RGB Image Dataset of Artificial Plants for Deep Learning Applications

The RGB image dataset of miniature artificial plants designed specifically for machine learning and deep learning applications in agricultural image analysis. The plant bases were fabricated using Plaster of Paris, shaped from vegetable trays to create a stable and uniform support structure. Artificial decorative plants made of plastic were inserted into these bases to simulate real plant structures under controlled conditions. This approach enabled the creation of a low-cost, reproducible, and highly variable dataset suitable for algorithm development and testing.
The dataset consists of three distinct plant types, differing in leaf morphology, size, color distribution, and structural arrangement. These variations allow the dataset to be flexibly used for multiple classification scenarios, such as three-class plant classification, binary classification (plant vs. weed), or multi-label categorization (two plant types vs. weeds). The dataset includes a total of 1,440 images, comprising 203 images of type-1, 843 images of type-2, and 394 images of type-3 plants.
All images were captured using a 16-megapixel RGB camera (Canon PowerShot SX170) under artificial lighting conditions, ensuring realistic illumination variability. Images were acquired from multiple angles and orientations to introduce geometric diversity and improve model generalization. The original image resolution was 1632 × 1553 pixels, preserving fine details such as leaf edges, texture, and spatial arrangement.
To facilitate efficient training of deep learning models, particularly convolutional neural networks (CNNs), all images were resized to 224 × 224 pixels. This standardization reduces computational complexity while retaining essential visual features. The dataset incorporates variations in lighting, viewpoint, and plant structure, making it suitable for robust model training even with limited data.
Overall, this dataset provides a controlled yet diverse platform for evaluating image processing, classification, and object detection algorithms in agricultural and plant phenotyping research.
The authors gratefully acknowledge financial support from ICAR–CIAE under the CRP-FMPF project.