Orange Detection & Ripeness Classification

Project overview

Images of oranges lying on the ground were annotated for a computer vision dataset focused on detecting and classifying oranges by condition and ripeness. Each visible orange was annotated with a bounding box and assigned an appropriate class.

The dataset was prepared as training data for an AI model capable of detecting oranges and distinguishing between rotten, ripe and green fruit.

Work performed

• Reviewed and organised the orange image dataset before annotation.
• Individual oranges lying on the ground were identified.
• Accurate bounding boxes were created around each visible orange.
• Every orange was classified as Rotten, Ripe or Green.
• Class labelling remained consistent throughout the dataset.
• Partially visible and overlapping oranges were annotated where applicable.
• Quality checks were performed, and inaccurate bounding boxes or class labels were corrected.
• Annotations were confirmed suitable for object detection and classification model training.

What made it hard

• Distinguishing between ripe and green oranges from visible colour and appearance.
• Rotten oranges had to be identified from visible damage, discolouration or decay.
• Overlapping oranges and partially occluded objects complicated boundary placement.
• Oranges needed detecting among grass, soil, leaves and other background elements.
• Consistent classification had to hold across different lighting conditions and viewpoints.

Result

A high-quality orange detection and classification dataset was completed containing the Rotten, Ripe and Green classes. Each visible orange was assigned the appropriate class and enclosed with an accurate bounding box, and the dataset was prepared for training an AI model capable of automatically detecting oranges and classifying their ripeness and condition.

Final deliverable

Accurately annotated orange images were delivered with bounding boxes for the Rotten, Ripe and Green classes, with consistent labelling and annotation quality maintained throughout the dataset. The final dataset was prepared according to project requirements and was ready for AI model training and development.