Introduction: The accuracy and real-time capability of hazard detection in transmission corridors directly impact grid safety, fault prevention, and emergency response efficiency. Traditional manual inspection is inefficient, and conventional image processing methods are easily disturbed in complex scenarios. To address this, this study employs the Task-aligned One-stage Object Detection (TOOD) algorithm to automatically identify typical external hazards in UAV aerial images.

Methods: We focus on four common hazards: balloons, kites, bird nests, and damaged insulators. A publicly available dataset is used to train the TOOD model, which leverages its task-aligned mechanism and feature fusion architecture to improve detection stability under strong sunlight, haze, and occlusion. Standard data augmentation is applied during training, and model generalization is evaluated on a validation set across various operating conditions. The output includes hazard location, category, and confidence score for risk analysis.

Results: Experiments show that TOOD outperforms most mainstream one-stage detectors in this task. It achieves stable performance in complex environments and basically meets the real-time requirements of field inspection.

Conclusions: Applying TOOD to transmission corridor hazard detection demonstrates reliable recognition capability under challenging conditions and possesses practical engineering value. It provides a feasible technical solution for intelligent UAV-based inspection and supports the automation of transmission line operation and maintenance.