Ash Gourd Leaf Condition Dataset (AGLCD-2025): A High-Resolution Benchmark for Multi-Class Leaf Disease Classification

The Ash Gourd Leaf Condition Dataset (AGLCD-2025) consists of 3,295 high-resolution images of Ash Gourd (Benincasa hispida) leaves, collected in 2025 from agricultural fields in Bangladesh (GPS: 23.8145°N, 90.4125°E; altitude: 26.4 meters). The original images were captured using a Xiaomi M2102J20SG smartphone under natural daylight conditions, retaining full EXIF metadata such as F-stop (f/1.8), ISO speed (ISO-100), exposure time (1/125 sec), focal length (5 mm), metering mode (center-weighted average), and white balance set to auto. The raw images are in JPG format, 24-bit sRGB color space, and 72 dpi resolution.

To prepare the dataset for deep learning and image classification tasks, all images were preprocessed through multiple stages:

Background removal to isolate leaf structures and enhance clarity.

Enhanced image folder containing contrast-adjusted, noise-reduced images for improved feature extraction.

Augmented image folder with balanced 1,000 images per class created via data augmentation techniques (rotation, scaling, flipping, etc.) to address class imbalance.

The dataset covers 7 different Ash Gourd leaf conditions, representing both healthy and diseased states of the plant:

Downy Mildew (456 images)

Dried Leaf (470 images)

Healthy Leaf (547 images)

Leaf Blight (452 images)

Soft Rot (466 images)

Yellow Hallow (424 images)

Yellowing of Leaf (480 images)

This AGLCD-2025 dataset is built to support diverse applications such as leaf disease classification using CNNs and transformer-based models, mobile app development for plant disease diagnosis, explainable AI (XAI) research, and smart farming solutions. It is particularly valuable for fields like Agrotechnology, where it can aid in developing advanced crop management strategies, improving plant health monitoring, and optimizing sustainable agricultural practices. The goal of this dataset is to accelerate progress in precision agriculture, agro-tech innovation, and AI-driven plant pathology.

冬瓜叶片状态数据集（AGLCD-2025）共计包含3295张高分辨率冬瓜（Benincasa hispida）叶片图像，于2025年采集自孟加拉国的农田（GPS坐标：23.8145°N，90.4125°E；海拔：26.4米）。原始图像由小米M2102J20SG智能手机在自然日光环境下拍摄，保留了完整的EXIF元数据，包括光圈（f/1.8）、ISO感光度（ISO-100）、曝光时长（1/125秒）、焦距（5毫米）、测光模式（中央重点平均测光）以及自动白平衡设置。原始图像格式为JPG，采用24位sRGB色彩空间，分辨率为72 dpi。

为适配深度学习与图像分类任务，所有图像均经过多阶段预处理：
1. 背景移除：分离叶片结构以提升画面清晰度；
2. 增强图像文件夹：收录经对比度调整、降噪处理的图像，用于优化特征提取效果；
3. 均衡增强数据集文件夹：通过旋转、缩放、翻转等数据增强技术，生成每类各1000张的均衡数据集，以解决类别不平衡问题。

本数据集涵盖7种不同的冬瓜叶片状态，覆盖植物健康与染病两类场景：
- 霜霉病（Downy Mildew）：456张
- 叶片干枯（Dried Leaf）：470张
- 健康叶片（Healthy Leaf）：547张
- 叶枯病（Leaf Blight）：452张
- 软腐病（Soft Rot）：466张
- 黄化空洞（Yellow Hallow）：424张
- 叶片黄化（Yellowing of Leaf）：480张

AGLCD-2025数据集可支撑多类应用场景，例如基于卷积神经网络（Convolutional Neural Network，CNN）与Transformer模型的叶片病害分类、植物病害诊断移动应用开发、可解释AI（Explainable AI，XAI）研究以及智慧农业解决方案。其在农业技术领域具有较高应用价值，可助力开发先进的作物管理策略、优化植物健康监测流程，并完善可持续农业实践。本数据集旨在推动精准农业、农业技术创新以及AI驱动的植物病理学领域的研究进展。