Wheat Detection

Project overview

The Wheat Detection project involved annotating wheat plants according to the number of visible leaves. Different classes were assigned depending on whether an individual wheat plant carried one leaf, two leaves, or three or more leaves. The dataset was prepared for computer vision model training focused on wheat detection and leaf-count-based classification.

The three classes were Z11 for a wheat plant with one leaf, Z12 for a plant with two leaves, and Z13 for a plant with three or more leaves.

Work performed

• Reviewed wheat field images and identified individual wheat plants.
• Counted the visible leaves on each plant.
• Assigned Z11 to plants carrying one leaf.
• Plants with two leaves were assigned Z12.
• Three or more leaves placed a plant in the Z13 class.
• Created accurate bounding box annotations around individual plants.
• Classification was kept consistent across different plant sizes and growth stages.
• Reviewed the annotations and corrected missing or incorrectly classified plants.
• Prepared the dataset for wheat detection and leaf-count-based model training.

What made it hard

• Counting visible leaves accurately on small wheat plants.
• Distinguishing individual plants where they grew close together or overlapped.
• Partially visible leaves and plants complicated the count.
• Differentiating wheat leaves from surrounding vegetation.
• Separating plants correctly into the Z11, Z12 and Z13 classes.
• Keeping classification consistent across different field and lighting conditions.

Result

A high-quality annotated dataset was prepared for wheat detection and leaf-count-based classification, with three clearly defined classes based on the number of visible leaves.

Final deliverable

The final dataset contained accurate and consistent bounding box annotations for Z11, Z12 and Z13, enabling computer vision models to detect wheat plants and classify them by visible leaf count.