An Image Dataset of Seven Major Pulses Grown in Bangladesh for Smart Agriculture Applications
Description
This dataset is a novel, comprehensive image collection featuring seven major pulse crop species cultivated in Bangladesh: Lentil (Masur), Mung bean (Mug), Grass pea (Khesari), Chickpea (Chola), Black gram (Māshkalāi), Field pea (Motor), and Kidney bean (Rājmā). The primary purpose of this dataset is to facilitate research in Machine Learning (ML) and Deep Learning (DL) for agricultural technology, specifically for applications in smart agriculture, such as automated species identification, yield forecasting, and plant health monitoring. The collection comprises approximately 17,500 images across 11 distinct classes covering various species, varieties, and their associated disease/health states, all captured under natural field conditions in Bangladesh. This focus on locally grown varieties makes the dataset highly relevant for addressing regional agricultural challenges, unlike many global datasets that may not feature these specific landraces. The data is pre-processed and ready for direct use in image classification and object detection tasks.
Files
Steps to reproduce
The data was systematically collected from various pulse-cultivation fields across different regions of Bangladesh during the 2023-2024 growing season. The collection methodology was designed to ensure high variability and real-world applicability for computer vision models. Image Acquisition: Images were captured using high-resolution digital cameras (e.g., smartphone cameras or DSLRs) under natural daylight conditions. Sampling Strategy: Sampling was performed across different growth stages of the plants (vegetative, flowering, and pod formation) to capture phenotypical variation. Images were taken at various angles and distances (top-down, close-up, and oblique views) to simulate real-world surveillance scenarios. Classes and Labeling: The images were manually classified and labeled by agricultural experts into 11 distinct classes. These classes represent the seven pulse species and specific health conditions (e.g., Healthy Mung Bean, Lentil with Rust, etc.). Data Processing: All collected images were standardized to a uniform size (e.g., 224x224 or 256x256 pixels, depending on the final published version) and saved in JPG format. No further background removal or advanced image manipulation was performed to maintain the integrity of the natural field environment, preserving the presence of soil, shadows, and background leaves which are crucial for model robustness.
Institutions
- Southeast University