The agricultural contamination elements (ACE) dataset: Multi class annotated images

Area of Science:

• Agricultural Science
• Computer Vision
• Environmental Monitoring

Background:

• Agricultural fields are susceptible to contamination from various man-made materials.
• Automated detection of these contaminants is crucial for efficient field management and environmental protection.
• Existing datasets may lack diversity in terms of contamination types, imaging conditions, and temporal variations.

Purpose of the Study:

• To introduce the Agricultural Contamination Elements (ACE) dataset, a novel resource for training and evaluating object detection models.
• To provide a diverse collection of annotated images captured via unmanned aerial systems (UAS) in cotton fields.
• To facilitate research in automated detection of agricultural contaminants across different growing seasons.

Main Methods:

• Images were captured using a UAS over cotton fields in Mississippi across three growing seasons (2021-2023).
• Four classes of contamination elements (bag, bottle, can, trash) were annotated using bounding boxes.
• Data collection involved random placement of elements and varied flight parameters (height, speed) for enhanced variability.
• Full-size images (16 MP stills or 4K video) were pre-processed into 720x720 pixel tiles.

Main Results:

• The ACE dataset contains over 21,500 bounding box annotations.
• Annotations are categorized into bag (plastic bags, thin plastic sheet), bottle, can, and trash.
• The dataset includes images from different cotton growing stages, with 2021 data showing the most variation.
• Data distribution across years: 2021 (59%), 2023 (29%), and 2022 (12%).

Conclusions:

• The ACE dataset provides a valuable, large-scale resource for developing and validating AI models for agricultural contamination detection.
• The dataset's temporal and environmental variations support research into model robustness and performance across different conditions.
• This resource can aid in improving precision agriculture practices and environmental monitoring in agricultural settings.