Food Assessment Model for Indian Elderly Persons Using CNN and Image Processing Techniques Leena

K. Gautam and Vijay S. Gulhane Abstract Nutrient requirements differ from person to person and are
influenced by a variety of factors like age, physical activity, diseases, etc. In Indian elderly persons, loss
of apetite, isolation, depression, weak muscles are some of the important factors which not only lead to
malnutrition but also raise the risk of getting affected with other major diseases results in the worsening
of chronic conditions, frequent hospitalization causing burden to caregivers. In this paper, a user-
friendly and intelligent system is proposed that can accurately classify and estimate the food region size
of mixed/cooked food images, which can be further used to obtain calories and nutrient information
utilizing the Indian Food Nutrient Dataset (IFND). The proposed system uses convolution neural network
also known as CNN trained and tested on Indian Food Image Dataset (IFID) and integrated with refine
Image processing techniques for better classification accuracy. Experimental results show that the
adapted method leads to an accuracy of 97%. Keywords Convolutional neural network · Refine image
preprocessing techniques · Estimation of food region Introduction Food and nutrition play a crucial role
in our daily lives. The amount of nutrients required in a certain food item changes depending on age,
physical factors, and other disorders [1]. In elderly persons, isolation, depression, low immunity level,
and weak muscles are added factors determining nutrients requirements [2]. The main issue among
elderly people is a loss of appetite and low immunity, which not only leads to malnutrition but also
increases the chance of contracting other serious L. K. Gautam (B) · V. S. Gulhane Sipna College of
Engineering and Technology, Amravati, Maharashtra, India e-mail: kgautam@sipnaengg.ac.in V. S.
Gulhane e-mail: vsgulhane@sipnaengg.ac.in © The Author(s), under exclusive license to Springer Nature
Singapore Pte Ltd. 2022 V. Bindhu et al. (eds.), Proceedings of Third International Conference on
Communication, Computing and Electronics Systems, Lecture Notes in Electrical Engineering 844,
https://doi.org/10.1007/978-981-16-8862-1_73

diseases, worsening chronic conditions, and delaying recovery [3, 4]. The malnourished elderly are more
likely to require hospitalizations, cause a burden to caregivers, and have a gigantic monetary expense to
society [5]. To lead a healthy and active life, an efficient food assessment system is highly recommended
which classifies and estimates nutrient requirements from the given food image. The method which can
automatically assess, i.e., can classify and estimate the nutrient as well as calorie contents from a food
image allowing a person to learn from what food contains and how healthy it can be. Many researchers
[6–8] had already proved the accepted classification accuracy of the given food images using
convolutional neural network (CNN) using existing databases, while others had performed calorie
estimation of a given food using different tools and techniques [6, 9–14]. In all of these methods, raw
food was segmented using simple segmentation algorithms, which were then assessed for content
estimation using an existing, limited, and/or calorie annotated dataset featuring dishes belonging to that
particular region. Some methods [8, 15] used pre-trained CNN models to classify the food image and a
vector embedding technique to estimate food properties from the image. Estimation of attributes for its
ingredients contents and nutritional values can be efficiently applied to the raw foods but not on mixed
and cooked food. To reduce the computational time of convolutional neural network, region-based
classification called R-CNN was proposed [16]. R-CNN classifies images using a selective region-based
approach but for some data, it might consume more training time. The above mentioned strategies
increase complexities and cause usability issues. Moreover the classification and estimation were
performed using a preexisting monotonous database with some region-based added dishes. Indian
cuisine on the other hand is diverse, consisting of different ingredients with different cooking styles for
preparing the same food dish item. For example, “kanda-poha” is similar to “Indori-poha” and “batata-
poha” with a few minor ingredient differences. Similarly, “upma”, a popular Indian breakfast, is served
differently in different states, with varying ingredient variations and cooking methods and the same is
with “idlis”. Estimation of these commonly cooked Indian food items requires accurate classification,
which can certainly improve its estimation accuracy. Correct classification requires fine-tuned
segmentation approaches, as well as effective training and testing of CNN on a set of food images. In this
work, we propose a CNN-based food assessment system using refine preprocessing techniques for
accurate classification and estimation. The process begins by capturing an image of a food dish from
both, front and side views using a pre-registered reference object and a specified distance, as shown in
Fig. 1. To measure the real size of the food: its area, and depth, commonly used pre-registered reference
objects such as a coin, a cube, a card, and so on can be used. The reference object and the dish regions
are isolating and segmented. After segmentation, the top view of the food shot is utilized to classify it
using CNN, which has been trained and tested on a variety of Indian food images to ensure accurate
classification. Real size area and depth of the food dish are acquired from the top and side views,
respectively, with the help of the registered size of the reference object, and are then examined for
calorie and nutrient estimation

Conclusion and Future Work

In this paper, we proposed a user-friendly dietary assessment model designed specifically for Indian
elders that uses refine image processing techniques to classify images with remarkable accuracy and
also predicts the food region size by calculating area and volume, which can then be used to obtain
calories and nutrient information from IFND. To enchance the classification accuracy of CNN images
were preprocessed with edge detection and GrabCut segmentation techniques. We focused on

the recognition task of cooked and mixed foods and our results indicates 97% of accuracy which is highly
acceptable. Limitation observed in our proposed system is in exceptional circumstances, i.e., when the
color of plate region and food region are same, edge detection fail to distinguish the edges adequately.
In such cases an object detection step might be included before performing edge detection to improve
segmentation results. Future work includes implementation as well as expanding the data set to
encompass a wider range of Indian food images in each class

Dietary Assessment and Nutritional Analysis Using Deep Learnin

S. Madhumitha, M. Magimaa, M. Maniratnam, and N. Neelima

Abstract Nutrition management is an important feature in day-to-day life from prevention of obesity
and chronic diseases related to food intake. To facilitate proper and adequate nutrition intake, suitable
dietary assessment is essential. This dietary assessment method based on food recognition has the
ability to detect the volume, based on the area of the image taken through computer vision. There has
been many modern techniques developed recently on image-based dietary assessment. These
techniques have shown trustworthy results on overcoming issues on epidemiology on dietary studies
and multiple challenges. This model provides a detailed outline of how features are extracted from the
image using computing algorithms, methodology used, and mathematical methods for image
recognition. This model provides volume and mass estimation of the given food based on the area
occupied by the food. From this, the detailed nutritional analysis was calculated for dietary assessment.
The overall accuracy of 94.68% was achieved for classification of food which outperforms the existing
methods.

