A Dataset for Railway Catenary Foreign Object Detection in Complex Environments

It was constructed from real railway catenary background images through structurally constrained object insertion and constrained image-to-image environmental editing. The dataset was designed to preserve physically plausible spatial and attachment relationships between foreign objects and catenary components, including contact wires, droppers, and supporting structures, while providing controlled environmental diversity.The complete dataset contains 51,414 annotated images covering eight foreign-object categories: kite, banner, bird nest, balloon, tarp, branch, plastic bag, and clothing.

The main dataset comprises 48,736 images under four environmental conditions—sunny, rainy, foggy, and nighttime—and is divided into 38,976 training images, 4,864 validation images, and 4,896 internal test images. An additional external test subset contains 2,678 images constructed from independent background images and is intended for evaluating cross-background generalization.CatenaryFOD-8 provides predefined source-grouped data splits, object-detection annotations in both YOLO and COCO formats, class definition files, dataset configuration files, and non-sensitive image-level metadata.