Vision-Based Yakiniku Recognition System for Autonomous Cooking Robot

This paper presents a vision-based meat recognition system intended for cooking robots in the Japanese barbeque or ‘yakiniku’ practice. The objective of the recognition system is to directly recognize the meat slice image from the camera as one of three doneness classes, including undercooked, well-cooked, and overcooked, with almost no preprocessing to the input image. A simple Convolutional Neural Network (CNN) is employed as the main component of the proposed recognition system. The developed system is evaluated in two experiments: the experiment of the CNN with images in the test set, and the experiment of the entire recognition system in the real kitchen. Results from the experiments were evaluated through precisions and recalls. The experimental results show promising performance of the proposed system for recognizing doneness of meat slices using visual information in both the dataset and the real world.