Keywords Volume estimation · Food detection · Dietary assessment · Nutritional analysis

Introduction The world is hit by one of the deadly viruses, and the entire world has gone into lockdown.
One of the most common issues faced being at home is not having adequate physical exercise resulting
in overweight. Excessive fat accumulation which might harm the health of an individual can be defined
as overweight or condition of obesity [1]. A lot of health-related issues are born from merely having
extra fat. Few issues S. Madhumitha (B) · M. Magimaa · M. Maniratnam Department of Electronics and
Communication Engineering, Amrita School of Engineering, Bengaluru, India N. Neelima Amrita Vishwa
Vidyapeetham, Bengaluru, India e-mail: n_neelima@blr.amrita.edu © The Author(s), under exclusive
license to Springer Nature Singapore Pte Ltd. 2022 V. Bindhu et al. (eds.), Proceedings of Third
International Conference on Communication, Computing and Electronics Systems, Lecture Notes in
Electrical Engineering 844, https://doi.org/10.1007/978-981-16-8862-1_2

include cardiovascular disease and musculoskeletal disorders, and some types of cancer are serious
effect of non-healthy diet. Obesity is non-communicable, and yet it is one of the prime concerns of
almost all health organizations [2]. The proposed model aims to give nutritional reports on the food
which are consumed by the people, thus enlightening them on their dietary knowledge [3]. This analysis
will help in understanding the essential requirements and harmful substances in the food that are being
consumed. Now, with the onset of image processing and artificial intelligence, a lot can be done with
minimum human intervention. Food recognition and classification are the key factors in dietary
assessment and can be done using different methods like geometry features, statistical features and
methods based on machine learning. Multidimensional histogram can also be employed as a feature
vector for classification [4]. This model uses Inceptionv3 as CNN model, to draw out a feature graph of
the food image and classify the food image to the nearest food class. Each output from the CNN layer is
a feature map, and it contains parameters linked to the layers. It further processes the image and trains
it. The dataset used is Food-101, and it has 101. In this, mathematical models were used for image
recognition. However, the complexity of the architecture and the time required are the major setbacks.
Aiming to reduce the complexity of the architecture, the proposed model identifies the food by
extracting the features from the image using a computing algorithm. Main idea behind this model is to
help individuals monitor their food intake and to have a calorie intake count. The aim of dietary
assessment is to identify appropriate and useful areas of change in the individual’s diet and lifestyle,
thus aiming to improve a person’s overall health.
Conclusıon

Obesity is non-communicable, and yet it is one of the prime concerns of almost all health organizations.
One must monitor their individual dietary to maintain healthy lifestyle. Dietary assessment can play an
important role in every one’s individual lifewhich can assess every day’s intake and calorie management.
The proposed model classifies the good using deep learning approach and provides nutritional analysis
and dietary assessment using volume estimation methods. Once the food is recognized, area and the
corresponding volume are estimated. An accuracy of around 85% is obtained using this technique. A
database is made using the dataset that is used and with normalized values of fat, proteins, calories and
fiber of each food for 100 g

With respect to food class, from the corresponding database of nutritional particulars, the calorie value
is calculated, and the dietary output is obtained. This model can be used for health tracking and can be
used for leading a healthy life and promoting self-owned diet.

A Survey of IoT Based Pregnancy Woman Health Monitoring System

L. Jubair Ahmed1 Department of ECE,Sri Eshwar College of Engineering, Coimbatore, email id:
jubaircbe@gmail.com Madhumitha R 2 Department of ECE, Sri Eshwar College of Engineering,
Coimbatore, email id: madhumitha0803@gmail.com Mehana sree M 3 Department of ECE, Sri Eshwar
College of Engineering, Coimbatore email id: mehanamehu25@gmail.com Oviya K4 Department of
ECE,Sri Eshwar College of Engineering,Coimbatore email id: oviyakalimuthu@gmail.com Sharmila M5
Department of ECE,Sri Eshwar College of Engineering,Coimbatore, email id: sharmidia24@gmail.com

Abstract—In the Beginning of 2020, most of the countries were affected by COVID-19, especially India.
Our government announced lockdown, this situation became worse to many people mainly pregnancy
women who needed monthly health check-ups. Our solution for this problem is using IOT Based
Pregnancy Woman Health Monitoring System which helps woman in monitoring the health and getting
reports just sitting in home. The selected study issue presents an effective Monitoring System for
increasing the confidentially of pregnancy women health or otherwise, since it is an essential emotional
and psychological event a crucial component in the life of a married lady. This paper gives the survey
about the difference we bring from various health monitoring systems developed in recent years.
Keywords:COVID-19, IOT, Pregnant Woman, Sensors.
 I. INTRODUCTION
II. In December 2019, the first COVID-19 case was discovered in China and by January 2020, all
other 212 countries reported its first case. The virus spread increases vigorously. The live
case count and death count were increases tremendously all over the world. Some countries
were started to announce lockdown and stopped international flights, particularly .Indian
government announced country lockdown in March 2020 where there will be no state to
state and district to district travel, no restaurants for dine-in, grocery and medical shops
were opened only for particular hours. This environment mainly affected people who
needed monthly health check-ups especially pregnant women in rural areas. They have to
travel from their village to city for routine check-up. While travelling there may be a chance
that they can affected by virus. A lot of pregnant women skipped their health check-up
because of fear. In July 2021, medical council announced that pregnant woman also allowed
for Vaccination, they can choose between Covishield, Covaxin and Sputnik V. Many came
forward but many in rural areas did not get much awareness. And this doesn’t stop them
from fear. Again, in December 2021 other variant form of COVID-19 virus OMICRON was
identified in India and the spread again increasing rapidly.

To help pregnant women, we are going to develop a IOT Based pregnancy woman health monitoring
system which monitor and give report of both mother and child. In the wide area of Communication IOT
isa essential interconnecting medium.In this aspect of the IOT, Sensor devices are used to gather
information, respond to actuator signals[12], and deliver feedback via regulated and approved input.
The information is provided to the user by recording the acquired data

Human-machine interaction is enabled by the Internet of Things. There is quiet more health monitoring
systems were developed in recent years. This paper gives a survey about how we are different from
other developers and what are the features we added to help pregnant woman. Health-care systems
keep a tight eye on the patient's activities. Through the cloud server, this information is transferred to
the approved users' smart devices. All of the information that is available can be kept and analyzed to
aid in decisionmaking. The parameters may be used to access the person's essential details such as
Temperature, Oxygen, Heart rate, etc., which is useful in an emergency.The suggested monitoring
system's major goal is to integrate the newest trends and technologies that emerge in the field of
healthcare from time to time in order to strengthen the tracking device and plug any gaps in existing
systems.The suggested technique intends to increase the efficacy of the device designed to track
patients and offer timely preventative signals to the sufferers in question before serious damage occurs
to the adoptive parent and their baby.Various filtering methods for noise removal in biomedical images
[1][2].

VI.CONCLUSION

The Internet of Things (IoT) is a huge network of connected devices that supply crucial physical data,
which is subsequently processed in the cloud to generate business value. It expands the possibilities for
some important components across many companies and industries, along with the Health
Management System.This study looked into the usage of IOT in a variety of aspects of people’s lives as
well as the medical sector.A review of recent research in IOT based solutions was conducted like a
prelude towards the suggested experimental inquiry.The existing Smart Maternity process of finding is
assessed as successful, but the system needs to be enhanced further. As a result, a new version with the
most up-to-date scientific and technological inputs is required, and monitoring services must be
improved to make them more effective and foolproof in order to provide better healthcare to pregnant
women in rural areas during pandemics. The suggested tracking device's major goal was to include the
latest findings as well as innovation in the field of healthcare on a regular basis in order to enhance the
device and close gaps in current products[12]. The suggested approach intends to increase the efficacy
about the device designed to automatically screen mothers and make warnings available to the people
in question as soon as possible, whenever the expecting woman or her unborn suffers. Both before and
after birth, pregnant women and children would benefit enormously from this proposed userfriendly
paradigm based on specialised dependable IoT devices.

