You are on page 1of 7

A Survey on Nutrition Monitoring and Dietary

Management Systems

Abstract - A balanced diet with approximate nutrient Food-100, and UEC Food-256. The current dataset
intake is vital for infants and children which reduces related to food images have varied aspects namely
the risks of life-threatening diseases such as obesity, the number of food item groups, the total number
cardiovascular disease and cancer. Unlike adults,
of images for a specific food class and also the type
infants require some assistance in their food intake.
of cuisine [9, 10].
The survey provides valuable insights about the
advancements of IoT in healthcare and the need for B. Image Based Food Recognition
nutrition monitoring. Various nutrition monitoring
a. Food image segmentation
systems for calorie estimation, prediction have been
developed using machine learning techniques and also Segmentation is the process of decomposition of an
with advanced deep learning based techniques. A
image into a number of non-overlapping
comparative view of the previous works of
researchers in the recent times has been provided.
meaningful areas with the similar attributes. When
the complexity and difficulty are considered, the
Keywords – Nutrition Monitoring, Machine Learning, various segmentation algorithms have been
Neural networks, Deep Learning. successful in varied measures. The Clustering
analysis algorithm which has a wide application in
I. INTRODUCTION
image segmentation, divides the datasets into
Monitoring the daily intake of nutrients and different groups in accordance to a particular
ensuring a balance is very vital for the growth of standard. The food images are analysed to extract
children. Infant child mortality and morbidity, for the food items before the food classification [11-
children under the age of five is mainly due to 13].
undernutrition. Almost 90% of the children
b. Feature extraction
suffering from undernutrition are from the
developing countries [1]. Fulfilling the balanced Feature extraction is a dimensionality reduction
nutrition requirement of the children helps to process, where an initial set of raw variables is
reduce malnutrition and reduces the risk of diseases reduced to more manageable groups (features) for
that are encountered due to lack of nutrition. The processing, while still accurately and completely
IoT is an enabling technology which consists of describing the original dataset. The image can be
things connected in a network to communicate and recognized based on characteristics like colour,
share data. The life of the user becomes shape and the texture [14]. The selection of
comfortable by adopting the various technologies relevant features is important when a recognition
based on IoT [2]. The Internet of Things (IoT) model is built for identifying the food images [15].
provides a probable solution to suit the demands of
improving the healthcare systems and is also the c. Food classification
main focus of major research works [3-5]. The IoT
Food classification strategies for image recognition
has great impact in the healthcare industry and
are the machine learning based approach that
improves the lives of millions of people all over the
consists of features and the deep learning approach.
world enabling remote assistance. The IoT has the
The Machine learning based approach consists of
capability to safeguard the health of a patient but
feature extraction and use them to construct a
also helps the physicians to ensure care [6, 7]. An
prediction model based on algorithms like Support
analysis was conducted in India on the various
Vector Machine (SVM), Bag of Features and K
features that affect infant mortality rate in the
Nearest Neighbours. In contrast, the emerging deep
Indian states [8].
learning approach has a large number of connected
A. Food Image Databases layers that learn features, followed by a final layer
responsible for classification. Deep Learning based
A comprehensive collection of quality food images approaches have become more popular and
is key to train a food-classification model and effective.
benchmark the prediction performance on large
food image databases such as Food-101, UEC
d. Food volume estimation and nutrient The system takes an image of the food
analysis item and passes it through Mathworks
Image Processing which extracts the raw
After the food items have been identified, the features and improves the quality. The
nutrient content has to be analysed. At first, the image is passed through a compression
food candidate regions must be recognized, phase which helps to reduce the number of
segmented, and classified correctly. The volume features using the Principal Component
from each segmented item will be calculated; and Analysis (PCA) method and scale the
the nutrient can be estimated based on a nutritional subsequent learning phases. The food type
facts table, such as USDA Food Composition classification is done by inputting the
Database. The volume can either be underestimated compressed image to the classifier. The
or overestimated with interference from external food size prediction is done by passing the
factors, such as lighting conditions, blurred images, compressed image to a regressor. Calories
and noisy background [16-18]. Only a very few are predicted by passing the compressed
strategies were reported for estimation of food image and predicted values to another
volume and calorie intake, as presently the regressor. This is based on supervised
important focus of this domain still lies in the focus learning model. The dataset is limited to a
of food classification [19]. small category and the image cannot be
II. REVIEW OF LITERATURE diversified.
C. Kohila et al. [22] proposed a calorific
A. Prabha et al. [20] proposed a Smart Log value prediction mechanism using image
system that performs automated nutrition processing and machine learning. The
monitoring and meal prediction. The smart image of the food is transmitted through a
sensor board consisting of Piezo Electric mobile device and it initially undergoes
sensors is used for nutrition quantification. segmentation with Fuzzy C-means
The nutrient data acquisition is done using Clustering Segmentation which fixes the
Optical Character Recognition and by cluster centre based on the group data
linking open source Application Program unlike the K-means Clustering which can
Interfaces (APIs) through barcodes. The be erroneous if the cluster centre is not
meal prediction is done by collecting defined properly by the user. The
nutritional value of the leftover food along mathematical morphology is utilized as a
with the user’s feedback on the type of tool for extracting the image components
food that is desired. The SR8 database and the region shape description such as
available through the US Department of erosion, dilation, opening and closing.
