A Deep Learning Approach for Fruit Recognition and Nutrition Detection

Abstract

Fruit recognition and nutrition detection consumes time and requires labor while using traditional methods. Deep learning can acquire robust features from images that's why it is used for this application. In this paper, feature extraction from fruit images is done using six different deep learning approaches. Then the fruit images are classified into different fruit categories and the nutritional content is estimated for each fruit image. The six deep learning approaches are CNN 15-layer, CNN 8-layer, VGG16, ResNet18, ResNet50, and InceptionV3 applied on 18 fruit types. To overcome the limitations in each dataset on its own, two datasets, Fruit-360 and Fruits and Vegetables Image Recognition were combined. Four optimizers are employed for evaluating each deep learning approach. These optimizers are Adam, Nadam, RMSprop, and SGD. This study highlights the importance of using diverse optimization algorithms within each deep learning approach and combining datasets to enhance the fruit recognition models accuracy. Results showed that the 15- layer CNN model achieved the highest accuracy for both fruit recognition and nutrition detection at 99.12%. This paper shows the potential of deep learning in automizing fruit recognition and nutrition detection which saves time and labor in food industry. These findings will promote healthy eating habits and facilitate dietary management.