Efficient Machine Learning for Malnutrition Prediction among under-five children in India Saksham Jain∗
Department of Electrical Engineering Aligarh Muslim University Aligarh, India saksham1598@gmail.com
Tayyibah Khanam∗ Department of Electrical Engineering Aligarh Muslim University Aligarh, India
tayyibahkhanam@zhcet.ac.in Ali Jafar Abedi Department of Community Medicine Aligarh Muslim
University Aligarh, India ajabedi@myamu.ac.in Abid Ali Khan Department of Mechanical Engineering
Aligarh Muslim University Aligarh, India abid.khan.me@amu.ac.in Abstract—Child malnutrition is
considered to be one of the leading causes of infant mortality and malnutrition. This study was aimed to
leverage the advantages offered by machine learning models in terms of determining and accurately
predicting significant factors of malnutrition. For this study, the Children’s recode files from the Indian
Demographic and Health Survey (IDHS) datasets from 2005-2006 and 2015-2016 were used. To examine
the nutritional status of children aged 0-59 months, this study looks at stunting (Height-for-age), wasting
(Weight-forHeight), and concurrent stunted wasting (Height-for-age-Weightfor-Height). Regular
Machine Learning models, Tabular Deep Learning frameworks, H2O base models, and AutoML models
are the four types of machine learning models employed in our research. This research found that
Automated machine learning algorithms and Tabular Deep Learning frameworks, in general,
outperformed other models in terms of speed and efficiency, as well as Accuracy (up to 96.46%) and
AUC-ROC scores (up to 99.95%), which are important in classification problems like this one. Following a
graphical representation of the importance of numerous drivers of malnutrition for all three
anthropometric indices, we concluded our findings by comparing the performances of several models
and determining the topperforming algorithms. This paper significantly contributes to the possibilities of
using machine learning in identifying probable correlates of malnutrition for the effective prevention,
cure, and identification of target groups. Index Terms—AutoML algorithms, Feature Importance,
Malnutrition, Tabular Machine Learning I. INTRODUCTION Starvation, dietary deficiency and sanitation
are some of the major obstacles to achieving economic and social development goals in any developed
nation. According to the World Health Organization, there were 149.2 million stunted children under the
age of five, 45.4 million wasting children, and 38.9 million overweight children in 2020. With the
exception of Africa, the number of stunted children is decreasing in all regions. Southern Asia is home to
more than half of all wasting children, while Asia as a whole is home to more than threequarters of all
children suffering from severe wasting. In terms *Equal Contribution of the targets, the stunting aim is
making the most progress at the country level, with approximately two-thirds of countries experiencing
at least some progress [12].The world’s most malnourished people are in developing countries in
southern Asia and Africa and hence, malnutrition rates in Southern Asia and Sub-Saharan Africa are
particularly high, among the highest ones in the world [5]. Thus, malnutrition among children under the
age of five is a major public health issue in India and thus is also responsible for about half of all under-5
child deaths in India. The graveness of this situation is evident by the fact that India has one of the
highest rates of stunted children in the world, approximately 46.6 million which is double that of Sub-
Saharan Africa [14] and also accounts for one-third of the global total according to Global Nutrition
Report 2018 [15]. Poor nutrition in the first few years of a child’s life can also contribute to restricted
growth, which is directly linked to poor cognitive abilities, reduced understanding and attention at
school and thus, is a major hindrance to a child’s holistic progress and development. Furthermore, in
countries where malnutrition is prevalent, the implications of malnutrition extend far beyond the
person, and thus contribute in reduction of total labor-force productivity and economic growth. Other
significant indicators of malnutrition in children aged 0 -5 are low birth weight, diarrhoea within the last
six months, and physical developmental delay in most underdeveloped nations, including India. While
socioeconomic status of the family is one of the major factors contributing to health issues in a child and
directly affects the child’s dietary score, sanitation and lifestyle, other external factors such as
environmental pollution, political commitment and the role of governmental policies can’t be ignored.
To lessen the risk of malnutrition and its dire consequences in the Indian society, it is thus important to
identify other significant concentrations through background factors and independent factors, the level
of importance of each contributing factor and the correlation between these factors. With the 978-1-
6654-5883-2/22/$31.00 © 2022 IEEE 2022 IEEE Delhi Section Conference (DELCON) | 978-1-6654-5883-
2/22/$31.00 ©2022 IEEE | DOI: 10.1109/DELCON54057.2022.9753080 Authorized licensed use limited
to: Hanyang University. Downloaded on December 14,2023 at 10:13:45 UTC from IEEE Xplore.
Restrictions apply. goal of proper identification and treatment of malnutrition, researchers in the past
have aimed to accurately predict the linked condition through machine learning techniques, with the
primary aim of accurate malnutrition prediction to reduce the risks. While most existing research
attempts to predict malnutrition through one or more anthropometric indices, we consider a concurrent
stunted-wasting index based on the studies and evidence found in the literature regarding its relevance.
Finally, our study extends a novel empirical approach through Machine Learning algorithms to be able to
flexibly determine to what extent malnutrition is driven by several detected social, economic, and
environmental factors. This also gives us the opportunity to examine the effect of other undetected
factors such as culture, politics, and conflict. The rest of the paper is sectioned as follows. Section II
discusses the background behind Machine Learning and Malnutrition. Section III discusses related works
and the methodology of our approach with its novel contributions. Section IV, V & VI give overview on
the data-sets, discusses the implementation & results and summarize the paper findings respectively

VIII. CONCLUSION

This research presents evidence for combining wasting and stunting as a hybrid index of Stunted-
Wasting, that demonstrates significantly better results than stunting and wasting considered
individually.Even though Machine Learning algorithms are currently in practice, by utilizing open-source
tabular deep learning libraries and AutoML algorithms, we attempted to fasten up the process while also
making it much more efficient. Ultimately, we also demonstrate results of the AUC-ROC scores to give a
complete idea of the performances of these models. We conclude that the best performances are
achieved on TabNet, an interpretable Tabular Deep Learning framework, followed by H2O Random
Forest base model and AutoML algorithms. Some of the notably important features present in our
dataset found by the feature importance plots in Figures 3,4 & 5 are - Child’s age, Mother’s BMI,
Mother’s age, Toilet Facility, Wealth Index, Birth weight, Mother’s Education, Sex of Child, State,
Religion, Caste, Exposure to Mass Media and Immunization. With this background, we can now predict a
child’s health beforehand given information about important health, family, and personal factors.

