| With people’s attention to healthy diet and balanced nutrition intake,and the continuous development of deep learning technology,more and more scholars begin to conduct research on food image recognition and detection.However,in the current research on food image recognition and detection algorithms,the differences between Chinese dishes and dishes from other countries are not fully taken into account,and the image characteristics of Chinese dishes,such as small color difference and great influence by cooking methods,are not well studied.At the same time,in practical applications,more dishes are displayed in the same image.This paper conducts research on this situation,and the main research work is as follows:Firstly,in view of the current situation that there is no large multi-objective Chinese dish detection data set,in order to verify the performance of the Chinese dish image detection network proposed in this paper,this paper created a Chinese dish detection data set chinesefood-50 containing 50,000 images.Secondly,considering the multi-scale features of multi-objective Chinese dish images,a Chinese dish detection network based on feature fusion is proposed to complete the multi-objective Chinese dish detection task.The overall design of the algorithm is based on the recursive idea,and the recursive pyramid network and cascade multi-level detector are adopted to gradually improve the detection effect of the detector.Compared with the existing target detection network,it has more accurate positioning effect and higher recognition accuracy.At the same time,in order to better extract more discriminative dish image features,an attention network structure based on global-local features was proposed to complete the task of Chinese dish recognition.The whole network consists of two sub-networks,which respectively focus on the global features and local features of Chinese food images,and then fuse the visual features extracted from the two sub-networks to obtain the final feature representation.Experimental results show that the proposed algorithm is better than the existing algorithms.Finally,the above dish recognition algorithm is packaged as a service to create a nutritional diet management system for athletes.The system includes two parts:web terminal and mobile terminal,which is convenient for athletes and dietitians to fully understand their daily nutrition intake and reasonably arrange daily dietary intake. |
