A Mobile Application for Food and Its Ingredients Detection Using Deep Learning

Food identification has become a demand not only for people with special dietary, but also for religious or healthy reason. Other than that, ones with culinary interest would also like to have the information of ingredients or the name of a cuisine to satisfy their strong desire. However, it is not easy to get this information as it is usually a secret recipe or a restaurant ultimate formula. Moreover, people hesitate or unsure to ask about this in a public place. As a result, we need a method that can help us to detect, both the name and the ingredients. In general, machine learning methods might be available to create a model that can classify cuisine. Some optimizer are available to create a machine learning model that also provide learning rate which determine variable of how fast the model can learn throughout the data. In this work, we present an empiric result on performance of MobileNets as learning resources for Deep Learning (Tensorflow). We retrain our data - that consist of 32.194 images - from food data set that is retrieved via Google search images, to specifically detect the name and the ingredients which is suitable for specific optimizer. In order to get the result, we retrain our model using 4 optimizer, which are Gradient Descent Optimizer, Adagrad Optimizer, RMSProp Optimizer and Adam Optimizer with 4 different learning rate which are 0.001, 0.01, 0.1, and 1 to find which optimizer produce the best for food recognition.