Prediction of Child Malnutrition using Machine Learning Shubham Kar Dept. of CSE University of
Engineering & Management, Jaipur, India shubhamkar1999@gmail.com Sauvik Bal Dept. of CSE
University of Engineering & Management, Jaipur, India sauvikbal@gmail.com Susmita Pratihar Dept. of
CSE University of Engineering & Management, Jaipur, India susmita.pratihar1409@gmail.com Gururaj H
L Dept. of CSE Vidyavardhaka College of Engineering, Mysuru, India gururaj1711@vvce.ac.in Subhadip
Nayak Dept. of CSE University of Engineering & Management, Jaipur, India
subhadipnayak56@gmail.com Ravikumar V Dept. of CSE Vidyavardhaka College of Engineering, Mysuru,
India ravikumarv@vvce.ac.in Abstract— Sometimes malnourished children fall into some serious health
issues. And doctors are unable to find out the root causes of their illness, but they used to apply some
practices which were not appropriate for every child. Children often die because of this reason. So, it is
very dangerous for malnourished children. Along these lines, the fundamental point of our review is to
anticipate hunger status of a 1 to 5 years more established kid in Asia by utilizing AI. Looked for ongoing
examination papers (2010 - 2020) which identified our point and combined outcomes into a synopsis of
what is and isn't known and tried to find out advantages and drawbacks. As explained in the
introduction part, to do so, selected a suitable dataset from open source. And they went through many
articles to know about Machine Learning Algorithms like their advantages and drawbacks. So, four
widely used Machine Learning classifiers like Random Forest, Support Vector Machine, K-Nearest
Neighbors, Logistic Regression have been considered to predict a good accuracy score of malnutrition
status among under 5 years old. At last, they looked for the best algorithms according to their accuracy
score. Based on various performances of Machine learning Algorithms, the best results were performed
with Random Forest and Logistic Regression, which demonstrate an accuracy of 91.11 % and 89.88 %,
Train accuracy of 1.000 and 0.847. Additionally, a most extreme discriminative ability appeared by
Random Forest classification. Here they analyzed those 4 ML algorithms to find which one is performing
best. Among them Random Forest and Logistic Regression performing very well. And in future they will
do some beneficial work for malnourished children. Keywords— Child Malnutrition, Random Forest,
SVM, Logistic Regression. I. INTRODUCTION Primarily, malnutrition is a disruption of the body's normal
growth and development. Usually, children aged 0-5 years are malnourished in most cases. In a single
word, malnutrition is the lack of nutrition in the body [1]. There are many different reasons behind child
malnutrition. Lack of protein and vitamins in the diet is a major cause of malnutrition. In most cases, the
baby will not be able to get the required protein because the amount of nutritious food like meat, egg,
fish in the diet is low enough. Moreover, the population of India and Asia has increased significantly in
the last few decades. The supply of cereals, fruits, fish, meat, and milk has declined, which is a major
problem for the economically backward people. Often because of absence of cash they can't devour
adequate protein and nutrients needed by the body. Children are also malnourished due to problems
with food intake, digestion, absorption, and metabolism. Being infected with any complex disease like
diarrhea, jaundice, typhoid etc. after birth is also one of the major causes of malnutrition. Malnutrition
problems also depend on the rate of family education [2]. The number of malnourished children is
increasing every year. Ailing health obstructs the psychological and actual development of youngsters,
prompting an increment in the quantity of kids with gathered insight. Children also suffer from serious
diseases like Kwashiorkor, Marashmas, Rickets, Pneumonia, Malaria, Measles, and even anemia. And
most of the time die due to lack of proper treatment [3]. As of now, serious hunger is the reason for
around 33% of all kids passing consistently. The number of deaths of girls is comparatively higher than
that of boys. Essentially because of destitution and absence of training to get appropriate nutritious
food the rate of disease among youngsters living in rustic and metropolitan ghetto regions is altogether
higher than in metropolitan regions. The growth of the country will be measured not only based on
resource and total capita, it will also be measured based on the health of the nation. The basic problem
in India is malnutrition among the children. The main problem in India is malnutrition in children under 5
years of age. In the Statistics they show that the number of underweight children in India is the highest
level and twice as high as in Africa [4]. The problem of malnutrition in India has been considered as a
concentrated phenomenon. In emergent nations, around 2.3 million children between the ages of 6 and
60 die each year, mainly because of malnutrition. It presumes that the breadth of gentle underweight
requests more mindfulness as a helpful sign of changing general medical issue among preschool
youngsters in developing countries [5][6]. That recognize lack of healthy sustenance at a crude stage for
the anticipating and usage of ideal intercession at the association for the wellbeing framework. As the
author discussed in the introduction part, child Malnutrition will be a huge disaster for the children
under 5 years. So, from our project they will be able to know about the main reason for malnutrition,
most affected places, and others. and if those things are known to us also, they can handle the situation.
Then they can save those affected children’s lives. So, they were motivated for that reason.

V. CONCLUSION

In this study, we compared those four machine learning algorithms based on their accuracy score, train,
and test accuracy. So that they can predict accurately whether a child is malnourished or not. Among
those algorithms Random Forest and Logistic Regression performed very well with the most accurate
prediction in Asia's dataset. In this research basically run our machine learning algorithms all over Asia.
To foresee if a kid is malnourished. But in future we will do the same for India. So that they will be able
to know India’s current scenario about malnutrition. It is known that the approximate amount of the
children who are suffering from this situation and they can find out some main reasons behindit. Our
main aim is to control the situation. So, they will arrange some campaigns to make people aware about
this and to make them sensible. They will provide them proper guidance to overcome this situation.
Also, in the future it will be a very good and valuable discovery to differentiate between children
whether they are affected by malnutrition or not. And they will create an instrument which can identify
malnourished children who are at higher risk. Thus, later on they will provide them with nutritional food,
required medicines and other useful health supplements. In this way, our model will be valuable for
controlling the situation and repeal of child malnutrition
An Intelligent Image Classification Approach for Food Items Nutrition Facts Preethi Harris(B) , M. Nithin,
S. Nithish Kannan, R. Gokul Prasanth, and M. Kissore Kumar Sri Ramakrishna Engineering College,
Coimbatore, India preethi.harris@srec.ac.in Abstract. Nutrition is essential for health and wellbeing. The
state of health is determined by what one eats. Knowing the nutrient contents of food create awareness
of what to eat and how much to eat. People are growing fond of the modern lifestyle and they merely
eat to fill up their stomach. The UN report states that over 25 percent of the world’s population lack
nutritious food. Lack of balanced diet may lead to obesity and other health related issues. Study further
reveals that through appropriate diet and daily food consumption tracking, even obese people showed
remarkable reduction in weight. Inspired by these findings in literature, a mobile application has been
proposed to identify and classify Indian breakfast food items using a derivative of the Convolution
Neural Network model called MobileNet. The identified images are further classified to determine the
nutritive content in the food identified. In addition, the mobile App also summarizes the daily nutritional
value of food consumed against the recommended nutritional value based on gender. Thus, the
proposed work can be extended to monitor the daily food habits of people with health conscious and
related issues. Keywords: Health · Balanced diet · Obesity · Food item · Mobile application · Convolution
neural networks · Nutrition 1 Introduction Image classification is a concept where a system can analyze
an image and identify the ‘class’ to which the image falls [1]. A class/classification is nothing but a label
such as car, flower, plant and so on. Image classification is the process of the system analyzing the image
and telling you the label under which it can be clubbed. With deep learning [2], image classification has
become an area widely explored. Deep learning which is an important part of ML has triggered
significant progress in image classification in recent years with Convolutional Neural Networks (CNN) [3,
4] models. One popular model among App developers is the MobileNet model of CNN for Image
Classification, especially in Computer Vision. There are many other models but what makes MobileNet
[5] special that it requires very less computation power to run or apply machine learning with optimum
number of layers. So it is best suited for portable and hand held devices such as Mobile phones, tablets,
devices with embedded systems etc. It is also compatible with web browsers as browsers but with
limitation on the computation power, graphic processing and storage. © The Author(s), under exclusive
license to Springer Nature Switzerland AG 2022 J. I.-Z. Chen et al. (Eds.): ICIPCN 2021, LNNS 300, pp.
422–430, 2022. https://doi.org/10.1007/978-3-030-84760-9_37

