YOLOv5-Based Image Processing for Pineapple Rind Defect Detection

This study implemented YOLOv5 (You Only Look Once) as its algorithm to detect pineapple rind defects using known image processing techniques such as preprocessing and augmentation. Python language was used to integrate the YOLOv5 algorithm, and a Graphical User Interface (GUI) was designed to allow the system to communicate with the user. Using the 5MP OV5647 Camera, the software can process source inputs such as image capture and video. The dataset contents follow, where 70% of training data (350) images, 20% of valid (100) images, and 10% of testing (50) images have been used to create the dataset. Additionally, the dataset had undergone preprocessing and augmentation to maximize the convergence rate of the model. As the system has been implemented and tested throughout the 40 images of pineapples that overall contain 20 rind defects and 20 with no rind defects, the trained model can identify pineapple rind defects at a default confidence threshold of 0.60. With that, the proponents obtained an accuracy score of 88.10^%, which can be considered an ideal and capable model for capturing and detecting pineapple rind defects.