• An efficient object detection model based on an improved RT-DETR is proposed. • The model is used for insulator defect detection in electrical transmission lines. • A reparam window attention is added to enhance the foundational blocks of ResNet-18. • The small object detection layer can integrate features into the pyramid network. • The feature fusion process is further refined via selective boundary aggregation. In this paper, an efficient object detection method based on an improved real-time detection transformer is proposed to address the insulator defect detection in electrical transmission lines. The proposed method incorporates a reparam window attention block to enhance the foundational blocks of residual network-18. This modification significantly enlarges the receptive field, thereby enhancing feature extraction and improving representation capability. Moreover, the method processes the small object detection layer using space-to-depth convolution, effectively integrating features into the feature pyramid network. The method further refines the feature fusion process through selective boundary aggregation, improving the discriminative power of the learned features. The experimental analysis corroborates that the improved method attains elevated accuracy levels in comparative studies, evidencing its competence in insulator defect identification across diverse and complex settings.
Yao et al. (Tue,) studied this question.