Recipe recommendation system via object detection

Cooking is one of the most important daily activities in our lives. However, it takes a long time for people, such as employees and students, to decide what to cook and what they can do with what they have in their hand for those with limited time and little experience in the kitchen. Even if a decision is made, it is not an easy task to achieve harmony between ingredients when trying to do something with the ingredients at hand. For this reason, the recipes provided by the food sites available on the internet are browsed by the people, but since all the ingredients are included in those food sites, they direct people to buy additional ingredients. This study eliminates this problem by searching recipes according to the ingredients at hand. In this thesis, the input image taken from the user; firstly the image is processed through the image pre-processing steps. As a result of these operations, more than one contour is detected as output and these contours are saved as separate images. These saved images are given as an input to the deep convolutional neural network model and recognition process is started. At the recognition stage; Xception was selected from deep convolutional neural network architectures trained in ImageNet dataset with transfer learning. It is customized by fine tuning on the Xception architecture. In this new network model created: accuracy; 0.9452, loss; 0.4930, validation loss; 0.4883, validation accuracy; 0.9362 has been achieved. The recognition tag of each image that comes out of this network is saved in a list. When the recognition process is completed, the values in this list are sent to the database as queries. As a result of the query, a recipe suggests which recipes can be made with the ingredients in that list. As a result, the fact that people can search only according to the ingredients in their kitchens will prevent the person who will cook to go out and go shopping for the moment. Thus, they will be able to save time and cost.