Lightweight Transformer-Based Bird Nest Detection on Power Transmission Lines

To enhance the accuracy and speed of bird nest detection on power transmission lines, we present a lightweight transformer-based detection model. Our model features an improved EfficientViT backbone, which reduces the computational complexity and the number of model parameters. Additionally, we introduce a novel Cascade Detail Infusion (CDI) module that effectively fuses multi-scale features, thereby improving detection precision in complex environments. The EMASlideVarifocalLoss function further optimizes the model’s performance by dynamically adjusting IoU thresholds. Experimental results indicate that our model achieves a mean Average Precision (mAP) of $\mathbf{9 1 . 5 \%}$, a detection speed of 43 FPS, and a computational cost of 8.4 GFLOPs. When compared to baseline models such as RT-DETR, Faster-RCNN, and Swin Transformer, our model demonstrates an accuracy improvement of up to $10 \%$ while reducing computational costs by $\mathbf{6 9 - 9 2 \%}$. These results confirm the robustness of our model and its potential for real-time detection applications.