Fruit Image Classification Using CNN With EfficientNet Architecture for Visual Education

Abstract

Advancements in artificial intelligence and computer vision have significantly influenced education, particularly by enhancing visual-based learning for young learners. One promising application is fruit image classification, which helps students recognize and differentiate fruits through visual cues. Traditional methods often struggle with varied backgrounds and lighting conditions, making deep learning models more suitable. This study aims to develop an efficient fruit classification system using the EfficientNetB0 architecture within a convolutional neural network (CNN) framework. This study evaluates the model's effectiveness as a visual learning tool in educational contexts while ensuring computational efficiency. The dataset, sourced from Kaggle, consists of eight fruit categories: apples, bananas, kiwis, lemons, passion fruits, peaches, pineapples, and raspberries. It was split into training and validation sets with an 80:20 ratio using stratified random sampling to ensure balanced class representation during evaluation. Preprocessing steps included resizing images to 224×224 pixels, normalization with EfficientNet preprocessing, and data augmentation techniques to improve generalization. A custom classification head was added, and the EfficientNetB0 base was frozen. Training employed the Adam optimizer, categorical cross-entropy loss, early stopping, and class weighting across 30 epochs. The model achieved a validation accuracy of 99%, with near-perfect precision, recall, and F1-score across all classes. The confusion matrix showed minimal misclassification, indicating strong generalization even among visually similar fruits. In conclusion, the EfficientNetB0-based model demonstrates high accuracy, balance, and computational efficiency. It is ideal for integrating interactive visual learning tools to enhance concept recognition in educational settings, particularly among early learners.