Food Recognition based on Deep Learning Algorithms

Accurate methods can help the technology nowadays to keep improving and provide a reliable system for the people to use. In this paper, two different image classification systems; Convolutional Neural Network (CNN) and Residual Neural Network (ResNet) were proposed in order to recognize six food classes; Apple, Orange, Avocado, Milo, Vico and Koko based on color features. Then, the overall performance for both classifications were analyzed in the end of this paper. Datasets of food images were collected from various sources consisting 400 images for each food classes to test the robustness of each classification system. The data were split into 60% training data, 20% validation data and 20% testing data. The system that is proposed in this paper consist of 4 layers for Convolutional Neural Network (CNN) while Residual Neural Network (ResNet) consist of 50 layers. The color feature extraction that is involved for both classifications, RGB values (Red, Green, Blue) are highly considered in order to determine the category of the food. Overall, this experimental results on food recognition showed 100% training accuracy and 98.67% overall testing accuracy for CNN while 99.87% training accuracy and 96.67% overall testing accuracy for ResNet.