Food image recognition based on Mobile NetV2 using support vector machine
Synopsis
The rapid growth in deep learning has made convolutional neural networks deeper and more complex to realize higher accuracy. But many day-to-day recognition tasks need be performed in a limited computational platform. One of the applications is food image recognition which is very helpful in individual’s health monitoring, dietary assessment, nutrition analysis etc. This task needs small convolutional neural network based engine to do computations fast and accurate. MoblieNetV2 being simple and smaller in size can incorporate easily into small end devices. In this paper, MobileNetV2 and support vector machine are used to classify the food images. Simulation results show that the features extracted from Conv_1 layer, out_relu layer and Conv_1_bn layer of MobileNetV2 and classified using Support Vector Machine have achieved classification accuracies of 84.0%, 87.27% and 83.60% respectively. Because of fewer parameters, smaller size and lesser training time, MobileNetV2 is an excellent choice for real-life recognition tasks.
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.