St. John's Wort Segmentation

Project overview

Agricultural field images were annotated for a computer vision dataset focused on segmenting St. John's Wort plants in real-world field environments. Precise polygon segmentation masks were created around visible plants to generate high-quality training data for an AI-based segmentation model.

The main objective was to prepare a reliable dataset that would allow an AI model to identify and segment St. John's Wort accurately while distinguishing it from surrounding crops, grass, weeds and other vegetation.

Work performed

• Reviewed and organised the agricultural field image dataset before annotation.
• Identified visible St. John's Wort plants in each image.
• Traced precise polygon segmentation masks along the visible boundary of every plant.
• Applied the STW class consistently across the dataset.
• Annotated partially visible plants where applicable.
• Differentiated St. John's Wort carefully from surrounding weeds, grass, crops and vegetation.
• Handled overlapping plants and complex vegetation with detailed polygon annotations.
• Performed quality checks and corrected inaccurate or incomplete segmentation masks.
• Ensured the final annotations were consistent and suitable for training a segmentation model.

What made it hard

• Differentiating St. John's Wort from visually similar weeds and surrounding vegetation.
• Small plants and thin plant structures made precise segmentation difficult.
• Dense field vegetation often left plant boundaries unclear.
• Overlapping leaves and plants required careful polygon placement.
• Partially visible plants had to have their visible boundaries estimated accurately.
• Different growth stages produced wide variation in plant size and appearance.
• Shadows and changing lighting conditions affected plant visibility.
• Holding polygon boundaries precise while excluding surrounding vegetation took significant attention to detail.

Result

A high-quality St. John's Wort segmentation dataset was completed with precise polygon annotations under the STW class. The dataset was prepared for training a computer vision model capable of automatically identifying and segmenting St. John's Wort in real-world agricultural and natural environments.

Final deliverable

Accurate and consistent polygon segmentation masks for the STW class were delivered across the dataset. The final annotations were prepared to the project requirements and were ready for AI model training and automated weed detection and monitoring applications.