Nowadays, the trend of consuming clean food in Thai society become very popular since everyone pay more attention on their health especially elderly who need to focus on the amount of calories and the diversity of nutrients that they take for each meal. So we come with an idea that wouldn’t it be great if we create an AI that can classify and localize ingredients, and also calculate the amount of nutrients in Thai dishes. Overview FOOD ANALYSIS SYSTEM USING A DEEP LEARNING TECHNIQUE Related Theory Convolutional Neural Networks (CNNs) In deep learning, a convolutional neural network (CNN, or ConvNet) is a class of deep neural networks, most commonly applied to analyzing visual imagery. CNNs are regularized versions of multilayer perceptrons. Multilayer perceptrons usually refer to fully connected networks, that is, each neuron in one layer is connected to all neurons in the next layer. The "fully-connectedness" of these networks make them prone to overfitting data. Typical ways of regularization includes adding some form of magnitude measurement of weights to the loss function. However, CNNs take a different approach towards regularization: they take advantage of the hierarchical pattern in data and assemble more complex patterns using smaller and simpler patterns. Therefore, on the scale of connectedness and complexity, CNNs are on the lower extreme. Experiments Secondly, the ingredients localization networks was trained by feeding 400 labeled images into You Only Look Once (YOLO) networks which is a neural network that works efficiently when it comes to classification and localization. Finally, the food classification server and mobile application were created in order to make our project more accessible. By using image from mobile application as an input and send the image to server, then the server will return all of the important information back to mobile application. Results of our Project Conclusion Firstly, the food recognition networks was trained by feeding 400 images in order to train our networks. We use tensorflow which is an open source library to achieve this problem Food is one of the most important things for human life but traditional nutrition estimation via comparison book has limitation and hard to be understand for normal people. Food types are categorized and ingredients for a specific food dish are importance especially for elderly or diabetics. This research proposed the easy way to determine nutrition by using Convolutional Neural Networks (CNNs). MobileNet network is selected as food categorization network and YOLO v3 tiny network is selected as ingredients classifier and localizer. In order to make our project more accessible, food classification and mobile application were created to make user feels more convenience. By using application on mobile phone instead of carrying hardware along with them. Project Authors : Project Advisor : Kittinon Sawanyawat 5810554300 Dr. Kanjanapan Sukvichai Warayut Muknumporn 5810552722 Faculty of Electrical Engineering Kasetsart University

FOOD ANALYSIS SYSTEM USING A DEEP LEARNING TECHNIQUE

Download PDF Report

Upload
others
View
0
Download
0

Embed Size (px)

Citation preview

Overview

FOOD ANALYSIS SYSTEM USING A DEEP LEARNING TECHNIQUE

Related Theory

Convolutional Neural Networks (CNNs)In deep learning, a convolutional neural network (CNN, or ConvNet) is a class of deep neural

networks, most commonly applied to analyzing visual imagery.CNNs are regularized versions of multilayer perceptrons. Multilayer perceptrons usually refer to fully

connected networks, that is, each neuron in one layer is connected to all neurons in the next layer. The"fully-connectedness" of these networks make them prone to overfitting data. Typical ways ofregularization includes adding some form of magnitude measurement of weights to the loss function.However, CNNs take a different approach towards regularization: they take advantage of the hierarchicalpattern in data and assemble more complex patterns using smaller and simpler patterns. Therefore, onthe scale of connectedness and complexity, CNNs are on the lower extreme.

Experiments

Secondly, the ingredients localizationnetworks was trained by feeding 400labeled images into You Only Look Once(YOLO) networks which is a neural networkthat works efficiently when it comes toclassification and localization.

Finally, the food classification server and mobileapplication were created in order to make our projectmore accessible. By using image from mobile applicationas an input and send the image to server, then the serverwill return all of the important information back to mobileapplication.

Results of our Project Conclusion

Firstly, the food recognition networkswas trained by feeding 400 images in orderto train our networks. We use tensorflowwhich is an open source library to achievethis problem

Food is one of the most important things for human lifebut traditional nutrition estimation via comparison book haslimitation and hard to be understand for normal people. Foodtypes are categorized and ingredients for a specific food dishare importance especially for elderly or diabetics. Thisresearch proposed the easy way to determine nutrition byusing Convolutional Neural Networks (CNNs).

MobileNet network is selected as food categorizationnetwork and YOLO v3 tiny network is selected as ingredientsclassifier and localizer.

In order to make our project more accessible, foodclassification and mobile application were created to makeuser feels more convenience. By using application on mobilephone instead of carrying hardware along with them.

Project Authors: Project Advisor:

Kittinon Sawanyawat 5810554300 Dr. Kanjanapan SukvichaiWarayut Muknumporn 5810552722

Faculty of Electrical Engineering Kasetsart University