Food Genre Classification from Food Images by Deep Neural Network with Tensorflow and Keras

Different varieties of food and its quality speak the culinary tendencies and religious beliefs of people from different regions of the world. The food reflects culture and each food uses unique and combination of ingredients. It is a fashion and trend now a day to develop new food varieties, cooking methods and trying out cross culture recipe. Handling the food serving for a multi ethnic, multi culture and different belief people in a common environment like conferences, air lines, quarantine centers, even refugee camps etc. is a challenge and need concentration and attention. A safer way of using untiring intelligent machines to replace the food distribution to ease and protect chefs and kitchen staff is felt essential. We developed a system of intelligent visual food identification in order to avoid errors that may happen when handling big numbers of food distribution by deploying smart machines. We choose 17 assorted varieties of food images and trained the system to identify its correct type and nature. We used Convolution Neural Network with Tensor-flow and Keras and trained the network with 170 different food images. We trained the system with online and offline images and achieved training accuracy of percentage. The experiment results show 80 percentages and are not only limited to single entity but describing multiple possible labeling of food according to its nature.