LitchiLeaf4001: A Comprehensive Dataset of Lychee Leaf Diseases for AI-Based Visual Diagnosis

LitchiLeaf4001 is a curated image dataset focused exclusively on lychee (Litchi chinensis) leaves. It was collected to support research and development in computer vision, machine learning, and deep learning for plant disease detection. This dataset is designed to empower the agricultural AI community, particularly in Bangladesh, where lychee is a commercially important fruit crop. The dataset consists of 4,001 images captured from lychee orchards in three agriculturally diverse districts: Dhaka, Manikganj, and Gaibandha, between December 2024 and April 2025. It includes both healthy and diseased leaves affected by common visual conditions found in lychee plants. Class Distribution: 1. Anthrax – 620 images 2. Curly Leaf – 368 images 3. Dried Leaf – 555 images 4. Healthy Leaf – 662 images 5. Insect Hole – 1,157 images 6. Yellow Mosaic Virus – 639 images. Location: 1. Dhaka: [Latitude (°N): 23.8103, Longitude (°E): 90.4125] 2. Manikganj: [Latitude (°N): 23.8617, Longitude (°E): 89.9333] 3. Gaibandha: [Latitude (°N): 25.3287, Longitude (°E): 89.5284] Potential Applications: 1. Computer Vision: - Disease region detection and segmentation - Leaf health visual feature extraction - Real-time visual monitoring of lychee trees 2. Machine Learning: - Multiclass classification of lychee leaf diseases - Development of explainable AI (XAI) models for plant health assessment - Decision support tools for agricultural advisors 3. Deep Learning: - CNN-based disease recognition (e.g., InceptionV3, ResNet, MobileNet) - Attention-based DL models for fine-grained disease spotting - Integration with GANs for synthetic data augmentation 4. Smart Agriculture / AgriTech: - Mobile-based plant disease diagnostic apps - IoT-integrated crop monitoring systems - Early warning systems for lychee disease outbreaksLitchiLeaf4001是一款专为荔枝（Litchi chinensis）叶片打造的精选图像数据集。本数据集的采集旨在支撑植物病害检测方向的计算机视觉、机器学习与深度学习相关研究与开发工作，旨在赋能农业AI社区，尤其是在荔枝作为重要经济果树的孟加拉国。 该数据集共包含4001张图像，采集自孟加拉国达卡（Dhaka）、马尼格甘杰（Manikganj）与盖班达（Gaibandha）三个农业多样性各异的区县的荔枝果园，采集时段为2024年12月至2025年4月。数据集涵盖健康叶片与受荔枝常见病害侵染的病叶样本。 类别分布： 1. 炭疽病叶 — 620张 2. 卷叶病叶 — 368张 3. 枯叶病叶 — 555张 4. 健康叶片 — 662张 5. 虫孔叶 — 1157张 6. 黄色花叶病毒病叶 — 639张。 采集地点： 1. 达卡（Dhaka）：[北纬：23.8103°，东经：90.4125°] 2. 马尼格甘杰（Manikganj）：[北纬：23.8617°，东经：89.9333°] 3. 盖班达（Gaibandha）：[北纬：25.3287°，东经：89.5284°] 潜在应用场景： 1. 计算机视觉领域： - 病害区域检测与语义分割 - 叶片健康视觉特征提取 - 荔枝树实时视觉监测系统 2. 机器学习领域： - 荔枝叶片病害多分类任务 - 面向植物健康评估的可解释AI（Explainable AI，XAI）模型开发 - 面向农业咨询人员的决策支持工具 3. 深度学习领域： - 基于卷积神经网络（Convolutional Neural Network，CNN）的病害识别研究（如InceptionV3、ResNet、MobileNet） - 基于注意力机制的深度学习细粒度病害识别模型 - 结合生成对抗网络（Generative Adversarial Networks，GANs）的合成数据增强技术 4. 智慧农业/农业科技领域： - 移动端植物病害诊断应用程序 - 物联网（Internet of Things，IoT）集成式作物监测系统 - 荔枝病害爆发早期预警系统