A novel detector for the tilt inspection of power transmission towers based on principal component analysis

The inspection of power transmission towers, crucial for grid maintenance, has seen advances through unmanned aerial vehicles (UAVs) and neural networks. However, complex backgrounds, inaccurate semantic segmentation, and challenges in tilt calculation limit their practical use. This study introduces a non-contact tilt detection and quantification method based on an improved U-Net architecture. The model integrates Squeeze-Excitation (SE) and Convolutional Block Attention Module (CBAM) to enhance channel and spatial feature representation, addressing issues such as blurred edges and structural loss. Using the segmented binary mask, contour points are extracted, and Principal Component Analysis (PCA) determines the tower’s main orientation to compute its tilt relative to the vertical axis. Experimental results show that the proposed U-Net+SENet outperforms the baseline, increasing mean intersection over union (mIoU) and Dice coefficients by 1.08% and 4.57%, respectively. In complex cross-arm scenarios, exceeding those of both the baseline and CBAM-enhanced models. This approach improves segmentation accuracy and enables precise axis extraction and tilt estimation. Visualisation confirms the model’s superior detail extraction, verifying its effectiveness and reliability.