Solar Panel Defect Detection on Infrared Images Based on Improved Yolov8

To address the problems of low detection accuracy, difficulty in detecting small targets, and large model computation in solar panel infrared image defect detection methods, this paper proposes an algorithm based on YOLOv8 by introducing new structures and loss functions to improve detection effect and reduce parameters. Firstly, this study designs the C2f-EMA module, which introduces the EMA attention mechanism to improve the Bottleneck module in the C2f module, enriching the multi-scale receptive field features and enhancing the model's ability to detect small targets. Secondly, the WIOU loss function is used to further improve detection accuracy. Finally, ADown lightweight downsampling is used to replace part of the downsampling in the main network to reduce the number of model parameters and improve the size of the model. The experimental results show that the improved algorithm achieves a mean accuracy (mAP) of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{8 3. 6 \%}$</tex> compared to the original model by <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{5. 5 \%}$</tex>, but the number of model parameters is about 20 w, which is sufficient for the requirements of detection accuracy and model size.. The algorithm can detect microscopic defects on the surface of solar panels accurately while reducing the number of parameters, providing a new method for improving the detection accuracy of solar panel defects and outdoor deployment.