Professional Documents
Culture Documents
An Internship
On
“REAL TIME PULSE DETECTION USING DEEP LEARNING”
Submitted in partial fulfillment of the requirement for the award of the degree of
Bachelor of Engineering
in
Computer Science and Engineering
Submitted By
Sheetal D
2JH20CS084
CERTIFICATE
Certified that the Internship / Professional Practice Entitled “REAL TIME PULSE
DETECTION USING DEEP LEARNING” carried out by PRIYANKA R BADIGER,
bearing USN 2JH20CS063, bonafide student of Jain College of Engineering and Technology, is in
partial fulfillment for the award of the BACHELOR OF ENGINEERING in Computer Science and
Engineering from Visvesvaraya Technological University, Belagavi during the year 2022-2023. It
is certified that all the corrections/suggestions indicated for Internal Assessment have been incorporated
in the report submitted in the department library. The report has been approved as it satisfies the
academicrequirements in respect of the Internship / Professional Practice prescribed for the said Degree.
1.
2.
JAIN COLLEGE OF ENGINEERING AND TECHNOLOGY
Department of Computer Science and Engineering
Sai Nagar, Hubballi – 580 031
DECLARATION
I, SHEETAL D, bearing USN 2JH20CS084, student of Eighth Semester B.E, Department of
Computer Science and Engineering, Jain College of Engineering and Technology, Hubballi,
declare that the Internship/ Professional Practice entitled “REAL TIME PULSE
DETECTION USING DEEP LEARNING” has been carried out by me and submitted in
partial fulfillment of the course requirements for the award of degree in Bachelor of
Engineering in Computer Science and Engineering from Visvesvaraya Technological
University, Belagavi during the academic year 2023 - 2024. The matter embodied in this
report has not been submitted to any other university or institution for the award of any other
degree.
SHEETAL D 2JH20CS084
Place: Hubballi
Date:
ABSTRACT
In recent years, deep learning techniques have revolutionized various fields, especially computer vision,
with their remarkable performance in tasks such as image classification and object detection. This paper
presents a novel approach for image classification and live detection using deep learning methods. The
proposed model leverages convolutional neural networks (CNNs), a powerful class of deep learning
architectures, to automatically learn hierarchical representations of image features.
The image classification component of the model involves training a CNN on a labeled dataset to
recognize and categorize different classes of images. Transfer learning techniques are employed to fine-tune
pre-trained CNN models, enabling efficient training even with limited annotated data. The trained model
achieves high accuracy in classifying images into predefined categories.
The effectiveness of the proposed image classification and live detection model is demonstrated through
comprehensive experiments on benchmark datasets and real-world scenarios. The results showcase the model's
robustness, accuracy, and efficiency in both offline image classification and real-time object detection tasks.
The proposed approach holds promise for various applications in fields such as security, automotive,
healthcare, and entertainment, paving the way for intelligent systems capable of understanding and interacting
with the visual world in real time.
i
ACKNOWLEDGEMENT
The satisfaction and the euphoria that accompany the successful completion of any task would be
incomplete without the mention of the people who made it possible. The constant guidance of these persons
and encouragement provide, crowned our efforts with success and glory. Although it is not possible to thank
all the members who helped for the completion of the internship/ Professional Practice work individually, I
take this opportunity to express my gratitude to one and all.
I express my sincere gratitude to Dr. Prashanth Banakar, Principal, Jain College of Engineering and
Technology for the support and encouragement.
I wish to place on record, my grateful thanks to Prof. Maheshkumar Patil. HOD, Department of CSE,
Jain College of Engineering and Technology, for the constant encouragement provided to me.
I am indebted with a deep sense of gratitude for the constant inspiration, encouragement, timely
guidance and valid suggestion given to me by my guide Prof. Rajeshwari Patil, Assistant Professor,
Department of CSE, Jain College of Engineering and Technology.
I am thankful to Svasti Innovation Labs for providing me an opportunityto carry out the internship/
Professional Practice work in their esteemed organization.
I am thankful to all the staff members of the department for providing relevant information and helping
in different capacities in carrying out this project.
Last, but not least, I owe my debts to my parents, friends and also those who directly or indirectly have
helped me to make the project work a success.
SHEETAL D 2JH20CS084
ii
TABLE OF CONTENT
1. Abstract i
2. Acknowledgement ii
4. List of Figures iv
3.1 Project 11
18
10. References
iii
LIST OF FIGURES
iv
REAL TIME PULSE DETECTION USING DEEP LEARNING
CHAPTER 1
Svasti Innovation Labs is a technology solution partner to provide innovative, cost-effective SMEs operation
solution to solve numerous thread level problems at different stages in supply chain domain.
The team started out journey in 2019 solving retail sector problems for mid-sized business in grocery and
FMCG in semi-urban and rural areas. Subsequently pivoted to build a technology stack including hardware to
tackle the thread level problems in supply chain.
