AI Based Automated Food Recognition Using CNN for Fitness Mobile Application

Eating a healthy, balanced diet with regular exercise is important for maintaining physical and mental health. Healthy diet and exercise promise a healthier lifestyle. Other than that, people competing in different sports need a customized, strict workout and diet plan. Trainees offer their best at gymnasium by following a strict workout plan recommended by their coaches. However, there is no way till now to determine if the trainees are taking the prescribed diet strictly. Therefore, an optimized application with the feature of real time scanning of the prescribed meal can be helpful to overcome this problem. For this purpose, we have used Convolutional Neural Networks (CNN) along with Image Processing. Using ResNet50 with activation functions of ReLU and Softmax, our trained CNN model and the proposed architecture achieved an accuracy of 82.70 percent. We used CNN pre-trained models such as InceptionV3 with LeakyReLU and Softmax, InceptionV3 with ReLU and Softmax, VGG16 with ReLU and Softmax, VGG16 with ReLU and Softmax, and ResNet50 with ReLU and Softmax, so the achieved accuracy, respectively, is 78%, 71%, 58%, 56%, and 82.7%. Therefore, the proposed model has the potential to achieve high performance on real time food scanning which can be used in online fitness platforms(mobile applications and websites) and fitness gadgets.