Research on Construction and Data Augmentation of Hidden Danger Target Dataset for Transmission Corridors

The detection of hidden danger targets in transmission corridors is crucial for ensuring the safe operation of the power grid. However, the scarcity of samples under special weather conditions has severely restricted the training effect of deep learning models. This study focuses on constructing a dataset of hidden danger targets in transmission corridors and proposes a special weather sample augmentation method based on the pix2pix algorithm. By collecting 1,727 original monitoring images and generating 267 rainy and 267 hazy images using pix2pix, a multi-weather dataset containing 2,258 images is constructed. Using the Labelme tool for detailed annotation, a standardized dataset in COCO and YOLO formats is formed, and a hybrid data augmentation strategy is introduced to enhance data diversity. Experimental results show that the special weather images generated by pix2pix are highly realistic, and the augmented dataset significantly improves the generalization ability of target detection models. The research provides high-quality data support for hidden danger target detection in transmission corridors and is of great significance for the practical application of intelligent monitoring systems.