Agriculture website is also analysed using Feature extraction is performed to retrieve
their API which provides a food report of interesting parts of the image and then
associated nutrient values for a particular calorie measurement is done. It has
food item and a nutrient report which limited scalability and diversely mixed
gives an extensive list of food and their food images have not been considered.
nutrient values for a selected amount of
nutrients. The results have been analysed D. Sangita et al. [23] proposed a nutritional
by creating an Attribute-Relation File status investigation system based on
Format which inputs the Waikato machine learning. A logical regression
Environment for Knowledge Analysis model was considered for the major four
(WEKA) tool which builds a better variables namely BMI, HAZ, WHZ and
prediction model and is observed that the WAZ individually. The study predicts the
Bayesian classifiers provided better nutrition state of the child in two phases.
results. The open dataset consisted of Phase I pre-processes the dataset using
multiple redundant logs and psychological SMOTE Resampling method. Feature
monitoring mechanisms have not been extraction is done using machine learning
incorporated which in turn leads to lack of techniques with Entropy based Gain Ratio
accurate prediction. concept. Phase II uses Nominal logistic
regression using iterative reweight least
B. Manal et al. [21] proposed a machine squares algorithm which predicts the
learning based pipelined approach for characteristic features. The dataset is
predicting the calories from food images. based on a very specific geographical area
and over a particular period of time and it phone camera captures the FDA-mandated
also considers only the basic features. Nutritional Facts Label and the other
method is by linking open source
E. Oscar et al. [24] proposed a menu-match: Application Program Interfaces (APIs)
restaurant-specific food logging from through barcodes. The meal prediction is
images. An image recognition framework done by collecting nutritional value of the
based on the bag of visual words approach leftover food along with the user’s
which extracts the base features from the feedback on the purpose of the meal. Food
images and then encoded with locality- Classification is done through an
constrained linear coding (LLC). The algorithm which is built on a Bayesian or
extracted features are pooled using max- Belief Network (BN) with constraint
pooling in a rotation-invariant pooling based method. The algorithms used are
scheme. A regression based method Nutritional Balance Network Algorithm
estimates the calories and along with for building a multi-layer perceptron
feature representation mapped the feature neural network using Stochastic Gradient
space to calories using Support Vector Descent and Food classification algorithm
Regression. The approach is limited for for building Bayesian Network using hill-
discrete serving sizes and custom menu climbing search. The system is designed
and is also dependent on the GPSS of food only for the children and the dataset
consumption. The system lacks user contained multiple redundant logs.
customization and requires cost-sensitive
learning to directly minimize calorie H. Parisa et al. [27] proposed a method for
estimation errors during the training. food calorie measurement using deep
learning neural network. The system uses
F. Kiran et al. [25] proposed a method for food images and the features are extracted
measuring the calories and nutrition from using Graph Cut Segmentation. A pre-
food images using machine learning trained model file is generated with the
techniques. The images got from the help of Convolutional Neural Network
mobile device are pre-processed followed (CNN). It is loaded into the application
by the segmentation step to extract the and tested to perform image recognition
colour and texture features through K process. The stochastic gradient descent
Means clustering. The extracted options algorithm is used to get an improvement
are used for food classification using in the cost. The calorie measurement is
Support Vector Machine (SVM). The food done by two approaches namely finger
portion volume measurement is done by based calorie measurement and calorie
superimposing a grid of squares onto the measurement using distance estimation.
image segment which matches the The system lacked analysing mixed food
irregular shape of the food images easily. portion and only limited samples were
The calorie measurement is done based on considered.
the food mass and nutritional tables. The
system has limited cuisine varieties mixed III. COMPARISON OF RESEARCH WORKS
food images have not been considered. TABLE 1: COMPARISON OF THE RESEARCH
WORKS
G. Prabha et al. [26] proposed a Smart Log
system based on deep learning for Title Approach Advantag Limitation
automated nutrition monitoring system in es
IoT. The system can be considered as a Prabh Machine Automate Psychological
product which includes a smart sensor a et learning d nutrition monitoring
board with a food weighing sensor and a al. based monitorin mechanism is
smart phone application. Automatic [20] prediction g and not
using OCR meal incorporated
Nutrition Quantification is the first step
and APIs prediction
which collects the quantity of nutrient
Mana Machine The The dataset is
values along with the timestamp and l et al. learning pipelined limited to a
transmits it to the cloud. The nutrient data [21] based approach small category
acquisition is done using Optical pipelined is very and lacks
Character Recognition in which the on-
approach effective realistic Kohil Machine The Lacks mobility
using PCA when scenario a et learning approach and there is a
compared al. based Fuzzy has deviation in
to the [22] c-means efficient calorific value
baseline clustering feature between the
for extraction observed and
segmentatio mechanis calculated
n and m values
Morphologi
cal operation
for
extracting
image
components.
