2026-10-12T02:16:06Z https://www.scidb.cn/oai

10.57760/sciencedb.44706 2026-07-31T16:19:32Z

2026-07-31 LitchiPestDisease: A Multi-Illumination Public Dataset for Visual Detection of Litchi Fruit Borer Infestation and Litchi Anthracnose doi:10.57760/sciencedb.44706 en Litchi is an economically vital tropical fruit native to South China. Litchi fruit borer (Conopomorpha sinensis) creates tiny boreholes on fruit pericarp, rendering fruits unmarketable; litchi anthracnose (Colletotrichum gloeosporioides) produces necrotic lesions and shortens shelf life. Traditional manual sorting is costly, subjective and low-precision. Existing public agricultural datasets mostly focus on crop leaf diseases, lack dual-stress (pest + disease) fruit samples, and fail to provide standardized multi-lighting image groups for postharvest sorting scenes. This dataset fills the above research gaps.Data Collection & SamplingSampling location: Core litchi production areas of Maoming City, Guangdong, China.Sampling time: Litchi ripening seasons (May–July) of 2024 and 2025.Litchi cultivars: Pest subset only contains 'Feizixiao'; disease subset covers 'Feizixiao' and 'Baitangying'.Shooting environment: Controlled laboratory fixed uniform lighting, mainstream smartphones for multi-angle fruit photography; all samples transported to lab within 24h after picking.Raw image volume: Total 6,705 original JPG images, including 3,061 pest infestation images and 3,644 anthracnose disease images.Multi-illumination data provision: We directly attach three groups of enhanced multi-illumination images for every original image, including front lighting, side lighting and back lighting variants generated by standardized brightness & contrast adjustment parameters.Front Lighting: Brightness=1.4, Contrast=1Side Lighting: Brightness=1.2, Contrast=1.3Back Lighting: Brightness=0.6, Contrast=1.2 All multi-illumination images are released together with original images and annotations, no extra code execution required for users. The linear brightness-contrast adjustment simulates workshop light changes, but cannot reproduce complex real optical phenomena (color temperature, specular reflection, multi-source mixed shadow).Annotation System (Multi-Granularity Labels)Litchi Fruit Borer Subset (3,061 original + matched multi-illumination images): LabelImg bounding box annotation for tiny borer boreholes (average diameter <1mm). Dual annotators cross-check, IoU threshold ≥0.8 for consistency verification.Litchi Anthracnose Subset (3,644 original + matched multi-illumination images):Bounding box for whole litchi fruit; three severity grades by lesion coverage: Mild (<10%), Moderate (10%–30%), Severe (>30%). Class distribution: Mild 1004, Moderate 1310, Severe 1330.2,048 images with Labelme polygon instance segmentation for lesion pixel contours, labeled with corresponding severity.Annotation format: Standard YOLO .txt label files, fully compatible with all three groups of multi-illumination images (illumination adjustment does not change target position/size). Litchi pest and disease detection; litchi fruit borer; litchi anthracnose; multi-illumination images; object detection; instance segmentation; agricultural computer vision; postharvest fruit sorting; small object detection 徐兵 苏雪萍 马泽杰 PUBLIC https://creativecommons.org/licenses/by/4.0/ dataset Science Data Bank