CVRP: A rice image dataset with high-quality annotations for image segmentation and plant phenomics research

Area of Science:

• Agricultural Science
• Computer Vision
• Plant Biology

Background:

• Machine learning models are crucial for precision agriculture and plant breeding, but their development is limited by a lack of high-quality, annotated image datasets.
• Existing datasets often lack the diversity and detail required for advanced crop analysis.

Purpose of the Study:

• To introduce the Comprehensive Multicultivar and Multiview Rice Plant image dataset (CVRP).
• To provide a valuable resource for developing and evaluating machine learning models in rice research.
• To facilitate advancements in rice phenomics, breeding, and precision agriculture.

Main Methods:

• Collected images from 231 rice landraces and 50 modern cultivars under dense planting conditions in paddy fields.
• Captured both natural environment and focused indoor images of rice panicles.
• Employed a semi-automatic annotation process using deep learning, followed by manual curation.

Main Results:

• The CVRP dataset contains diverse, high-quality images with detailed annotations.
• Evaluated the performance of four state-of-the-art semantic segmentation models using the CVRP.
• Successfully demonstrated 3D plant reconstruction with organ segmentation using the dataset.

Conclusions:

• The CVRP dataset significantly enhances capabilities for image-based panicle identification and segmentation.
• It serves as a valuable resource for automated rice cultivar identification, panicle/grain counting, and 3D plant reconstruction.
• The dataset and annotation model are publicly available to foster further research.