Determining the freshness of fruits in the food industry by image classification using transfer learning

• Published in: Multimedia tools and applications Vol. 81; no. 6; pp. 7611 - 7624
• Main Authors: Kazi, Aafreen, Panda, Siba Prasada
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.03.2022; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Artificial neural networks; Computer Communication Networks; Computer Science; Computer vision; Data Structures and Information Theory; Freshness; Fruits; Image classification; Machine learning; Multimedia Information Systems; Neural networks; Object recognition; Special Purpose and Application-Based Systems; Classical convolutional networks; CNN architectures; Residual networks; Deep learning algorithms; Food industry; Image classification
• ISSN: 1380-7501; 1573-7721
• DOI: 10.1007/s11042-022-12150-5

Convolutional neural networks (CNNs) play a significant role in a number of computer vision applications like image classification and object detection. Past studies have focused on the application of deep neural networks and convolutional neural networks for detecting the freshness of perishable goods like fruits and vegetables. However, this study tries to explore the potential of transfer learning with respect to the CNN models in image classification of fruits rather than implementing the traditional CNN architectures. In this study, the determination of three different types of fruits and its relative freshness was classified using various architectures of classical convolutional neural networks and a residual convolutional neural network. The performance of each model of the convolutional neural network was evaluated on the given set of data. The study tried to evaluate the best performing model on image data that consisted of six different types of fruits based on their test scores. The results suggest that the freshness of fruits could be determined with high accuracy by using traditional and residual convolutional neural networks. However, the residual networks perform extremely well on the fruits dataset which was considered with a test accuracy score greater than 99%. A glimpse on the implementation of the given algorithm in the food industry was also explored. Thus, this study attempts to select the CNN model that could be applied in the food industry in complementation with other computer vision techniques like background reduction, image augmentation methods.