Tobacco Stand Count

Project overview

Agricultural field images were annotated for a computer vision dataset focused on counting tobacco plants, commonly known as stand count. Individual tobacco plants were identified and accurate annotations created to generate high-quality training data for an AI-based detection and counting model.

The main objective was a reliable dataset able to automate tobacco plant stand counting in agricultural fields and provide accurate plant population estimates.

Work performed

• Reviewed and organised the agricultural field image dataset before annotation.
• Identified individual tobacco plants throughout the field images.
• Created accurate bounding boxes around each visible tobacco plant.
• Maintained consistent labelling using the Tobacco class.
• Partially visible plants were annotated wherever applicable.
• Distinguished tobacco plants from weeds, soil, crop residue and surrounding vegetation.
• Each plant was annotated separately so that stand counts stayed accurate.
• Carried out quality checks for missed, duplicated or incorrectly positioned annotations.
• Corrected inaccurate annotations and kept the dataset consistent throughout.
• Prepared the final dataset for training a computer vision model for tobacco plant detection and stand counting.

What made it hard

• Counting individual tobacco plants in dense field environments.
• Distinguishing tobacco plants from weeds and surrounding vegetation.
• Closely positioned and overlapping plants were hard to separate.
• Small plants at early growth stages were easy to miss.
• Partially visible plants had to be annotated accurately.
• Complex backgrounds contained soil, crop residue and other vegetation.
• Differing plant sizes and growth stages called for consistent annotation standards.
• No plant could be missed or counted twice without skewing the stand count.

Result

A high-quality tobacco stand count dataset was completed with accurate and consistent annotations. The dataset was prepared for training a computer vision model capable of automatically detecting and counting tobacco plants in agricultural fields.

Final deliverable

Delivered accurate and consistently labelled bounding box annotations for individual tobacco plants throughout the dataset. The final output matched the project requirements and was ready for AI model training and automated tobacco stand count and crop monitoring applications.