Integrating YOLOv8 and Cosine Similarity for Food Recipe-Specific Ingredient Collection and Recommendation

Artificial Intelligence (AI) is revolutionizing all the industry, including the culinary domain. The paper aims to develop an innovative food recommendation system by integrating the YOLOv8, for object detection algorithm and the Cosine Similarity Algorithm, for ingredient detection and recipe suggestions. A system which the researchers have developed can identify ingredients and suggest corresponding recipes. A custom dataset was made consisting of twenty-one (21) ingredients for training the algorithm YOLOv8. For the Cosine Similarity, a dataset consisting of ten (10) dishes such as “Chicken Tinola”, “Chicken Adobo”, “Chicken Afritada”, “Chicken Curry”, “Sinigang na Baka”, “Nilagang Baka”, “Beef Kare-Kare”, “Sinigang na Baboy”, “Pork Humba”, and “Pork Menudo” was utilized. The researchers utilized NVIDIA Jetson Nano for the main computing, and Raspberry PI V2 Camera for capturing the ingredients. The results of the detected samples were then shown on 5-inch LCD Display. The accuracy of the system that yielded 84.76%, has been determined by using a confusion matrix. The precision of the system resulted in 91.75%, while the recall was 91.75%. The F1 score of the system is determined to be 91.75 percent Additionally, a percentage error formula was utilized for recipe recommendations, resulting in a 12% error rate.