Aerial Cotton Field Images with Plastic Contamination Annotations

Three growing seasons of cotton fields in Mississippi from 2021 to 2023 are documented in this dataset. It contains 720 x 720 pixel aerial image tiles with XML annotations for four contamination classes: bags, bottles, cans, and trash. Sean Donohoe created this public domain dataset to support vision systems for detecting field contamination before harvest.

Use Cases

• Train object detection models to locate bags, bottles, cans, and trash in 720x720 aerial image tiles using bounding box annotations.
• Benchmark vision system performance for detecting plastic sheet-like materials (bag class) that intertwine with cotton plants.
• Develop models to distinguish ground-level contaminants like drink containers (bottle/can classes) from other foreign materials (trash class).
• Create systems for pre-harvest quality control by identifying contamination that can diminish cotton bale value.

Strengths

• Contains data from three distinct growing seasons (2021, 2022, 2023).
• Annotations use precise bounding boxes in XML files for four specific object classes.
• Images are high-resolution square tiles of 720 x 720 pixels.

Limitations

• Sample size (number of images/rows) is not specified in the provided description.
• Geographic coverage is limited to cotton fields in Mississippi, USA.
• Contamination was artificially placed and removed for imaging, which may not reflect real-world distribution.

Provenance

Source

Sean Donohoe via figshare.

Collection Method

Aerial images of cotton fields with artificially placed and subsequently removed contamination, annotated with bounding boxes.

Time Range

2021, 2022, 2023 growing seasons.

Freshness

Last updated in April 2026, with data covering up to the 2023 growing season.

Geography

Cotton fields in Mississippi, USA.