| With the accelerated aging of the population and the improvement of people’s living standards,more and more Chinese people are paying attention to healthy diet,especially the elderly living alone.Therefore,a home social robot that can monitor and analyze diet every day is necessary.Through the monitoring and analysis of the user’s diet,valuable information can be provided for disease diagnosis and dietary habits management.In view of this,in order to solve the problem of diet quality assessment of the elderly,this paper designs a user-based intelligent perception social robot based on machine vision,which can monitor and analyze the user’s diet information.First,in order to realize the monitoring of users’ diet by social robots,a single-dish classification and recognition algorithm based on improved VGGNet and a multi-dish recognition algorithm based on YOLOv5 were proposed successively.Secondly,in order to obtain the user’s dietary information,a visionbased nutritional component autonomous perception algorithm is proposed,which establishes the intake requirement standard of healthy dietary components,calibrates the nutritional component information of the dishes,identifies and records the user’s dietary behavior,and counts each daily user dietary information.Finally,in order to analyze the user’s diet information,an intelligent perception system of user’s diet information is developed,and integrated into the social robot platform,to realize the test of the prototype system on the social robot platform.The main research contents of this paper are as follows:1)Firstly,a single dish classification and recognition algorithm based on improved VGGNet is proposed for the identification of a single dish;then,a Chinese dish recognition and classification data set CFNet-232 is built on the basis of the public Chinese dish data set Chinese Food Net;finally,we get The average recognition accuracy rate of the algorithm tested on the self-built data set is 86.13%,and compared with the other four network models,the Top-1 and Top-5 accuracy rates of the test set are higher than those of the other four network models,indicating that The recognition and classification accuracy of the single dish classification and recognition algorithm based on the improved VGGNet is better than that of the comparison algorithm.Considering the actual family dining environment and the requirement of real-time detection and lightweight of the algorithm,a multi-dish recognition algorithm based on YOLOv5 is proposed,and the recognition accuracy of the algorithm is 89.70%.2)In order to obtain the user’s dietary information,a vision-based nutritional composition autonomous perception algorithm is proposed to realize the autonomous perception of the nutritional composition information ingested by the user.In order to facilitate quantitative analysis of nutritional components,based on this algorithm,a standard for the intake of healthy dietary components was established,the nutritional information of food ingredients was calibrated,the user’s dietary behavior was identified and recorded,and the daily user’s dietary information was counted.3)A home social robot platform is built for the proposed intelligent perception method of user diet information,and the above algorithms are integrated to realize the test of user diet information intelligent perception system on the social robot platform.The system performance test results show that the average nutrient composition perception accuracy of the system is 90.1%,the response time is less than 6ms,and the speed is greater than 18 fps,which has good robustness. |
