CNN Comparative Study for Apple Quality Classification

Abstract

A reliable and efficient automated fruit grading process is needed to meet the market's increased demand for good quality fruits. Automated systems based on image processing technology are being developed to reduce reliance on manual expertise, which is often time-consuming, expensive, and biased. The propose of this paper is to construct and evaluate different types of CNN models, including conventional CNN, GoogLeNet and AlexNet, for categorising fresh and rotten appearances using a dataset of apple images. All the datasets had performed the operations of pre-processing steps with scaling, rotating, and cropping. All models were trained and tested using the datasets provided by the KAGGLE network. The accuracy and loss performance of all the models are measured. The result shows that the GoogLeNet achieved 100% accuracy, which has better performance compared to the AlexNet and conventional CNN. Hence, the GoogLeNet model has the potential for integration into an automatic fruit grading system.