Abstract
Fruits play a vital role in the food and agricultural industries, where maintaining their quality is essential for food safety, consumer satisfaction, and reducing economic losses. Fresh fruits retain their original physical and textural characteristics, while stale fruits do not. Manual fruit quality assessment remains common but is time-consuming, subjective, and prone to inconsistencies, particularly for large-scale operations. These limitations underscore the need for automated, efficient, and accurate solutions to classify fruits effectively. This study presents a Deep Learning (DL)-based approach for classifying fruit images as fresh or stale using the MobileNetV2 architecture. Leveraging advancements in computer vision and Machine Learning (ML), we explored automated methods for fruit quality assessment. A dataset of 40,933 fruit images, encompassing a wide variety of fruits at different freshness levels, was collected and preprocessed. The MobileNetV2 model was fine-tuned for binary classification using this dataset, achieving remarkable performance: 95.66% accuracy, 96.23% precision, 95.02% recall and 95.62% f1 score on the test set. The lightweight architecture of MobileNetV2 ensures computational efficiency, making it suitable for real-time applications and deployment on edge devices. Comparative analyses with other pre-trained models, including InceptionV3, Xception, VGG16, VGG19, ResNet50, and EfficientNetV2L, demonstrated that MobileNetV2 excels in both accuracy and efficiency, solidifying its potential for real-world implementations. This research contributes to smart agriculture by providing a scalable and automated solution for fruit classification, paving the way for efficient and accurate quality control systems in agricultural and food industries. Our findings support advancements in agricultural automation, enabling consistent and real-time fruit quality assessment across diverse settings.
