ARTEN-Enhanced Multi-Crop Disease Dataset
Description
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.
Files
Steps to reproduce
The enhanced dataset was compiled from the previously created multi-crop disease image analysis of healthy and infected leaf samples under genuine agricultural field circumstances. The raw images were first classified into crop and disease classes. Then, synthetic degradation processes were used to produce the training pairs necessary for learning the enhancement model. Image quality evaluation was conducted using Laplacian variance-based clarity estimate approach to determine improvement goals. Then, the Attention-guided Residual Texture Enhancement Network (ARTEN) was trained on the paired degraded and reference images for 100 epochs. Following conventional model optimisation, the trained ARTEN framework was used on the complete source dataset to provide improved outcomes. The augmentation process preserved the original spatial resolution of 640×640 pixels, while increasing the image quality while maintaining the disease-specific visual information. The completed dataset was then saved with the same class labels and folder structure as the original dataset thus ensuring repeatability and compatibility with established research methodologies.