Thistle Detection

Project overview

Agricultural field images were annotated for a computer vision dataset focused on detecting thistle plants in real-world agricultural environments. The project involved identifying visible thistle plants and creating accurate bounding box annotations to generate high-quality training data for an AI-based object detection model.

The main objective was to prepare a reliable dataset for detecting thistle plants automatically while distinguishing them from crops, grass and other surrounding vegetation.

Work performed

• Reviewed and organised the agricultural field image dataset before annotation.
• Visible thistle plants were identified throughout the field images.
• Accurate bounding boxes were drawn around individual plants.
• Consistent labelling was maintained using the thistle class.
• Annotated partially visible plants where applicable.
• Thistle was carefully differentiated from crops, grass, weeds and other vegetation.
• Plants at different growth stages and sizes were handled individually.
• Quality checks corrected inaccurate, missing or poorly positioned annotations.
• Confirmed annotations were consistent and suitable for object detection model training.

What made it hard

• Differentiating thistle from other weeds and surrounding vegetation.
• Small plants in dense agricultural fields were difficult to detect.
• Partially visible or overlapping plants complicated box placement.
• Different growth stages produced wide variation in plant size and appearance.
• Complex field backgrounds made plant identification harder.
• Similar-looking vegetation occasionally made class identification difficult.
• Varying lighting conditions and shadows affected plant visibility.
• Keeping bounding boxes tight around irregularly shaped plants required careful attention to detail.

Result

A high-quality thistle detection dataset was completed with accurate and consistent bounding box annotations. The dataset was prepared for training a computer vision model able to detect thistle plants automatically in real-world agricultural environments.

Final deliverable

Accurate and consistent bounding box annotations for the thistle class were delivered throughout the dataset. The final output met the project requirements and was ready for AI model training and automated weed detection applications.