Sangi Machine Explored The data is
ta et learning the limited to a
al. with possibiliti particular time,
[23] Entropy es of context and
based Gain using only limited
Ratio artificial features are
concept and intelligenc considered.
Nominal e in
logistic identifyin
regression g probable
using correlates
iterative of
reweight malnutriti
least squares on.
algorithm
Oscar Machine Automate It lacks user
et al. learning d customization
[24] based SVM computer and is GPSS
classifier vision dependent
and LLC system for
logging
food and
robust
calorie
estimation
Kiran Machine IMg2 Has limited
et al. learning calories cuisine
[25] based K app that varieties and
Means determine mixed food
clustering s the images have
and SVM carlorie not been
intake and considered.
estimation
.
Prabh Deep The Psychological
a et Learning prediction monitoring
al. based novel accuracy mechanism is
[26] 5-layer of Smart- not
perceptron Log is incorporated
neural 98.6%.
network and
a Bayesian
Network
based
prediction
algorithm
Parisa Deep The The database [6] Catarina I.Reis,Marisa da silva Maximiano,
et al. Learning accuracy is limited and “Internet of Things and Advanced Application in
[27] based for single mixed food Healthcare”, Oct 2016, ISBN13: 9781522518204.
Convolution food portions have
al neural portions is not been [7] P. Sundaravadivel, E. Kougianos, S. P.
networks 99% considered. Mohanty, and M. Ganapathiraju, “Everything You
with Wanted to Know about Smart Healthcare,” IEEE
virtualizatio Consum. Electron. Mag., vol. 8, no. 1, pp. 18–28,
n January 2018.

[8] Suriyakala Vijayakumar, Deepika M G,


IV. CONCLUSION Amalendu Jyotishi and Deepa Gupta, “Factors
Affecting Infant Mortality Rate in India: An
The Internet of Things is a booming technology
Analysis of Indian States”, International
that spans across various domains and its
Symposium on Intelligent Systems Technologies
incorporation with healthcare is increasing day to
and Applications, ISTA; Advances in Intelligent
day. This survey presents a comparative view on
Systems and Computing, Volume 530, Pages 707-
the research works in the nutrition monitoring,
719, Jaipur; India. Publisher Springer, 2016.
calorie estimation and dietary systems. The
machine learning techniques are being replaced by [9] Ciocca G, Napoletano P, Schettini R., “Food
the deep learning techniques that outperforms the recognition: a new dataset, experiments and
traditional methodologies. There is also an increase results”, IEEE Journal Biomed Health Informatics,
in the integration of wearable sensors along with vol.99, pp. 1-1, 2016.
smartphone through mobile technologies which
will be a revolution in food monitoring systems [10] Matsuda Y, Hoashi H, Yanai K., “Recognition
which helps in disease prediction and prevention. of multiple-food images by detecting candidate
regions”, IEEE International Conference on
REFERNCES Multimedia Expo vol.25, pp. 25-30, 2012.
[1] Markos Z, Doyore F, Yifiru M, Haidar [11] Xin Zheng, Qinyi Lei, Run Yao, Yifei Gong
J,“Predicting Undernutrition Status of Under-Five and Qian Yin, “Image Segmentation based on
Children Using Data Mining Techniques: The Case adaptive K-means algorithm”, EURASIP Journal
of 2011 Ethiopian Demographic and Health on Image and Video Processing, Publisher
Survey”, Journal Health Medical Information, 5: Springer, 3 August 2018.
152, 2014.
[12] Hamirul Binti Hambali, Salwa Khalid
[2] Sajjad Hussain Shah, Ilyas Yaqoob, “A Survey: Abdulateef and Massudi Mahmuddin,
Internet of Things (IOT) Technologies, “Segmentation of Multi Food Images Using
Applications and Challenges”, 4th IEEE Integrated Active Contour and K-means”, Journal
International Conference on Smart Energy Grid of Engineering and Applied Sciences, pp. 3146-
Engineering, 2016. 3151, 2018.
[3] P. Gope and T. Hwang, “BSN-Care: A Secure [13] Yu Wang, Fengqing Zhu, Carol J. Boushey
IoT Based Modern Healthcare System Using Body and Edward J. Delp, “Weakly Supervised Food
Sensor Network,” IEEE Sensors Journal, vol. 16, Image Segmentation Using Class Activation
no. 5, pp. 1368–1376, 2016. Maps”, Proceedings of IEEE International
Conference on Image Processing, pp. 1277-1281,
[4] N. Zhu, T. Diethe, M. Camplani, L. Tao, A.
