Deep convolutional networks for food detection and classification

Mohammed A. Subhi, Sawal Hamid Md Ali, Mohammed Abdulameer

Research output: Contribution to journalArticle

Abstract

Automatic food image recognition systems have been the focus of recent works in the field of dietary assessment. The accuracy of such systems alleviates the process of estimating the nutrition information. However, due to the nature of food images, detection of multiple food items in the same scene is particularly a challenging task. For this reason traditional machine learning approaches achieved low classification accuracy when food items are involved. In the recent years, deep convolutional neural networks (DCNN) have outperformed existing solutions by introducing better feature extraction structure and more accurate classification strategy. In this paper, we implement deep learning classification and detection method using Google's machine learning and Tensor flow library. For classification purposes, the images were collected from existing datasets including 2000 categories of (Food/Non-food) classes. When the food is identified within the image, multiple CNN object detection models are applied to recognize individual food items within a single image. Since food detection requires more sophisticated training and annotation, selected food images were used and manually annotated for this purpose. The testing results show that all three models are capable of detecting individual food items within a single image with high accuracy.

Original languageEnglish
Pages (from-to)2433-2438
Number of pages6
JournalJournal of Computational and Theoretical Nanoscience
Volume16
Issue number5-6
DOIs
Publication statusPublished - 1 Jan 2019

Fingerprint

food
Machine Learning
machine learning
Image Recognition
Nutrition
Object Detection
Learning systems
Feature Extraction
Annotation
High Accuracy
nutrition
annotations
Tensor
Image recognition
Neural Networks
Testing
pattern recognition
learning
Tensors
Feature extraction

Keywords

  • Deep convolutional neural networks
  • Deep learning
  • Food classification
  • Food detection
  • Food image dataset

ASJC Scopus subject areas

  • Chemistry(all)
  • Materials Science(all)
  • Condensed Matter Physics
  • Computational Mathematics
  • Electrical and Electronic Engineering

Cite this

Deep convolutional networks for food detection and classification. / Subhi, Mohammed A.; Md Ali, Sawal Hamid; Abdulameer, Mohammed.

In: Journal of Computational and Theoretical Nanoscience, Vol. 16, No. 5-6, 01.01.2019, p. 2433-2438.

Research output: Contribution to journalArticle

@article{98b54cae4d2b45f192693289771abdec,
title = "Deep convolutional networks for food detection and classification",
abstract = "Automatic food image recognition systems have been the focus of recent works in the field of dietary assessment. The accuracy of such systems alleviates the process of estimating the nutrition information. However, due to the nature of food images, detection of multiple food items in the same scene is particularly a challenging task. For this reason traditional machine learning approaches achieved low classification accuracy when food items are involved. In the recent years, deep convolutional neural networks (DCNN) have outperformed existing solutions by introducing better feature extraction structure and more accurate classification strategy. In this paper, we implement deep learning classification and detection method using Google's machine learning and Tensor flow library. For classification purposes, the images were collected from existing datasets including 2000 categories of (Food/Non-food) classes. When the food is identified within the image, multiple CNN object detection models are applied to recognize individual food items within a single image. Since food detection requires more sophisticated training and annotation, selected food images were used and manually annotated for this purpose. The testing results show that all three models are capable of detecting individual food items within a single image with high accuracy.",
keywords = "Deep convolutional neural networks, Deep learning, Food classification, Food detection, Food image dataset",
author = "Subhi, {Mohammed A.} and {Md Ali}, {Sawal Hamid} and Mohammed Abdulameer",
year = "2019",
month = "1",
day = "1",
doi = "10.1166/jctn.2019.7913",
language = "English",
volume = "16",
pages = "2433--2438",
journal = "Journal of Computational and Theoretical Nanoscience",
issn = "1546-1955",
publisher = "American Scientific Publishers",
number = "5-6",

}

TY - JOUR

T1 - Deep convolutional networks for food detection and classification

AU - Subhi, Mohammed A.

AU - Md Ali, Sawal Hamid

AU - Abdulameer, Mohammed

PY - 2019/1/1

Y1 - 2019/1/1

N2 - Automatic food image recognition systems have been the focus of recent works in the field of dietary assessment. The accuracy of such systems alleviates the process of estimating the nutrition information. However, due to the nature of food images, detection of multiple food items in the same scene is particularly a challenging task. For this reason traditional machine learning approaches achieved low classification accuracy when food items are involved. In the recent years, deep convolutional neural networks (DCNN) have outperformed existing solutions by introducing better feature extraction structure and more accurate classification strategy. In this paper, we implement deep learning classification and detection method using Google's machine learning and Tensor flow library. For classification purposes, the images were collected from existing datasets including 2000 categories of (Food/Non-food) classes. When the food is identified within the image, multiple CNN object detection models are applied to recognize individual food items within a single image. Since food detection requires more sophisticated training and annotation, selected food images were used and manually annotated for this purpose. The testing results show that all three models are capable of detecting individual food items within a single image with high accuracy.

AB - Automatic food image recognition systems have been the focus of recent works in the field of dietary assessment. The accuracy of such systems alleviates the process of estimating the nutrition information. However, due to the nature of food images, detection of multiple food items in the same scene is particularly a challenging task. For this reason traditional machine learning approaches achieved low classification accuracy when food items are involved. In the recent years, deep convolutional neural networks (DCNN) have outperformed existing solutions by introducing better feature extraction structure and more accurate classification strategy. In this paper, we implement deep learning classification and detection method using Google's machine learning and Tensor flow library. For classification purposes, the images were collected from existing datasets including 2000 categories of (Food/Non-food) classes. When the food is identified within the image, multiple CNN object detection models are applied to recognize individual food items within a single image. Since food detection requires more sophisticated training and annotation, selected food images were used and manually annotated for this purpose. The testing results show that all three models are capable of detecting individual food items within a single image with high accuracy.

KW - Deep convolutional neural networks

KW - Deep learning

KW - Food classification

KW - Food detection

KW - Food image dataset

UR - http://www.scopus.com/inward/record.url?scp=85068583928&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85068583928&partnerID=8YFLogxK

U2 - 10.1166/jctn.2019.7913

DO - 10.1166/jctn.2019.7913

M3 - Article

VL - 16

SP - 2433

EP - 2438

JO - Journal of Computational and Theoretical Nanoscience

JF - Journal of Computational and Theoretical Nanoscience

SN - 1546-1955

IS - 5-6

ER -