Tobacco Damaged Leaf Count

Project overview

Tobacco plant images were annotated for a computer vision dataset focused on detecting and counting tobacco leaves according to their condition. Individual leaves had to be identified and then separated into normal and damaged categories.

The main objective was high-quality training data for an AI model capable of automatically detecting, classifying and counting normal and damaged tobacco leaves.

Work performed

• Reviewed and organised the tobacco image dataset before annotation.
• Identified individual tobacco leaves in each image.
• Classified every leaf according to its visible condition.
• Created accurate bounding boxes around individual leaves.
• Healthy leaves were labelled Normal and visibly affected leaves Damage.
• Maintained consistent class labelling throughout the dataset.
• Partially visible leaves were annotated wherever applicable.
• Assessed visible damage such as holes, discolouration and damaged leaf areas.
• Carried out quality checks and corrected inaccurate or missing annotations.
• Checked that every target leaf was properly annotated to support reliable counting.
• Prepared the dataset for training an AI model for tobacco leaf detection, classification and counting.

What made it hard

• Distinguishing normal leaves from damaged leaves based on subtle visual differences.
• Minor or partially visible damage was easy to overlook.
• Overlapping tobacco leaves complicated box placement.
• Leaves appeared at different growth stages and sizes.
• Varied types and levels of damage made classification harder.
• Shadows and changing lighting conditions altered the appearance of damaged areas.
• Complex backgrounds made individual leaves difficult to isolate.
• Every visible leaf had to be annotated correctly for the counts to hold up.

Result

A high-quality tobacco damage detection and counting dataset was completed with accurate annotations for the Normal and Damage classes. The dataset was prepared for training a computer vision model capable of detecting, classifying and counting tobacco leaves according to their condition.

Final deliverable

Delivered accurate and consistently labelled annotations for Normal and Damage tobacco leaves throughout the dataset. The final output matched the project requirements and was ready for AI model training and automated tobacco leaf damage assessment and counting.