Palm Oil Fruit Classification

Project overview

Palm oil fruit images were annotated for a computer vision dataset focused on classifying fruits according to their ripeness and condition. The work involved identifying fruits at different maturity and quality levels and creating accurate annotations for each category.

The main objective was to develop high-quality training data for an AI model able to recognise the condition of palm oil fruits automatically in agricultural environments.

Work performed

• Reviewed and organised the palm oil fruit image dataset before annotation.
• Individual fruits were identified and their visible condition assessed.
• Classified each fruit into the appropriate maturity or quality category.
• Accurate annotations were drawn around the target fruits.
• Consistent labelling was maintained across all four classes.
• Normal, underripe, overripe and rotten fruits were distinguished by their visual characteristics.
• Handled fruits that were partially visible or clustered together.
• Quality checks corrected inaccurate or inconsistent annotations.
• Confirmed the final dataset was suitable for training a computer vision classification and detection model.

What made it hard

• Telling normal, underripe and overripe fruits apart relied on subtle visual differences.
• Rotten fruits were hard to identify when damage or discolouration was partially hidden.
• Densely packed fruit bunches had to be worked through carefully.
• Overlapping and partially visible fruits complicated boundary placement.
• Lighting conditions altered the appearance of fruit colour and maturity.
• Variations in fruit size, colour and surface condition made consistent classification difficult.
• Keeping class labels accurate across a large dataset required careful quality control.

Result

A high-quality palm oil fruit classification dataset was completed, covering four classes: overripe, underripe, rotten and normal. The dataset was prepared for training a computer vision model able to identify palm oil fruit maturity and quality conditions automatically.

Final deliverable

Accurate and consistently labelled annotations were delivered for overripe, underripe, rotten and normal palm oil fruits. The final dataset met the project requirements and was ready for AI model training and automated palm oil fruit maturity and quality assessment.