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

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Soybean and cotton are vital crops, but volunteer plants and weeds challenge sustainable management.
• Accurate plant recognition in complex canopies is crucial for effective weed control.
• Existing deep learning datasets lack the complexity of real-world agricultural environments.

Purpose of the Study:

• To create a comprehensive, high-resolution dataset for leaf-level instance segmentation of soybean and cotton plants.
• To address the limitations of current datasets in capturing real-world agricultural field complexities.
• To facilitate the development of advanced AI-driven crop management strategies.

Main Methods:

• Collected 640 high-resolution images from a commercial farm across various growth stages, weed pressures, and lighting conditions.
• Annotated 7,221 soybean and 5,190 cotton leaves with bounding boxes and segmentation masks at the leaf-instance level.
• Validated the dataset's utility using the YOLO11 deep learning model for object detection and segmentation.

Main Results:

• The dataset successfully captures challenging scenarios like overlapping foliage, small leaf sizes, and morphological similarities.
• Validation with YOLO11 demonstrated state-of-the-art performance in identifying and segmenting overlapping foliage.
• The annotated data enables precise leaf-instance segmentation in complex agricultural settings.

Conclusions:

• The developed dataset provides a robust foundation for advancing AI in agriculture, specifically for soybean and cotton.
• Public availability of this dataset will foster research in selective herbicide application and pest monitoring.
• This resource supports the creation of more effective, data-driven strategies for sustainable soybean and cotton management.