USU-Corn-WeedDB: A UAV RGB Image Dataset for Multi-Species Weed Detection in Forage Corn

2026-06-04 · Source: Computer Vision and Pattern Recognition · Field: Agriculture & Food Systems — Precision Agriculture & Smart Farming, Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Advanced, quick

Summary

USU-Corn-WeedDB is a new publicly available UAV RGB image dataset designed for multi-species weed detection in forage corn, addressing the scarcity of field-representative training data for site-specific weed management. Collected on 27 June 2025 from a commercial field in Cache Valley, Utah, using an Autel EVO II Dual 640T V2 drone at ~10m AGL, it features a 0.48 cm/pixel ground sampling distance. The dataset comprises 366 full-resolution images, tiled into 8,800 patches of 640x640 pixels. Of these, 800 images are manually annotated with 10,539 bounding-box instances across three weed species: common lambsquarters, redroot pigweed (53.86% of instances), and green foxtail. The remaining 8,000 tiles form an unlabeled pool for semi-supervised learning. Validation with 28 object detection models, including YOLOv8, YOLOv9, YOLOv10, YOLO11, YOLO26, and RT-DETR, yielded mAP@0.5 scores from 0.773 to 0.840, demonstrating competitive performance even with lightweight architectures suitable for edge-deployed UAV systems.

Key takeaway

For Machine Learning Engineers developing precision agriculture solutions, USU-Corn-WeedDB offers a critical resource for advancing site-specific weed management. You should integrate this field-representative dataset to train and validate multi-species weed detection models, especially exploring semi-supervised learning with its unlabeled image pool. This can improve model robustness and deployment efficiency, particularly when evaluating lightweight architectures for edge-deployed UAV systems in real-world agricultural settings.

Key insights

A new UAV RGB dataset enables multi-species weed detection in forage corn, supporting supervised and semi-supervised deep learning.

Principles

• Field-representative datasets are critical for effective site-specific weed management.
• Preserving natural class imbalance mirrors real-world agricultural conditions.
• Lightweight models can achieve competitive performance for UAV edge deployment.

Method

Acquire UAV RGB imagery at ~10m AGL, tile into 640x640 patches, manually annotate weed species with bounding boxes, and reserve unlabeled data for semi-supervised training.

In practice

• Train multi-class weed detection models using the USU-Corn-WeedDB dataset.
• Utilize the unlabeled image pool for semi-supervised learning experiments.
• Benchmark lightweight object detection models for UAV edge deployment.

Topics

• UAV Imagery
• Weed Detection
• Forage Corn
• Deep Learning
• Object Detection
• Semi-Supervised Learning
• Precision Agriculture

Best for: Computer Vision Engineer, AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• USU-Corn-WeedDB: A UAV RGB Image Dataset for Multi-Species Weed Detection in Forage Corn — A new UAV RGB image dataset addresses data scarcity for multi-species weed detection in forage corn.
• Carbon Robotics built an AI model that detects and identifies plants — A new AI model enables instant, real-time weed recognition for autonomous agricultural robots, eliminating retraining.
• SWNet: A Cross-Spectral Network for Camouflaged Weed Detection — Cross-spectral fusion and edge refinement are crucial for accurate camouflaged weed detection in agriculture.
• Tomato Multi-Angle Multi-Pose Dataset for Fine-Grained Phenotyping — TomatoMAP dataset enables accurate, reproducible fine-grained tomato phenotyping using multi-angle imagery and AI models.
• An Open Dataset for the Acoustic Monitoring of Nocturnal Migratory Birds in Europe — Crowd-sourced acoustic data and a specialized deep learning model enhance nocturnal migratory bird monitoring.
• Contrasting Local and Global Modeling with Machine Learning and Satellite Data: A Case Study Estimating Tree Canopy Height in African Savannas — Local SatML models can outperform global or fine-tuned global models for specific regional tasks.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.