The current generation people are more diet conscious, either to maintain a good figure or owing to
ailments. The dependence of smart devices for food habit tracking and diet is growing tremendous in
usage with the fast paced life. The use-of MobileNet model for food classification could aid the
development nutria-apps based on customized features [6]. Some of the related contribution in
classification especially related to food items in literature is summarized as follows: Derek in his work
classified grocery items of Freiburg groceries dataset using CNN [7]. This work revealed that the state-of-
the-art research to classify grocery items had mostly focused only on Computer Vision and not neural
networks. In this work the authors used a greedy algorithm called guided_pruning along with
Convolution Neural Network (CNN) to improve the accuracy. However, one of the biggest challenges
faced in this work was the large resources needed to run such models. David et al. [8] proposed food
item classification using CNN unlike the Artificial Neural Network. 2D Convolution Neural Network with
Max-Pooling function was proposed to train the FOOD-101 dataset for image classification. Multiple
layers were concatenated to form the final output to classify multimedia dataset. However, there were
some mis-classification in the images and the computational time was significantly large due to more
number of prominent features. At the outset Mohd Azlan [9] used TensorFlow for Deep Neural Network
to classify five different types of flowers. The role of epochs played a major role in determining the
accuracy of the classified image. This work used Python along with TensorFlow framework. Sirawan and
Olarik [10] took-up the challenge of classifying real-world food image using MobileNet. In this work food
image identification and classification was proposed using a modified MobileNet architecture comprising
of global average pooling layers. This customization could address overfitting the food images, batch
normalization, rectified linear unit, dropout layers and the last layer is usage of softmax layer. They
further concluded that data augmentation could give better accuracy with MobileNet. The survey done
researchers at Duke University [11] for overweight/obese people revealed that when the daily food
intake was tracked using an app, significant loss in weight was reported. Not everyone wants or have the
time for a weight-loss treatment hence it is important to create alternate approaches that can
accommodate these people. Digital health approaches have potential for that and researchers have
been focusing on app development augmented with other technologies for users to record their daily
food intake and weight. The dieters can receive in-app reminders to log their meals daily through such
apps. The adoption of mobile applications has created an opportunity for improving nutrient intake as
they can be used for digitalizing dietary assessments. Many nutrition-related apps have been launched
in which only few of them effective and installed by the users. Some of the apps recording of food intake
using a food diary feature. Previously food selection was done using text search or barcode scanner
technologies with the common attributes focusing on weight, height, gender and age. Some apps
offered the possibility of creating personal reminders, which could be used, for example, to remind
users of consuming snacks during the day time. Some apps proposed diet challenges to users. Emerging
technologies, such as image recognition were not identified. Understanding 424 P. Harris et al. these
findings, the proposed system has been implemented with deep learning image classification with
classification of Indian food. With image classification, the user does not need to waste time. With just
one click, the nutritional content is displayed. Thus, literature survey done revealed the various
techniques used for food recognition. However, a system that can accurately, efficiently and quickly
detect and classify food is in need. It should use an alternative efficient approach to make the process
responsive and reliable. That can be achieved by the proposed system which uses deep learning-based
approach for image classification and tracking. The next section describes the proposed work done
explaining image classification, library used for model creation, the framework used to design the UI of
the application, the database used and the features of the application

4 Conclusion

Eating a healthy and consuming a balanced diet has become a topic of interest among the current
generation. This work utilizes the concepts of MobileNet Model as it has reduced number of layers and
performs classification in few seconds. The proposed work focuses on recognizing and classifying five
Indian food items along with daily nutrition calculation against the nutrition consumed. The focus of the
proposed work is classifying the food item images under the five classes based on the primary
ingredients without considering the toppings in the food item. The ease of use and tracking the nutrition
consumed against the daily nutrition chart, is also based on gender, with access from anywhere and
anytime. The system can be further enhanced by training the model to classify more food items like
regional food categories, tracking the diet accurately driven by Robotic Process Automation

Nutrition Monitoring and Health Awareness System based on Crowd Sourced Food Images using
Transfer Learning Serene Ravi John∗, Rahul Krishnan Pathinarupothi∗, Uma Gopalakrishnan∗, Srisairam
Achuthan† and Nivedita Pavithran‡ ∗Center for Wireless Networks & Applications (WNA), Amrita
Vishwa Vidyapeetham, Amritapuri, India † IntegraConnect, West Palm Beach, Florida, USA ‡Department
of Clinical Nutrition, Amrita Institute of Medical Sciences and Research Center, Amrita Vishwa
Vidyapeetham, Kochi, India Abstract—With a worldwide appeal towards the need for a healthy lifestyle,
food plays a major role in contributing towards it. A real time system that can provide the nutritional
details of everyday meals from images taken can be useful. The existing systems and applications have
an image database majority of which are homogeneous in nature, and are not what is expected to be
seen in a real world application. Also, there is a lack of regional food and dish data sets from the
southern part of India. With the goal of developing a real-world, region specific solution, we embarked
upon building a crowd sourced data set and utilised transfer learning model for identification of food
and nutritional data giving an average validation accuracy of 90%. Index Terms—Convolutional neural
network, Food images, Transfer learning, Inception v3 I. INTRODUCTION Nutrition is a critical part of
health, which is defined as the study of nutrients in food, how the body uses them and the relationship
between diet, health and disease [1]. The nutrients are grouped into two on the basis of quantity
required as macro nutrients (carbohydrates, fats, proteins and water) and micro nutrients (minerals,
vitamins). Finding a balance in the nutrition intake is difficult, and if not properly maintained could lead
to various common lifestyle diseases and create underlying complications like heart diseases ,stroke,
obesity and type 2 diabetes. Apart from a well-managed diet, physical exercise and medications support
the individual to keep these conditions under control. Many individuals struggle to maintain a balanced
diet and also keep a track of their food intake. Hence, identifying the nutritional intake from food for
individuals with chronic conditions (e.g. diabetes) who require strict supervision on their diet is
necessary as their conditions can be adversely affected by it. Current methods include self report, 24-
hour recall and food frequency questionnaires (FFQs). There are also various electronic dietary
assessment tools available which allows various trained staff to assess the intake of food, allergies and
identify food patterns during primary health care and develop treatment plans. There are many
applications available on mobile platforms to track food intake and monitor calories, providing both free
and paid services to their users. With rapid development in the field of deep learning and artificial
intelligence, food image analysis has become computationally highly accurate. However, the state-of-art
systems use image databases in which the majority of images are of homogeneous images. For instance,
the images in the databases have white plates, standard bowls, and the photos are taken from standard
top-view positions. However, in the real world such ideal conditions can seldom be expected. In our own
initial study collecting crowd sourced images, it became evident that such assumptions limit the utility of
food recognition systems, which could perhaps explain the lack of popular adoption of such apps in our
daily lives. Another drawback we observed in the state-of-art systems is the lack of support for diversity.
India is known for its varied cultures and even small regions have developed their own diversity in food.
This poses a major technological challenge when it comes to the identification and nutritional
assessment of Indian food items. It is also worthwhile to note that Indian dietary habits have evolved
over millennia to suit the bodily nature of the tropical climate and hence, majorly consists of cooked
items rather than raw vegetables/fruits/meat. Moreover, we noted that there is also a significant
absence of food image data sets from the southern regions of India. In this paper, we report our project
of building a nonhomogeneous, regional data set specific to regional cuisine. To ensure real world
adoption, the images used for training were collected through crowd-sourcing and hence represented
the wide variations seen in images taken in diverse environments. We then develop a CNN based food
image recognition system that uses transfer learning models that are re-trained on a variety of food
images seen generally in the southern regions of India. The rest of the paper contains the related work,
the various challenges, the problem definition and the main objectives. Section 2 compares the related
works, section 3 describes the regional food classification system, followed by its implementation
reported in section 4. We discuss the results in Section 5, and finally the conclusion and future works in
Section 6