Over last one year, the organization made speedy progress and has hooked out a place in cutting edge
technologies such as AI/ML, smart device in automation, usage of cloud etc.
Currently foot stepping as technology solution provider for different business models providing a tailor-made
software stack for the sector like E-Commerce, FMCG/Grocery commodities.
An intense focus on innovation in automation across all stage of operation through IoT, AI/ML, Blockchain,
cloudDB, Analytics to put Svasti Innovation Labs in a unique position to constantly add innovative solutions
to its portfolio.
WAREHOUSE AUTOMATION
The ability to accurately forecast upcoming demand is crucial for avoiding inventory overruns, overstaffing
costs, lost revenue due to customer dissatisfaction, and employee safety hazards. Today's shipping and
logistics management software relies almost exclusively on AI-driven solutions to aggregate data industry data
from multiple sources and develops robust, predictive analytics that can help avoid the common problems
mentioned above.
VIRTUAL ASSISTANT
A virtual assistant is an administrative professional who works remotely for a client (SMEs) or set of clients
(SMEs), handling appointment setting, inbound and outbound phone calls, scheduling, travel arrangements,
email, government compliances, and more.
Svasti Innovations Labs providing this unique service exclusively for SMEs to reduce burden of employees for
legal compliance, internet based work, software related works etc. On subscription based Svasti labs provide a
virtual private assistant
CHAPTER 2
Artificial intelligence (AI) is the ability of machines to replicate or enhance human intellect, such as reasoning
and learning from experience. Artificial intelligence has been used in computer programs for years, but it is
now applied to many other products and services. For example, some digital cameras can determine what
objects are present in an image using artificial intelligence software. In addition, experts predict many more
innovative uses for artificial intelligence in the future, including smart electric grids.
AI uses techniques from probability theory, economics, and algorithm design to solve practical problems. In
addition, the AI field draws upon computer science, mathematics, psychology, and linguistics. Computer
science provides tools for designing and building algorithms, while mathematics offers tools for modeling
and solving the resulting optimization problems.
Machine Learning (ML) makes computers learn from data and experience to improve their performance on
some tasks or decision-making processes. ML uses statistics and probability theory for this purpose. Machine
learning uses algorithms to parse data, learn from it, and make determinations without explicit programming.
Machine learning algorithms are often categorized as supervised or unsupervised. Supervised algorithms can
apply what has been learned in the past to new data sets; unsupervised algorithms can draw inferences from
datasets. Machine learning algorithms are designed to strive to establish linear and non-linear relationships in
a given set of data. This feat is achieved by statistical methods used to train the algorithm to classify or predict
from a dataset.
ML can help predict which of your customers are at risk for leaving in advance, and give youan edge by
pre-empting with action.
1. The Broad: Machine Learning is the process of predicting things, usually based on what they’ve
done in the past.
2. The Practical: Machine Learning tries to find relationships in your data that can help you
predict what will happen next.
3. The Technical: Machine Learning uses statistical methods to predict the value of a target variable
using a set of input data.
4. The Mathematical: Machine Learning attempts to predict the value of a variable Y given an input
offeature set X.
Machine Learning allows us to accurately predict things using simple statistical methods, algorithms,
and modern computing power.
In the era of digitalization, the exponential growth of image and video data has fueled the demand for
sophisticated computer vision techniques capable of understanding and interpreting visual information. Image
classification and object detection are two fundamental tasks in computer vision with numerous practical
applications across various domains, including healthcare, autonomous driving, surveillance, and
entertainment.
Image classification involves categorizing images into predefined classes or categories based on their visual
content. This task plays a pivotal role in content-based image retrieval, automatic tagging, and image
organization. Traditional image classification methods relied heavily on handcrafted features and shallow
learning algorithms, which often struggled to capture complex patterns and variations in visual data.
In contrast, recent advancements in deep learning, particularly convolutional neural networks (CNNs), have
revolutionized image classification by automatically learning hierarchical representations of image features
directly from raw pixel data. Deep learning-based approaches have demonstrated superior performance on
benchmark datasets, achieving state-of-the-art accuracy in image classification tasks.
Object detection, on the other hand, extends beyond image classification by not only identifying the objects
present in an image but also localizing their spatial positions with bounding boxes. This task is crucial for
applications such as video surveillance, autonomous navigation, and augmented reality, where the precise
identification and localization of objects are essential.
Traditional object detection methods often involved complex pipelines consisting of handcrafted feature
extraction, followed by separate object localization and classification steps. However, these approaches were
limited by their reliance on manual feature engineering and struggled to generalize across diverse object
categories and complex scenes.
