Terra Weed Segmentation

Project overview

Agricultural field images were annotated for a computer vision dataset focused on detecting and segmenting weeds in real-world field environments. Visible weeds were identified and precise annotations created to generate high-quality training data for an AI-based computer vision model.

The main objective was a reliable dataset for automatically identifying weeds while distinguishing them from crops, soil and surrounding vegetation.

Work performed

• Reviewed and organised the agricultural field image dataset before annotation.
• Identified visible weeds throughout the field images.
• Created precise polygon segmentation masks around individual weeds.
• Maintained consistent labelling using the Weed class.
• Followed the visible boundaries of each weed carefully.
• Differentiated weeds from crops, soil, grass and other surrounding vegetation.
• Partially visible weeds were annotated wherever applicable.
• Carried out quality checks and corrected inaccurate or incomplete segmentation masks.
• Kept annotations consistent and suitable for computer vision model training.

What made it hard

• Differentiating weeds from crops and other vegetation.
• Dense field environments made individual weed boundaries difficult to identify.
• Overlapping plants required careful polygon placement.
• Small and partially visible weeds were challenging to annotate accurately.
• Similar-looking plants sometimes made weed identification difficult.
• Soil, crop residue and natural field backgrounds added further complexity.
• Varying lighting conditions and shadows affected weed visibility.
• Maintaining precise segmentation boundaries across the dataset demanded close attention to detail.

Result

A high-quality Terra Weed segmentation dataset was completed with precise polygon annotations for the Weed class. The dataset was prepared for training a computer vision model capable of automatically detecting and segmenting weeds in agricultural field environments.

Final deliverable

Delivered accurate and consistent polygon segmentation masks for the Weed class throughout the dataset. The final annotations matched the project requirements and were ready for AI model training and automated agricultural weed detection and monitoring applications.