VI. CONCLUSION With nutrition monitoring from food images being an active area of research that has
been being carried out on multiple large data sets of images. It was noted that in these data sets there is
a lack of representation of Indian and other region specific food items as well as of images that are taken
in a real world perspective. In order to bridge that gap we have assembled a crowd sourced region
specific image data set on 10 most commonly consumed food items in Kerala. By studying the various
existing models and methods for nutritional value calculation and monitoring, InceptionV3 was selected
to carry out multi-class classification as it is a pretrained model with major accuracy of 78% on ImageNet
data set of approx 1 million images. Learning of the different architectures and varying the parameters
of the model as an application of transfer learning testing and evaluation of the model was carried out.
The custom data set was used to train and validate the InceptionV3 network for which the model gave
an average validation accuracy of 90% .The different variations made while creating the models have
also been studied.

Article South Indian Cuisine with Low Glycemic Index Ingredients Reduces Cardiovascular Risk Factors in
Subjects with Type 2 Diabetes Nivedita Pavithran 1 , Harish Kumar 2 , Arun Somasekharan Menon 2 ,
Gopala Krishna Pillai 3 , Karimassery Ramaiyer Sundaram 4 and Omorogieva Ojo 5,* 1 Department of
Clinical Nutrition, Amrita Institute of Medical Sciences and Research Centre, Amrita Vishwa
Vidyapeetham, Kochi, Kerala 682041, India; brinivedita@aims.amrita.edu 2 Department of
Endocrinology, Amrita Institute of Medical Sciences and Research Centre, Amrita Vishwa Vidyapeetham,
Kochi, Kerala 682041, India; harishkumar@aims.amrita.edu (H.K.); drmenon.arun@gmail.com or
arunsmenon@aims.amrita.edu (A.S.M.) 3 Department of General Medicine, Amrita Institute of Medical
Sciences and Research Centre, Amrita Vishwa Vidyapeetham, Kochi, Kerala 682041, India;
mgkpillai@aims.amrita.edu 4 Department of Biostatistics, Amrita Institute of Medical Sciences and
Research Centre, Amrita Vishwa Vidyapeetham, Kochi, Kerala 682041, India;
krsundaram@aims.amrita.edu 5 School of Health Sciences, Faculty of Education, Health and Human
Sciences, University of Greenwich, London SE9 2UG, UK * Correspondence: o.ojo@greenwich.ac.uk; Tel.:
+44-20-8331-8626; Fax: +44-20-8331-8060 Received: 4 July 2020; Accepted: 20 August 2020; Published:
27 August 2020 Abstract: Background: Inflammation is considered as a predictor of cardiovascular
diseases in type 2 diabetes mellitus. No previous studies have investigated the effect of low glycemic
index (LGI) recipes of South Indian cuisine on the risk factors of cardiovascular disease in patients with
diabetes. Aim: The aim of this randomized controlled trial was to evaluate the improvement in
cardiovascular risk factors and blood glucose control, in patients with type 2 diabetes, after intervention
with recipes of Kerala cuisine, from locally available whole grain cereals, low in glycemic index. Method:
This was a prospective and randomized controlled study that was conducted over a period of 24 weeks.
A total of 80 participants were recruited from the Department of Endocrinology and Diabetes
Outpatient in Kerala, South India. All 80 patients had type 2 diabetes, and were aged between 35 and 65
years. Participants were randomly assigned and advised to follow either a LGI diet plan (n = 40) or their
usual diet, which served as a control group (n = 40). The advice was reinforced throughout the study
period. Anthropometric, biochemical parameters which included glycemic and cardio-metabolic
parameters were measured according to standard procedures. T-tests were conducted to compare the
differences between intervention and control groups, and the Pearson correlation coefficient was used
to evaluate associations between the variables. Results: There were significant differences (p < 0.05)
between the intervention and control groups with respect to weight, HbA1c, insulin, triglycerides,
Homeostatic Model Assessment of Insulin Resistance (HOMA-IR), high sensitivity C-reactive protein (hs-
CRP) and apolipoprotein B (ApoB). There was also a positive correlation between weight and blood
glucose variables. ApoB was positively correlated with lipid profile and insulin levels. Conclusions: The
long-term implementation of LGI diet of Kerala cuisine has been found to promote weight loss, enhance
insulin sensitivity and reduce the cardiovascular risk. Keywords: low GI diet; south Indian (Kerala)
cuisine; hs-CRP; HOMA-IR; ApoB; glycemic control Int. J. Environ. Res. Public Health 2020, 17, 6232;
doi:10.3390/ijerph17176232 www.mdpi.com/journal/ijerph Int. J. Environ. Res. Public Health 2020, 17,
6232 2 of 17 1. Introduction Diabetes mellitus is a well-established risk factor for cardiovascular disease,
and an increased incidence of coronary artery disease and cardiovascular mortality has been strongly
associated with the presence of metabolic syndrome [1] This linkage has been corroborated by the
notably higher morbidity and mortality rate in patients with diabetes compared to non-diabetic
individuals. Besides, metabolic syndrome has been reported to be a predictor of subclinical
atherosclerosis. Some of the factors associated with the increased risks of cardiovascular disease in
diabetes mellitus are higher levels of low density lipoprotein (LDL) cholesterol, elevated blood pressure,
lower levels of high density lipoprotein (HDL) cholesterol, insulin resistance, hyperglycemia,
inflammation, hypertension, dyslipidemia and obesity [2]. Vascular dysfunction resulting from type 2
diabetes has been shown as the causative factor for cardiovascular disease [3]. Inflammatory-response
proteins are considered as predictors of cardiovascular events. C-reactive protein (CRP) is defined as a
sensitive, but non-specific marker of inflammation [4]. The atherosclerotic process is now considered as
an inflammatory disease. Compared to other ethnic groups, South Asian community is highly vulnerable
to higher mortality rates from coronary heart disease (CHD) and stroke. Among Asians, incidence of
CHD, 5–10 years earlier, have been ascribed to lower HDL, higher triglyceride and higher LDL levels and
higher blood pressure (BP). Furthermore, a strong correlation has been found between diabetes and
higher rates of heart failure (HF) and all-cause death (ACD) [5]. Higher incidences of cardiovascular (CVD)
complications in the Asian population with type 2 diabetes mellitus (T2DM) calls for systematic
treatment strategies tailored to Asian ethnicity [6]. Cereals are the staple diet of South Indians and
carbohydrate constitutes about 55–65% of their daily intake of calories. There has been a transition in
the quality of grains consumed from whole grains to highly polished grains. Consumption of whole
grains is highly recommended unlike refined grains, which comprise endosperm that are known to
stimulate adverse effect on the cardio-metabolic risk factors including diabetes [7]. Ingestion of high
carbohydrate diets can raise blood glucose, insulin and triglycerides, leading to insulin resistance.
Glycemic index (GI), which gauge the quality of carbohydrates, is considered as an important factor in
the management of non-communicable diseases, particularly type 2 diabetes. Glycemic index is a
relative ranking of carbohydrate in foods according to how they affect blood glucose levels.
Carbohydrates with a low GI value (55 or less) are more slowly digested, absorbed and metabolized and
cause a lower and slower rise in blood glucose and, therefore, insulin levels. High glycemic index
foods(>70) have been shown to trigger rapid increase in the post-prandial blood sugar and insulin [8]
and have shown to increase the risk of type 2 diabetes mellitus and CVD [9,10] in the western
population as well as in Asian population, specifically Indians [11] and Chinese [12]. Traditional Indian
diets are rich in dietary fiber, which could explain the lower incidence of type 2 diabetes mellitus in India
in the 1960s–1970s. Studies in western countries have shown that the replacement of refined grains
with whole grains have resulted in risk mitigation of type 2 diabetes and CVD in hyperinsulinemic and
impaired glucose tolerant patients. In our previous study [13] that investigated the effect of a low GI
(LGI) diet with local recipes of South Indian, Kerala cuisine on body composition and glycemic control in
people with diabetes, it was demonstrated that there was a significant reduction (p < 0.05) of truncal
obesity and glycated hemoglobin. However, there is minimal published testimony that shows the
efficacy of long-term intervention, triggered by smaller changes in the intake of traditional red rice and
whole wheat flour that is low in glycemic index for the risk reduction of CVD in South Indians with type 2
diabetes. The current study is a follow up component of a larger study directed to assess the
improvement in cardiovascular risk factors and blood glucose control among type 2 diabetes subjects,
pursuant to a 24-week dietary consumption of Kerala cuisine from locally available LGI whole grain
cereals

