Ground-based imagery dataset for early weed classification in tomato crops.

Area of Science:

• Agricultural Science
• Computer Vision
• Data Science

Background:

• Early weed identification is crucial for precision agriculture and sustainable crop production.
• Site-specific weed management reduces herbicide use and environmental impact.
• Developing accurate AI models requires high-quality, annotated datasets.

Purpose of the Study:

• To introduce a novel, publicly accessible dataset of annotated RGB images for weed species identification.
• To support the training and evaluation of deep learning models for early-stage weed detection.
• To advance the development of image-based monitoring systems in precision agriculture.

Main Methods:

• A curated dataset of 1217 high-resolution RGB images (5184 × 3888 pixels) was collected from commercial tomato fields.
• Images were captured during the 2021 and 2022 growing seasons in Badajoz, Spain.
• The dataset contains 21,208 manually annotated instances of weed species, divided into two subsets by year.

Main Results:

• The dataset comprises 1217 images with 21,208 labelled weed instances.
• Subset 1 (2021): 938 images, 9060 instances.
• Subset 2 (2022): 278 images, 11,931 instances.

Conclusions:

• The public release of this dataset facilitates research in AI-driven weed detection.
• This resource will accelerate the development of efficient and accurate weed classification models.
• The dataset supports advancements in sustainable precision agriculture through improved weed management strategies.