Rail Fastener Damage Detection Standard Reference Dataset

The Rail Fastener Damage Detection Standard Reference Dataset (RFDD-SRD) is a curated benchmark dataset designed to support the evaluation of automated rail fastener inspection systems. The dataset consists of 100 high-quality images collected under a unified inspection perspective, where each image contains six fasteners, including one damaged instance and five normal instances. To address the scarcity of real damaged samples, the dataset is constructed using a controlled synthetic generation strategy, ensuring comprehensive coverage of five representative damage types: Deformed, Displaced, Fractured, Inverted, and Missing. All samples are rigorously validated through a multi-dimensional consistency evaluation framework, which enforces geometric, optical, boundary, noise, textural, and occlusion-level realism. Each image is accompanied by pixel-accurate annotations and standardized bounding box labels, following a consistent category definition. Annotations are manually refined and verified through a multi-annotator cross-validation process to ensure high spatial precision and semantic clarity. The dataset is organized in a lightweight and widely compatible format, including PNG images, text-based annotations, and structured metadata. RFDD-SRD is intended for use as a standard reference benchmark, enabling objective and reproducible evaluation of detection algorithms and imaging systems. It can also support research on model robustness, failure analysis, and cross-domain generalization in safety-critical inspection tasks.