Recipe Space: Novel deep learning approach for Food recognition and Recipe generation

Abstract

"The growing problem of food waste, which affects both the environment and resource management, is a growing concern. This is often due to the fact that people don't know how to use ingredients creatively in their homes, leading to unnecessary waste of resources. In order to reduce food waste and promote healthy eating habits, the thesis addresses the need for a new solution. To address this challenge, a system is being developed that uses advanced deep learning technology to identify food items from images or videos provided by users. The YOLOv8 action, which can correctly identify the various food components. To find and retrieve matching recipes for ingredients, this is connected to a backend that interfaces with the Spoonacular API. This approach is not only a way to reduce food waste by suggesting immediate uses for available ingredients, but also supports healthier choices of foods through the provision of nutritional information on recipes. The performance of this system, in particular the accuracy of the YOLOv8 model for identification of ingredients, has been thoroughly evaluated. The model showed very good accuracy, with a ratio of 0.854. The system has also been positively received by users, who have highlighted its usefulness in reducing food waste and the possibility of healthy eating. These results show that the system is technically suitable and useful for users who wish to use kitchen ingredients in a way which minimises waste. "