Anode Effect Detection in Rare Earth Electrolysis Utilizing Efficient Channel Attention and Deformable Convolutional Networks

Addressing the challenge of monitoring anode effect during the electrolytic refining of rare earths, the present study proposes an improved YOLOv5 detection algorithm. The algorithm initially integrates ECA (Efficient Channel Attention) into the YOLOv5 model to assign higher weights to the molten salt area. It then incorporates DCN (Deformable Convolutional Network) to adapt to the irregular shapes of sparks during anode effects. Finally, the SIoU loss function is employed to accelerate the convergence of the predicted bounding boxes and enhance the regression accuracy. The improved YOLOv5 algorithm demonstrated mAP (mean Average Precision) and P (Precision) of 0.842 and 0.827, respectively, on the dataset, representing an improvement of 3.4% and 8.8% over the YOLOv5 model. The experiment has demonstrated the effectiveness and superiority of the improved YOLOv5 algorithm in monitoring the anode effect in the rare earth electrolysis process, providing valuable reference data for the smelting and purification of these elements.