The SD-R3det-Based Transmission Tower Tilt Detection Model

Power transmission towers, due to their unique structure, often exhibit a high aspect ratio and are prone to missed detections of small targets during object detection tasks. To address this challenge, this paper proposes an enhanced transmission tower tilt detection model based on an improved R3Det algorithm, named SD-R3Det. The model incorporates a Squeeze-and-Excitation (SE) attention mechanism prior to the feature fusion stage to more effectively emphasize critical features. Furthermore, depthwise separable convolutions are adopted at the output of the Feature Pyramid Network (FPN) to reduce computational cost and enhance detection accuracy. To evaluate the performance of the proposed model, a dedicated detection dataset for transmission towers was constructed. Experimental results demonstrate that the improved model outperforms the original R3Det in complex scenarios, achieving a mAP@ 0.5 of 0.903 and a [email protected]:0.95 of 0.699, while also increasing the FPS by 3.4. This work significantly improves the accuracy of transmission tower tilt detection and provides strong technical support for intelligent monitoring of transmission infrastructure.