5. Conclusions

The inclusion of LGI diets of Kerala cuisine is an effective strategy of alleviating cardiovascular risk
factors in patients with type 2 diabetes. In particular, the long-term implementation of LGI diet of Kerala
cuisine has been found to control blood sugar and has been found to promote weight loss, enhance
insulin sensitivity and minimize cardiovascular risk. Author Contributions: N.P. conceptualized, collected,
and analyzed data and wrote the initial draft; H.K. guided the study and revised the manuscript; A.S.M.,
G.K.P. and O.O. revised the manuscript; K.R.S. performed the statistical analysis. All authors have read
and agreed to the published version of the manuscript

---------------------------------------------------------------------------------------------------------------------------------------.
Advancing Smart Malnutrition Monitoring: A Multi-Modal Learning Approach for Vital Health Parameter
Estimation Ashish Marisetty, Prathistith Raj M, Praneeth Nemani, Venkanna Udutalapally, Debanjan Das
Abstract—Malnutrition poses a significant threat to global health, resulting from an inadequate intake of
essential nutrients that adversely impacts vital organs and overall bodily functioning. Periodic
examinations and mass screenings, incorporating both conventional and non-invasive techniques, have
been employed to combat this challenge. However, these approaches suffer from critical limitations,
such as the need for additional equipment, lack of comprehensive feature representation, absence of
suitable health indicators, and the unavailability of smartphone implementations for precise estimations
of Body Fat Percentage (BFP), Basal Metabolic Rate (BMR), and Body Mass Index (BMI) to enable
efficient smart-malnutrition monitoring. To address these constraints, this study presents a
groundbreaking, scalable, and robust smart malnutrition-monitoring system that leverages a single full-
body image of an individual to estimate height, weight, and other crucial health parameters within a
multimodal learning framework. Our proposed methodology involves the reconstruction of a highly
precise 3D point cloud, from which 512-dimensional feature embeddings are extracted using a headless-
3D classification network. Concurrently, facial and body embeddings are also extracted, and through the
application of learnable parameters, these features are then utilized to estimate weight accurately.
Furthermore, essential health metrics, including BMR, BFP, and BMI, are computed to conduct a
comprehensive analysis of the subject’s health, subsequently facilitating the provision of personalized
nutrition plans. While being robust to a wide range of lighting conditions across multiple devices, our
model achieves a low Mean Absolute Error (MAE) of ± 4.7 cm and ± 5.3 kg in estimating height and
weight. Keywords—Multi-modal Learning, 3D Reconstruction, Feature Fusion, Height and Weight
estimation, Smart Healthcare, Noninvasive. I. INTRODUCTION Malnutrition is an ailment caused by
consuming food that lacks an adequate quantity of essential nutrients. It is most commonly used in
reference to undernutrition, [1] which occurs when a person does not receive sufficient calories,
proteins, or micronutrients. A scarcity of a quality diet most commonly causes undernourishment or
undernutrition. According to a WHO survey, there are 178 million malnourished children globally, with
20 million suffering from severe malnutrition, contributing to 3.5 to 5 million deaths in children under
five each year. On a global scale, undernutrition is responsible for 45% of all casualties in children under
five and is widespread in developing nations, especially among women and children. Malnutrition also
poses a range of severe health problems that include anemia, diarrhea, disorientation, weight loss, night
blindness, anxiety, attention deficits, and other neuropsychologic disorders [2]. In the aftermath of the
COVID-19 outbreak, which caused significant concerns and stress regarding public health [3], the
traditional approach of measuring height and weight in public health centers has been impacted. During
the pandemic, strict social distancing measures were put in place to minimize the spread of infection,
making the conventional method of calculating essential health metrics through direct measurements
undesirable. In addition, pandemics like COVID-19, according to UNICEF, put malnourished children at
an ever-increasing danger of mortality, as well as impaired growth, development, and learning for those
who survive. Therefore, there is a dire need to identify important health indicators and monitor chronic
stress & uncontrolled or unmonitored food consumption integrated with data-driven approaches [4]. A
primary step in identifying or diagnosing malnutrition and the nutritional status of any person can be
determined by computing their Body Fat Percentage (BFP), Basal Metabolic Rate (BMR), and Body Mass
Index (BMI) and comparing it with standardized charts. It is more accurate to infer the risk of
malnutrition and various medical conditions from these metrics since they represent the human body’s
functionality in a well-oriented manner. In this work, we intend to predict the height, weight and
successively calculate the important health metrics as mentioned above from a sing-shot full-body
image by incorporating a holistic representation of prominent features under the multi-modal learning
paradigm. Fig. 1 illustrates a conceptual overview of the proposed method. Person Suffering from
Malnutrition ❏ Social and mobility problems ❏ Digestive disorders and stomach conditions ❏ Mental
health conditions Causes of Malnutrition Usage of our application Monitoring by Doctor Personalized
Nutrition Plan Fig. 1: Conceptual Overview In this paper, we propose a solution based on multi-feature
fusion that includes 3D, facial, body, and metadata features integrated with a smartphone application
prototype to estimate a human’s height, weight, and other health parameters. The smartphone’s
camera serves as a sensor to capture a full-body image of a human, and the height is estimated by
calculating the centimetre per pixel ratio using image processing techniques. Following that, the
captured image is pre-processed by detecting, cropping, aligning the face and body, reconstructing &
samping a 3D person mesh object, and feature extraction in a multi-modal framework

