Lychee Pest Damage images

Core samples were collected from Gaozhou City (21.78°N, 110.99°E), Maoming, Guangdong Province, China – a key lychee production region, capturing representative specimens during the peak infestation period of 2024. All fruits underwent standardized agricultural processing and were imaged at the Maonan District laboratory (21.63°N, 110.89°E) using four mainstream mobile phone sensors:iPhone 12 (Apple Custom sensor, 12MP@4032×3024, Deep Fusion optimization)Honor 50 (Samsung HM2 sensor, 108MP@12032×9024, AI multi-frame fusion)Honor X50 (Samsung HM6 sensor, 108MP@12000×9000, multi-frame noise reduction)realme GT Neo (Sony IMX682 sensor, 64MP@9280×6944, AI scene detection)The original 3,061 high-resolution images feature two specialized annotation types:Only Wormholes: Precise labeling of 0.3-1.2mm diameter borehole morphology for pixel-level pest segmentation research;Fruit Peel+Wormholes: Preservation of pathological spatial relationships between boreholes and fruit peel tissue to support multi-scale object detection.Through a Python-based illumination robustness enhancement model, 6,122 simulated frontlight/backlight images were generated (illuminance variation ΔLux 500-2000), resulting in a final enhanced dataset of 9,183 images.

The dataset's triple scientific value lies in:Geo-pathological traceability: Integration with Gaozhou climate parameters (June 2024: 83% avg. humidity, 31°C avg. high, 25°C avg. low) provides environmental variables for pest prediction models;Pathological feature decoupling: Dual annotation strategies separate core borehole features from pathological peel backgrounds, enhancing interpretability in agricultural AI research.