FruitsMultiNet: A deep neural network approach to identify fruits through multi-scale feature fusion using mobile interface

The breakthrough of smart supply chain systems and the rapid adoption of automation in the growing fruit production and processing sectors are driving the increasing demand for an automatic fruit classification approach. A reliable fruit classification system during harvest and post-harvest phases can minimize time, cost, and human error while modernizing the processes of sorting, labeling, and packaging. The proposed research suggests a groundbreaking autonomous fruit classification method grounded on convolutional neural networks (CNNs). The classification of fruits is often challenging due to variations that occur in the same fruit throughout its life cycle. Additionally, the diverse compositions of sizes, shapes, and colors further complicate the process and impact accuracy. This research worked with a unique dataset of 3240 images of Bangladeshi fruits, publicly available on Kaggle. The data were preprocessed before the feature extraction phase to achieve more accurate outcomes. MobileNet, VGG16, NasNetMobile, DenseNet201, InceptionV3, and Xception were experimented with in the feature extraction and performance evaluation process. The proposed research combines MobileNet and VGG16 to build a transfer learning (TL) principles-based framework named FruitsMultiNet. These models were chosen because they achieved the highest individual accuracy compared to other models. The data were fragmented into training, testing, and validation sets for eight different fruits. The proposed FruitsMultiNet achieved a 99.84 % accuracy rate, which is greater than the accuracy rates of individual Deep Learning (DL) models. The integration of FruitsMultiNet into a mobile application makes it easily accessible for individuals, adding consistency and providing consumers with a more accessible system to classify fruits automatically. The application also provides nutritional information and health benefits of the recognized fruit for humans.