RoseLeafInsight: A high-resolution image dataset for rose leaf disease recognition

The Rose (genus Rosa) has become a significant factor in the Bangladeshi flower industry, both in terms of exports and local consumption. However, rose farming in this country faces serious challenges due to diseases affecting its leaves, which weaken the plants and result in lower flower yields and financial losses for farmers. Rosa (genus Rosa) is one of the most attractive and commercially valuable flower genera. However, agricultural rose production faces several challenges, such as pesticide resistance, which affects plant growth and results in a reduced quantity and quality of healthy flowers. Several natural factors also cause interference with rose production. Most farmers involved in this industry have limited education, which hinders their ability to identify early-stage rose-leaf disease solely through visual inspection. Furthermore, limited communication with agricultural experts exacerbates the situation, leading to delayed interventions and economic losses. This study presents the rose leaf disease dataset, which would help enhance disease tracking, diagnosis, and research in roses. From October 2024 to January 2025, large-scale field surveys were conducted to capture quality images for each condition class in rose leaves. In this paper, four classes comprise 'Black Spot,' 'Insect Hole,' 'Yellow Mosaic Virus,' and 'Healthy,' representing different stages in disease progression. There are 3,228 original images, categorized as follows: Black Spot (409), Insect Hole (453), Yellow Mosaic Virus (680), and Healthy (1,686). During the pre-processing stage, the images are resized to 3000×3000 pixels, and low-quality, duplicate, or irrelevant images are removed to ensure high quality. We have employed various augmentation techniques, including rotation, flipping, contrast adjustment, blurring, shearing, zooming, and noise addition, to increase the dataset size and enhance model generalization. Datasets like this one are in high demand for agricultural research, leading to improved disease management and increased yields. These goals can be achieved through high-accuracy machine-learning models for early disease detection and cause identification. This gives the farmers more time to take necessary actions for disease prevention and pest control. This tech-based system combines the field of agriculture with the cutting edge of computer science and AI, making precision agriculture even more effective and efficient. Our dataset is designed to meet the need for data to train these models and provide a baseline benchmark for disease detection in our specific crop, the Rose. Improvements in different generations of models, as well as numerous other forms of scientific advancements, can lead to further increases in efficiency and ultimately result in better, smarter farms. In our initial testing for categorizing rose leaves, we employed two well-known transfer learning models. Among them, MobileNetV2 performed exceptionally well, achieving an accuracy of 96.79% in image classification. This dataset can be integrated with innovative farming equipment, such as drones and sensors, to monitor large fields in real-time. This dataset serves as a benchmark for training deep learning models, enabling enhanced automated monitoring and decision-making in precision agriculture.