Aayush Dey

Detecting Rotten Fruit with YOLOv8

Started Jan. 20, 2023

Abstract or project description

Sorting and donating fresh produce makes a significant contribution towards helping impoverished families get the food they need. However, manually checking for rotten fruits or vegetables while sorting can prove to be a time-consuming and labor-intensive process, reducing the potential benefit of these projects. To address this problem, we used YOLOv8 to create a self-trained object detection model, capable of locating fresh and rotten fruit of any quantity in a variety of backgrounds and lighting conditions. By detecting large amounts of rotten fruit at once, this model could make the sorting process far more efficient. Many related projects centered around fruit production and distribution have already utilized object detection algorithms to automate tasks like sorting and yield estimation. While their methods involve training on real-world images, we instead created a synthetic dataset that could mimic real-world conditions while granting us full control over its features. After testing the model on real-world images, we found that our synthetic data produced similar results to other authentic datasets. When a small number of real-world images were added to the model’s training, its accuracy increased considerably to about 99%, making its use practical in sorting operations.