Heritage Collection Pest-Related Arthropod Image Dataset

This dataset provides a de-identified benchmark for visual recognition of pest-related arthropods in cultural-heritage collection storage and laboratory monitoring settings. It contains 900 full-scene images for object detection, including 794 annotated images with 2,195 bounding-box instances and 106 pest-free negative scenes, together with 800 single-object images for taxonomic-category classification across ten operational categories and 778 anonymous specimen groups. The package includes JPEG images, YOLO-format annotations, image-level metadata, class definitions, detection partitions, group-safe classification random splits, five-fold cross-validation manifests, per-run and aggregate result tables, and SHA-256 checksums. The classification and detection tasks were evaluated separately. Original filenames, acquisition dates, collection-location codes, source paths, EXIF/GPS metadata, personal identifiers, and private source-to-anonymous crosswalks are excluded. Trained model checkpoints and complete intermediate prediction outputs are not included. The dataset is intended for research on computer vision, arthropod recognition, object detection, image classification, and preventive conservation. Model outputs should be reviewed by qualified conservation or pest-management personnel before any intervention.