FOOD RECOGNITION USING DEEP CONVOLUTIONAL NEURAL NETWORK
0 likes171 views
This document describes a study that uses a convolutional neural network (CNN) to recognize food images and estimate nutrition information. The researchers trained their CNN model on the Food-101 dataset, which contains 101,000 images across 101 food categories. They preprocessed the images and used convolutional, pooling, and fully connected layers in their CNN architecture. Their proposed model achieved over 80% accuracy in classifying food images from the test dataset. The study demonstrated that CNNs can accurately recognize food images and potentially provide nutrition information, which could help with dietary analysis and health monitoring. Future work includes recognizing nutrition quantities based on food amounts.
More Related Content
What's hot (20)
Similar to FOOD RECOGNITION USING DEEP CONVOLUTIONAL NEURAL NETWORK (20)
More from IRJET Journal (20)
Recently uploaded (20)
FOOD RECOGNITION USING DEEP CONVOLUTIONAL NEURAL NETWORK
- 1. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 09 Issue: 01 | Jan 2022 www.irjet.net p-ISSN: 2395-0072 © 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 174 FOOD RECOGNITION USING DEEP CONVOLUTIONAL NEURAL NETWORK PRIYANKA N1, Dr. GEETHA V2 1PG Scholar,Department of Electronics and Communication, UBDT College of Engineering, Davanagere-577004, Karnataka, India 2Associate Professor, Department of Electronics and Communication, UBDT College of Engineering, Davanagere- 577004, Karnataka, India ---------------------------------------------------------------------***---------------------------------------------------------------------- Abstract -Food monitoringandnutritional analysisassumes a main part in health-related issues; it is getting more essential in our everyday lives. In this paper, we apply a convolutional neural network (CNN) to the task of detecting and recognizing food pictures and to estimate nutrition in the food. Considering the wide variety of food, image recognition of food items is extremely troublesome. Food- 101 dataset are used for train and test the model, we are using 101 different classes of food images in order to improve accuracy of model. The proposed model provides more than 80% of accuracy. Key Words: convolutional neural network (CNN), food recognition, deep learning, convolution layers, nutrition information 1.INTRODUCTION In an era of mobility, individuals become more conscious about their diet and stay away from coming or existing infections. Accurate assessmentofdietarycalorie estimation is essential for proper analysis of dietary intake. The importance of these individual structures lies in the proper classification of food. Over the past two decades, research has focused on computer vision and artificial intelligence programs to capture images from food and its health data. An automated vision-based traditional meal assessment system based on image analysis starts with four basic steps: food detection, food type classification, Estimating quantity or weight, and finally food information. the development of image processing andobject detection,machinelearning and deep learning methods and their application to convolutional neural networks (CNN) has improved the accuracy of image recognition. In recentyears,CNN hasbeen widely used in food recognition applications, improving traditional machine learning methods. 1.1 Objectives The main objective of this is to recognize and detecting food. 1.To implement suitable Preprocessing of Image. 2.To Implement suitable method for Feature extraction 3.To design and implement suitable method for classification of food using CNN. 1.2 CNN architecture The convolutional neural network consists of the input layer, the invisible layer, and the output layer. In the convolutional neural network, the middle layer is called the invisible layer. Convolutional layers On CNN, the input takes the form of a tensor: (number of inputs) x (input height) x (input width) x (input channel). After going through the spiral layer, the image is summarized in a feature map called the Activation Map. The form is as follows. (Number of entries) x (height of feature map) x (width of feature map) x (feature map channel). Within the spiral layer. Pooling layers Convolutional networks may include convolutional layersas well as general and global layers. The next layer of a single neuron is a set of neurons that combine the results of the data with the size of the group layer. Fully connected layer The fully connected layer connects one layer of each neuron to another neuronal layer. This is similar to the old- style multilayer perceptron (MLP) neural net. Flat Medium classifies images through fully linked layers. Fig1: Basic CNN Layer Structure
- 2. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 09 Issue: 01 | Jan 2022 www.irjet.net p-ISSN: 2395-0072 © 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 175 2. SYSTEM DESIGN convolution neural networks (CNN) areusedtodifferentiate the food images based on their features. Food datasets are utilized to prepare and assess CNN models. Keras is used to implement trained model. Fig 2: Block Diagram of System Design 2.1 Dataset: Here we provide a food-101 dataset it contains 101 different classes of food images. we are using 101000 images to train the model. Images were taken from publicly available Internet resources, The image consists of a single foodstuff to improve the appearance, rotation, color, and accuracy of complexity recognition. Fig 3: Proposed Dataset Food Categories 2.2 Pre-processing First step involves preparing food dataset. The datasets in the form of images taken from food-101 dataset. They have 101000 images belongs to 101 classes. Fig 4: Block Diagram of Proposed System The next step is a pre-trained CNN model and a perfect training. After the successful training and learning phase of the system, the classification phase begins. Classification is also done by users. 2.2 RESULTS We embarked our study by implementing the CNN architecture. We modify the parameters such as choosing deeper CNN architecture, divisionofdatasetintotrainingand testing, and the number of epochs to train the model.
- 3. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 09 Issue: 01 | Jan 2022 www.irjet.net p-ISSN: 2395-0072 © 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 176 The proposed classification and characteristic extraction method achieve a high degree of accuracy of 80%. We also describe possible and future improvements to improve system usability and accuracy. 3. CONCLUSION In this we have implemented convolution neural network to identifying food images and nutrition information. we have presented dataset from food-101 for classifying images. we can classify different images among 101 classes of image which were trained. then identify nutrition information for each image for each training andtestingwegotabout 80%of accuracy. For the future work have to recognize nutrition according to the quantity of each food item. REFERENCES 1.H. Kagaya and K. Aizawa, “Highly accuratefood/non-food image classification based on a deep convolutional neural network,” in International ConferenceonImage Analysis and Processing, 2015, pp. 350–357. 2. Gianluigi Ciocca, Paolo Napoletano, and Raimondo Schettini” Food Recognition: A N e w Dataset, Experiments, and Results” IEEE 2017. 3. Shamay Jahan, Shashi Rekha. H, Shah Ayyub Quadri “Bird’s Eye Review on Food Image Classification using Supervised Machine Learning” IJLTEMAS 2018 4. Amatul Bushra Akhil, Farzana Akter Tania Khatun & Mohammad Shorif Uddin “Recognition and Classification of Fast Food Images” Global Journal of Computer Science and Technology 2018. 5. W R SAM EMMANUEL and S JASMINE MINIJA” Fuzzy clustering and Whale-based neural network to food recognition and calorie estimation for daily dietary assessment” Sådhanå 2018. 6. Akshada Gade, Dr. Arati Vyavahare “Dietary Assessment Methods Based on Image Processing: A Review” International Journal ofInnovativeResearchinScience, Engineering and Technology 2017.