Garlic Detection & Counting

Project overview

Agricultural images were annotated for a computer vision dataset focused on detecting and counting garlic plants and bulbs in field environments. The project was one of the most challenging annotation projects undertaken, owing to the small size of garlic plants, dense vegetation, overlapping leaves, and complex agricultural backgrounds.

The main objective was to create accurate annotations that could be used to train an AI model for automated garlic detection and counting.

Work performed

• Reviewed and organised the agricultural image dataset before annotation.
• Identified individual garlic plants and heads throughout the field images.
• Created accurate bounding boxes around visible garlic targets.
• Consistent labelling was maintained using the Garlic class.
• Partially visible garlic plants were annotated where applicable.
• Distinguished garlic plants carefully from surrounding crops, weeds, soil, and other vegetation.
• Ensured each visible garlic target was properly annotated to support accurate counting.
• Detailed quality checks were performed to identify missing, overlapping, or incorrectly positioned annotations.
• Corrected inaccurate annotations and maintained consistency across the dataset.
• Prepared the annotated dataset for training an AI model capable of detecting and counting garlic automatically.

What made it hard

• Garlic plants were often very small and difficult to identify.
• Dense and overlapping leaves made individual garlic plants difficult to separate.
• Some garlic plants were partially hidden by other plants or vegetation.
• Similar-looking weeds and surrounding crops created confusion during annotation.
• Complex field backgrounds made object boundaries difficult to identify.
• Different camera angles, lighting conditions, and image quality affected visibility.
• Maintaining accurate annotations while ensuring every garlic plant was counted required significant attention to detail.
• Closely positioned garlic plants made it challenging to create separate bounding boxes without overlap or missed objects.

Result

A challenging garlic detection and counting dataset was completed with accurate and consistent annotations. The dataset was prepared for training a computer vision model capable of automatically detecting and counting garlic plants in agricultural fields.

Final deliverable

A high-quality annotated garlic dataset was delivered with carefully reviewed bounding boxes designed to support both object detection and counting. Despite the significant challenges caused by dense vegetation, small objects, and complex field conditions, the final dataset was prepared according to project requirements and was ready for AI model training and agricultural monitoring applications.