Research And Development On The Food Recognized Mobile System Based On Deep Perception Learning

As the saying goes," Food is the paramount necessity of the people" food is crucial in people's daily life,and diet is closely related to people's life and health.With the advancement of time and the development of technology,people's demand for food has not just eliminated hunger,using modern technology to improve the quality of diet and the standard of living has been the target of modern people.In addition,for the patients with hypertension and hyperglycaemia,the content of diet is related to their daily life,even their life and health.Therefore,the research on food recognition technology has extremely important research value and practical value in current era,and a food recognition system also has a strong market prospect.In order to solve the problem of food recognition,people can better control the daily intake of food.This paper designs and implements a mobile food recognition system for food images.In order to realize the real-time recognition system of food,this paper uses deep learning technology to divide the project into three modules for implementation.Firstly,the food recognition model is designed and trained,and the concept of "jumping convolution" is proposed to learn the features of food images.Later,in order to enhance the stability and accuracy of recognition,the multi-network fusion method is used to improve the accuracy and fault tolerance of food recognition algorithm.Finally,food recognition system has been implemented on the Android mobile phone.This paper mainly includes the following contents*:1)As for the food recognition problems,this paper designs and implements a new convolution neural network.This network uses “Jumping convolution” to learn feature from food images and get the food recognition network.Aiming at the characteristics of food images,this paper also proposes a set of preprocessing process for food images.It can effectively reduce the interference factors in the training process for problems such as complex image background and occlusion of image text.Experimental results show that the food recognition algorithm designed in this paper can effectively identify food targets and achieve high recognition accuracy.2)In the process of food recognition with a single network,it is unavoidable that there will be recognizition errors.In order to enhance the fault-tolerance of the recognition network,this paper uses Boosting ensemble learning to train the results of a single network by repeating the misdivided images,training out different recognition models,and then merging the network models to obtain the final result.This method can effectively improve the recognition accuracy of the food recognition system.On the basis of Boosting ensemble learning,this paper proposes a numberreduction Boosting network ensemble strategy to improve the system fault-tolerance rate,while increasing the final output accuracy of the system by limiting the number of learning times for the same image.The experimental results show that the ensemble strategy has good effect on food identification,and the error rate and accuracy rate have some improvement.3)Based on the research of food recognition theory,this paper implements the food recognition system on Android mobile phone.This work includes the transplantation and debugging of convolutional neural networks,and the establishment of the food recognition system in Android Studio.One more thing,in order to improve the user experience,the food text recognition functions and food-related information search functions were added to the food recognition system.The Android OCR technology and the Android network communication technology were respectively used.The system can basically meet the needs of people in daily life to identify food.