Comparative study on fine-tuning deep learning models for fruit and vegetable classification

Abstract

Fruit and vegetable recognition and classification can be a challenging task due to their diverse nature and have become a focal point in the agricultural sector. In addition to that, the classification of fruits and vegetables increases the cost of labor and time. In recent years, deep learning applications have surged to the forefront, offering promising solutions. Particularly, the classification of fruits using image features has garnered significant attention from researchers, reflecting the growing importance of this area in the agricultural domain. In this work, the focus was on fine-tuning hyperparameters and the evaluation of a state-of-the-art deep convolutional neural network (CNN) for the classification of fruits and vegetables. Among the hyperparameters studied are the number of batch size, number of epochs, type of optimizer, rectified unit, and dropout. The dataset used is the fruit_vegetable dataset which consists of 36 classes and each class contains 1,000 images. The results show that the proposed model based on the batch size=64 and the number of epochs=25, produces the most optimal model with an accuracy value (training) of 99.02%, while the validation is 95.73% and the loss is 6.06% (minimum).