2017.
Burrows, N. Twomey, D. Kaleshi, M. Mirmehdi, P.
Flach, and I. Craddock, “Bridging e-Health and the [14] Hong Liang, Yuan Gao, Yunlei Sun and Xiao
Internet of Things: The SPHERE Project,” IEEE Sun, “CEP: Calories Estimation from food photos”,
Intelligent Systems, vol. 30, no. 4, pp. 39–46, 2015. International Journal of Computers and
Applications, 25 July 2018.
[5] S. H. Chang, R. D. Chiang, S. J. Wu, and W. T.
Chang, “A Context-Aware, Interactive M-Health [15] Hamid Hassannejad, Guido Matrella, Paolo
System for Diabetics,” IT Professional, vol. 18, no. Ciampoini, Ilaria De Munari, Monica Mordoini,
3, pp. 14–22, 2016. Stefano Cagnoni, “Food Image Recognition Using
very Deep Convolutional Networks”, Proceedings
of the 2nd International Workshop on Multimedia
Assisted Dietary Management, Amsterdam, The processing”, In Proceedings of the 2017
Netherlands, 16 October 2016. International Conference on Innovations in
Information, Embedded and Communication
[16] Jia W., Li Y., Qu R., Baranowski, T., Burke, Systems (ICIIECS), Coimbatore, India, pp. 1–4,
L., Zhang, H., Sun, M., “Automatic food detection 17–18 March 2017.
in egocentric images using artificial intelligence
technology”, Public Health Nutrition, pp. 1-12, [23] Sangita Khare, Kavyashree, S., Gupta, D.,
2018 doi:10.1017/S1368980018000538. Jyotishi, A., “Investigation of nutritional status of
children based on machine learning techniques
[17] Sirichai Turmchokkasam, Kosin using Indian demographic and health survey data”,
Chamnongthai, “The Design and Implementation In: 7th International Conference on Advances in
of an Ingredient-Based Food Calorie Estimation Computing and Communications, vol. 115, pp.
System Using Nutrition Knowledge and Fusion of 338–349. Elsevier Procedia Computer Science
Brightness and Heat Information”, IEEE Access, 2017. https://doi.org/10.1016/j.procs.2017.09.087
vol. 6, pp. 46863 – 46876, 15 May 2018.
[24] Oscar Beijbom , Neel Joshi , Dan Morris ,
[18] İbrahim Berkan Aydilek, “Approximate Scott Saponas , Siddharth Khullar, “Menu-Match:
estimation of the nutrition of consumed food by Restaurant-Specific Food Logging from Images”,
deep learning”, Proceedings of IEEE International Proceedings of the 2015 IEEE Winter Conference
Conference on Computer Science and Engineering on Applications of Computer Vision, pp. 844-851,
(UBMK), 02 November 2017. 05 - 09 January, 2015.
[19] Silva BVR, Cui J., “A Survey on Automated [doi>10.1109/WACV.2015.117]
Food Monitoring and Dietary Management [25] Kiran Abhore, Dawande N. A., “Measuring
Systems”, J Health Med Informat 8:272, 2017, doi: Calories and Nutrition from Food Image”,
10.4172/2157-7420.1000272 International Journal of Advanced Research in
[20] Prabha Sundaravadivel, Kavya Kesavan, Computer and Communication Engineering, vol. 5,
Lokeshwar Kesavan, Saraju P. Mohanty, Elias Issue 6, June 2016.
Kougianos, Madhavi Ganapathiraju, “Smart-log: [26] Prabha Sundaravadivel, Kavya Kesavan,
An automated, predictive nutrition monitoring Lokeshwar Kesavan, Saraju P. Mohanty, and Elias
system for infants through the IoT”, IEEE Kougianos, “Smart-Log: A Deep-Learning based
International Conference on Consumer Electronics Automated Nutrition Monitoring System in the
(ICCE), 2018. IoT”, IEEE Transactions on Consumer Electronics,
[21] Manal Chokr, Shady Elbassuoni, “Calories 2018. DOI 10.1109/TCE.2018.2867802
Prediction from Food Images”, Innovative [27] Parisa Pouladzadeh, P. Kuhad, S. V. B. Peddi,
Applications of Artificial Intelligence Twenty - A. Yassine, S. Shirmohammadi, "Food calorie
Ninth IAAI Conference, 2017. measurement using deep learning neural network",
[22] Kohila R., Meenakumari R., “Predicting Proc. IEEE Int. Instrum. Meas. Technol. Conf.
calorific value for mixed food using image Proc., pp. 1-6, 2016.

You might also like