VI. CONCLUSION AND FUTURE WORKS This research presents a novel approach for predicting height and
weight and inferring other health indicators, such as BMI, BMR, and BFP, from a single-shot full-body
image. The methodology employs a holistic feature representation within a multi-modal learning
paradigm. The proposed solution undergoes meticulous validation and testing using real-world images,
including the simulation of various lighting conditions. The study also systematically examines the
significance of 2D and 3D features. To further enhance the performance of weight and height prediction,
future investigations can explore more rigorous methods for training and converging the multi-modal
architecture. Additionally, efforts can be made to improve the extraction of FFs (Feature Fusion), DFs
(Depth Fusion), and BFs (Body Fusion) embeddings. Exploring sub-embedding representation fusion
methods and designing approaches to predict height without scale information or constraints could also
contribute to improved prediction accuracy. Furthermore, future app development endeavors can focus
on fostering communities and addressing security concerns related to the Machine Learning model and
databases. These aspects will contribute to a more comprehensive and impactful implementation of the
solution.

Malnutrition Detection using Convolutional Neural Network Arun Raj Lakshminarayanan Department of
Computer Science and Engineering B. S. Abdur Rahman Crescent Institute of Science and Technology,
Chennai, India arunraj@crescent.education Saravanan Parthasarathy Department of Computer Science
and Engineering, B. S. Abdur Rahman Crescent Institute of Science and Technology, Chennai, India.
saravanan_cse_2019@crescent.educati on Pavani B Department of Computer Science and Engineering,
B. S. Abdur Rahman Crescent Institute of Science and Technology, Chennai, India
bollinenipavani000@gmail.com A.Abdul Azeez Khan Department of Computer Applications, B. S. Abdur
Rahman Crescent Institute of Science and Technology, Chennai, India.
abdulazeezkhan@crescent.education Rajeswari V Department of Computer Science and Engineering, B.
S. Abdur Rahman Crescent Institute of Science and Technology, Chennai, India.
Rajeswari06.vaddi@gmail.com K.Javubar Sathick Department of Computer Applications, B. S. Abdur
Rahman Crescent Institute of Science and Technology, Chennai, India. javubar@crescent.education
Abstract — Malnutrition is directly or indirectly responsible for the deaths of children younger than 5
years in many countries. Identification of malnourished children will help to prevent the risk of death
and can reduce physical and health issues by taking necessary measures or treatment. The proposed
system uses a Convolutional Neural Network (CNN), a Deep Learning algorithm that takes input,
analyzes the images, and differentiates one from the other. The architecture we used here is AlexNet for
the training process and Transfer Learning. The system takes the image of a child as the input and
classifies the image into a malnourished or normal child by comparing the image with the trained model.
The objective of the system is to detect malnutrition in children that can help people and healthcare
providers to reduce the effects caused by malnutrition by automation implementation instead of a
manual process. Keywords — Malnutrition, Convolutional Neural Network, Alexnet. I. INTRODUCTION
Malnutrition is a condition that occurs due to fewer intakes or over intake of nutrients. This can lead to
health issues such as diabetes, heart disease, eye problems, and stunted growth. Malnutrition is directly
or indirectly responsible for the deaths of children younger than 5 years in many countries. According to
the World Health Organization’s (WHO) 2020 edition, stunting has affected 21.3% or 144 million children
under 5 years of age globally. Wasting has affected or threatened the lives of 6.9% or 47 million children
under the same age group. About 5.6% or 38.3 million children under the age of 5 years were
overweight around the world [1]. Yet, while there has been an improvement, it has been slow and
patchy as shown in the Fig 1. Undernutrition leads to physical health issues and growth issues. According
to UNICEF’s The State of the World’s Children 2019 report [2], globally out of 3 children at least 1 is not
growing well due to malnutrition, and 1 out of 2 children is suffering from hidden hunger. Fig. 1.
Percentage and Number (millions) of stunted, wasted and overweighed under age 5 globally between
the years 2000-2019. Source: UNICEF, WHO, World Bank Group joint malnutrition estimates, 2020
edition Approximately 45% of deaths of children under the age of five years are due to undernutrition
[3]. The National Family Health Survey 2005-06 (NFHS-3) reported that in India, 48.0% of children under
age 5 years were stunted, 19.8% were wasted, 6.4% were severely wasted and 42.5% were
underweight[4]. The National Family Health Survey 2015-16 (NFHS-4) reported that 38.4% were stunted,
21.0% were wasted, 7.5% were severely wasted and 35.8% were underweight [5]. Over the decade
there was a slight decline in stunted and underweight children but wasted remains alarming. Children
were suffered more with malnutrition and the detection of malnutrition will help to prevent the risk of
death and can reduce physical or growth issues by taking necessary measures. Detection of malnutrition
in children can help people and healthcare providers to take preventive measures and can reduce the
effect caused by malnutrition on children. To detect the malnutrition in children, a type of artificial
neural network called convolutional neural network (CNN) is used. The architecture we used here is
AlexNet, which is a convolutional neural network that is a powerful model capable of achieving high
accuracies on challenging datasets.

VI. CONCLUSION AND FUTURE WORK Malnutrition is extravagantly high and has affected many countries
in the world by one or more forms. Detecting or predicting malnutrition will help the government or
health services to take preventive measures. Conventional Neural Network (CNN or ConvNet) algorithm
is used to detect the children affected with malnutrition under age five. Images of children are used as
input. Alexnet is a CNN used to find patterns in images to recognize faces and objects and performs
classification tasks. By using parametric conditions, Alexnet architecture, and with the help of the
extracted features, the system predicts whether children are affected by malnutrition or not. It also
categorizes the children who are at the risk of malnutrition. In the next phase of study, we would apply
the CNN algorithm to detect the type of malnutrition which affected the children. As the treatment
differs for each type, it would be helpful to the parents and healthcare providers.

------------------------------------------------------------------------------------------------------------------------------------------