A Leaf-Level Dataset for Soybean–Cotton Detection and Segmentation

Overview
This repository provides a curated image dataset designed for leaf-level detection and segmentation of soybean (Glycine max) and cotton (Gossypium hirsutum) plants under real-world field conditions. Each image is annotated with both bounding boxes and instance segmentation masks, suitable for a broad range of computer vision tasks including volunteer plant detection, weed discrimination, and early disease surveillance in precision agriculture.

Key Features

• Real Field Images: All 640 images were captured in a commercial farm setting in São Paulo, Brazil, across diverse growth stages and lighting conditions.
• Dual Annotations: Both bounding boxes (for object detection) and instance-level masks (for pixel-level segmentation) are provided.
• Multiple Growth Stages: Includes early, mid, and dense canopy phases, ensuring robustness to occlusions and overlapping foliage.
• High-Quality Labels: Rigorous annotation process with human experts and post-processing.
• Soybean & Cotton Leaves: Over 12,000 annotated leaves in total—7,221 soybean leaves and 5,190 cotton leaves.
• Various Use Cases: Ideal for tasks such as selective herbicide application, volunteer crop monitoring, phenotyping, and canopy analysis.

Data Structure

• Images: 640 high-resolution RGB images (1600×1200 pixels) in PNG format.
• Annotations:
• • Annotations for both detection bounding boxes and segmentation masks are in COCO format.
  • Bounding boxes are in absolute pixel coordinates.
  • Instance segmentation masks are in Run-length encoding (RLE) format.

Suggested Applications

1. Leaf-Level Detection and Segmentation: Train and evaluate object detectors to localize individual cotton or soybean leaves.
2. Precision Agriculture Analytics: Use detections to quantify volunteer plants, gauge canopy overlap, or assess leaf area index over time.
3. Weed Management: Extend to weed–crop discrimination.