WIoU v3 loss function is used to pay more attention to small object defects and eliminate uneven label distribution in the dataset.The experiments show that the mAP@0.5 of the proposed model in the experimental dataset reached 89.2%,an improvement of 2.4%over that of the baseline model.Its precision and recall rate reach 82.5%and 87.3%,respectively,which are improved by 3.1%and 3.7%.The detection performance of the proposed model is improved significantly while keeping a small number of parameters.