Tassel Detection

Project overview

The Tassel Detection project covered agricultural field images in which tassels had to be identified alongside the other objects and vegetation that appear in a working field. Five classes were annotated in total, spanning target crop structures, competing vegetation and waste.

The dataset was prepared for computer vision model training focused on tassel detection and on separating target objects from weeds, waste and surrounding vegetation.

Work performed

• Reviewed agricultural field images and identified every object belonging to the defined classes.
• Created accurate bounding box annotations around tassels and other target objects.
• Differentiated tassels from Johnson grass, shrubs, dried leaves and other vegetation.
• Garbage was annotated separately whenever it appeared in the field images.
• Maintained consistent labelling across different image conditions and object sizes.
• Checked annotations for missing, incorrect or poorly positioned bounding boxes.
• Prepared the dataset so that it suited object detection model training.

What made it hard

• Differentiating tassels from surrounding grass and vegetation.
• Small or partially visible tassels were easy to overlook.
• Overlapping plants and dense field vegetation complicated box placement.
• Dried leaves proved difficult to separate from other plant material.
• Lighting and field conditions varied considerably between images.
• Objects differed widely in size and orientation, making consistent boxes harder to maintain.

Result

A high-quality annotated dataset was prepared for tassel detection and wider agricultural object detection, with clear separation between tassels, weeds, garbage, dried leaves and shrubs.

Final deliverable

The final dataset contained accurate and consistent bounding box annotations for all five classes and was prepared for use in computer vision model training and agricultural monitoring applications.