ARTEN-Enhanced Multi-Crop Disease Dataset

The ARTEN-improved Multi-Crop Disease Dataset comprises 21,875 improved leaf images of five primary crops: Banana, Chilli, Groundnut, Cauliflower and Radish. To construct the dataset, we used the Attention-guided Residual Texture improvement Network (ARTEN), a deep learning-based image improvement system, which aims to improve the quality of the image, lesion visibility, texture representation, and structural preservation. All images were improved without any change to the class names, folder structure and spatial resolution of 640 by 640 pixels. The dataset will be helpful for research on computer vision, plant disease diagnostics, precision agriculture and artificial intelligence-based crop monitoring system.
Research Hypotheses
The premise of this study is that deep learning-based image enhancement may increase the visual quality and diagnostic usefulness of crop disease images while keeping the disease-specific traits. Improved images may help construct more accurate machine learning and deep learning models for disease categorisation, detection, localisation and severity assessment.
What the Data Shows
The collection comprises 21,875 ARTEN enhanced images of healthy and diseased leaf samples of Banana, Chilli, Groundnut, Cauliflower and Radish crops. The enhancement technique increases the clarity, contrast, edge definition, and lesion visibility of the image while maintaining the disease's symptoms and biological components. We provide a large dataset covering a wide range of disease conditions and natural field variation that is well suited for training and assessing agricultural AI models.
Notable Features
The dataset has been constructed based on the ARTEN framework, which combines the residual learning and attention techniques to improve the image quality while preserving the disease-related information. All images are given in a common size of 640×640 pixels, but keep the original class names and structure. The dataset includes numerous crops and disease classes, which may be used for multi-crop disease study and model assessment.
How to Interpret and Use the Data
The images are organised by crop and disease class. For each source images there is a corresponding enhanced image. The dataset may be utilised for disease classification, object identification, semantic segmentation, lesion localisation, severity estimate and explainable AI investigations. It supports major frameworks as PyTorch, TensorFlow, Keras, YOLO, Faster R-CNN, Vision Transformers.
Potential Applications
Potential applications include automated crop disease diagnostics, smart farming systems, precision agriculture, agricultural robots, mobile disease detection apps, and AI-based decision-support tools. The dataset also acts as a standard resource for computer vision, image enhancement and plant pathology research.

ARTEN增强多作物病害数据集（ARTEN-improved Multi-Crop Disease Dataset）包含21,875张优化后的叶片图像，涵盖香蕉、辣椒、花生、花椰菜与萝卜5类主要作物。为构建该数据集，我们采用了注意力引导残差纹理增强网络（Attention-guided Residual Texture improvement Network，ARTEN）——一款基于深度学习的图像增强系统，旨在提升图像质量、病斑辨识度、纹理表现力与结构保留度。所有图像均经过增强处理，且未改动原有的类别名称、文件夹结构与640×640像素的空间分辨率。本数据集可用于计算机视觉、植物病害诊断、精准农业以及基于人工智能的作物监测系统等领域的研究。

研究假设
本研究的前提是，基于深度学习的图像增强技术可在保留病害特异性特征的前提下，提升作物病害图像的视觉质量与诊断实用价值。经过增强的图像有助于构建更精准的机器学习与深度学习模型，用于病害分类、检测、定位以及严重程度评估。

数据集内容说明
本数据集涵盖21,875张经ARTEN增强的图像，包含上述5类作物的健康与染病叶片样本。该增强技术可提升图像的清晰度、对比度、边缘锐度与病斑辨识度，同时保留病害的症状与生物学特征。本数据集规模庞大，涵盖多种病害场景与自然田间变异情况，适用于农业人工智能模型的训练与评估。

显著特性
本数据集基于ARTEN框架构建，该框架结合了残差学习与注意力机制，可在保留病害相关信息的同时提升图像质量。所有图像均统一为640×640像素尺寸，且保留原始类别名称与文件结构。数据集包含多种作物与病害类别，可用于多作物病害研究与模型评估。

数据解读与使用方法
图像按作物与病害类别进行组织，每张源图像均对应一张增强后的图像。本数据集可应用于病害分类、目标检测、语义分割、病斑定位、严重程度评估以及可解释人工智能（explainable AI）研究。其支持PyTorch、TensorFlow、Keras、YOLO、Faster R-CNN、视觉Transformer（Vision Transformers）等主流深度学习框架。

潜在应用场景
潜在应用包括自动化作物病害诊断、智能农业系统、精准农业、农业机器人、移动病害检测应用以及基于人工智能的决策支持工具。本数据集同时可作为计算机视觉、图像增强与植物病理学研究的标准资源。