Fruit and Vegetable Recognition System Using Deep Transfer Learning Approach

Effectively categorizing an extensive range of fruits and vegetables can be accomplished by leveraging computer vision and image processing techniques. This article presents the development of an automated system for recognizing fruits and vegetables, employing deep transfer learning models. To build our image dataset, we combined images sourced from the publicly available “Fruit 360 dataset” with images we captured ourselves. The dataset encompasses 26 different types of fruits and 34 distinct vegetable categories. For this research, we primarily leveraged three key techniques: Simple CNN, ResNet50, and MobileNetV2, to carry out the final classification task. The achieved accuracy rates for these models were as follows: 76.67% for Simple CNN, 78.63% for ResNet50, and an impressive 96.52% for MobileNetV2. Furthermore, we have harnessed the power of the MobileNetV2 model to create a novel Android smartphone application within Android Studio. This application is designed to automatically classify fruits and vegetables. Users can take advantage of this app to identify and categorize fruits and vegetables using either their device’s camera or images from their gallery. One notable feature of this app is its ability to provide information in three different languages: English, Khasi, and Garo, ensuring accessibility for a wide range of users. With this innovative application, anyone can easily and accurately identify a variety of fruits and vegetables, making it a valuable tool for both consumers and enthusiasts.