Discharge Severity Assessment of Porcelain Suspension Insulators Based on UV Imaging and DETR

In electrical systems, suspended porcelain insulators are widely used for their reliability. However, their insulating performance declines under prolonged exposure to electrical, thermal, and stress conditions. This decline threatens the safety and stability of power systems. To improve the accuracy of discharge severity assessment in complex environments, an improved DETR (DEtection TRansformer) algorithm was developed. This algorithm enables precise identification of discharge severity levels.Mosaic-4 data augmentation was employed to enhance the model's generalization capabilities. By combining multiple images, this technique increased the diversity of input data, improving the model's adaptability to various scenarios and backgrounds. The improved DETR algorithm was integrated with Iterative MultiScale Feature Aggregation (IMFA) technology. IMFA facilitated the recognition and classification of multi-scale features by sparsely sampling features at key locations and using previous detection predictions as guidance. This approach maintained computational efficiency while significantly improving the detection of discharge spots of varying sizes, especially in complex backgrounds. The ReLU activation function was replaced with GELU (Gaussian Error Linear Unit) to accelerate model convergence and enhance detection accuracy. GELU's superior non-linear characteristics allowed for more effective gradient transmission during training, thereby improving the overall training efficiency of the model.Experimental validation showed that the improved DETR algorithm outperformed the original, achieving an average accuracy of 85.1 %, a 5.2 % increase from the original DETR. The proposed method demonstrated significant advantages in assessing discharge severity, particularly in dealing with small targets and complex backgrounds. This enhancement improves the accuracy and reliability of discharge assessment in power systems.