Bruise Detection of Guava Fruit Based on Thermal Imaging

Abstract

A consistent and recurring problem in the agricultural sector is the bruising of fruits, which leads to many consumers basing their perception on the quality of fruit by the appearance of it. While external bruises on fruits are detectable, internal bruising is hard to detect using the available methods and traditional physical assessment methods, as most of these happen during the harvesting, transportation and storage of fruits. Undetected and internal damage leads to losses due to wastage, reduced shelf-life and other economic costs. Current industrial-grade detection methods are expensive and complicated: for example, X-rays and hyper-spectral imaging are unavailable to the general public. The issue of finding hidden injuries in Guava fruits at the point of sale needs a workable, reasonably priced, user-friendly solution. This presents a method for the detection of bruise in guava fruits using machine learning and thermal imaging. The study employed the YOLOv8 deep learning model for detecting bruises in guava fruits and classifying the thermal images. Thermal images derived from a publicly available dataset were preprocessed and used to train the model. The model showed 90% accuracy, which was depicted through a confusion matrix. Evaluation and testing of the model was carried out with tests datasets.