Deep learning-based object detection techniques, such as Region-based Convolutional Neural Networks (R-
CNN), You Only Look Once (YOLO), and Single Shot MultiBox Detector (SSD), have overcome these
limitations by jointly optimizing object localization and classification within a unified framework. These
models leverage CNNs to efficiently process image regions and predict bounding boxes and class probabilities,
achieving remarkable accuracy and speed in object detection tasks.
In this report, we explore the principles, methodologies, and applications of deep learning-based image
classification and object detection. We discuss state-of-the-art techniques, recent advancements, and
challenges in these domains, highlighting their significance in modern computer vision research and real-
world applications. Additionally, we present a novel approach for integrating image classification and live
object detection capabilities using deep learning, showcasing its potential for various practical scenarios and
future advancements in computer vision technology.
ARCHITECTURE DIAGRAM
• System architecture refers to the overall design and structure of a complex system, such as a
computer system, a software application, a building, or a network.
• It involves the organization of components, the relationships between them, and the principles and
guidelines that govern their design and implementation.
The above figure shows Online transaction fraud detection is a critical problem in e-commerce and other online
services. Machine learning techniques can be used to build models that detect fraudulent transactions in real-time.
Here is a possible architecture for an online transaction fraud detection system:
1)Data Collection: Collecting data is the first step in building any machine learning model. In the context of
online transaction fraud detection, data can be collected from various sources such as payment gateways, logs,
user profiles, and historical transaction data.
2)Data Pre-processing: Once data is collected, it needs to be cleaned and pre-processed. This step involves
removing duplicate or irrelevant data, handling missing values, and converting data into a suitable format for
machine learning algorithms.
3)Feature Extraction: Extracting relevant features from the data is an important step in building accurate models.
Features could include transaction amounts, device information, location, time of day, and user behavior patterns.
4)Model Training: After feature extraction, machine learning models can be trained on the pre-processed data.
Various supervised learning algorithms such as logistic regression, decision trees, random forests, and neural
networks can be used to build models.
5)Model Evaluation: The trained models need to be evaluated using appropriate metrics such as precision, recall,
accuracy, and F1-score. The evaluation results can be used to fine-tune the model parameters and select the best
performing models.
6)Model Deployment: Once the models are trained and evaluated, they can be deployed into production. In this
step, the models are integrated with the online transaction processing system and used to detect fraudulent
transactions in real-time.
7)Monitoring and Maintenance: The deployed models need to be monitored continuously to ensure that they
are performing accurately. If the performance drops, retraining the models with new data or modifying the model
parameters may be required.
In summary, an online transaction fraud detection system based on machine learning requires data collection, pre-
processing, feature extraction, model training, model evaluation, model deployment, and monitoring and
maintenance. The specific techniques and algorithms used in each step will depend on the nature of the data and
the requirements of the business.
ALGORITHMS USED
Convolutional Neural Networks (CNNs) have emerged as a cornerstone in the field of computer vision,
revolutionizing the way machines perceive and understand visual data. Inspired by the hierarchical
organization of the visual cortex in animals, CNNs are specifically designed to process structured grid data,
such as images, by learning hierarchical representations of features directly from raw pixel data. Their ability
to automatically extract meaningful features from images has led to significant advancements in tasks like
image classification, object detection, and image segmentation.
Convolutional Layers:
At the heart of CNNs are convolutional layers, which are responsible for extracting features from input
images. Convolution involves sliding small filters or kernels across the input image and computing dot
products between the filter weights and local input regions. These filters act as feature detectors, capturing
spatial patterns such as edges, textures, and shapes. By stacking multiple convolutional layers, the network
can learn increasingly abstract and complex features, enabling it to discriminate between different classes of
objects.
Following the convolution operation, an activation function is applied element-wise to introduce non-
linearity into the network. The most commonly used activation function is the rectified linear unit (ReLU),
which replaces negative values with zeros, effectively introducing sparsity and facilitating faster convergence
during training. The non-linear activation function allows CNNs to model complex relationships between
features, enabling them to learn highly discriminative representations of visual data.
Pooling Layers:
Pooling layers are interspersed between convolutional layers to progressively reduce the spatial dimensions
of feature maps while retaining important information. Pooling operations, such as max pooling or average
pooling,
Downs ample feature maps by selecting the maximum or average value within each pooling window. This
spatial reduction helps in controlling the model's complexity, improving computational efficiency, and
increasing the receptive field of higher-level features.
Towards the end of the CNN architecture, one or more fully connected layers are typically employed to
perform high-level feature aggregation and classification. These layers connect every neuron in one layer to
every neuron in the next layer, enabling the network to learn complex non-linear mappings between features
and class labels. Fully connected layers are often followed by a soft max activation function, which outputs
the probability distribution over the predefined classes, facilitating multi-class classification.
CHAPTER 3
TASK PERFOREMED DURING INTERNSHIP
3.1 Project 1: This project explains about the building of image classification model using
TensorFlow.
Explanation of code and how it works
1) IMPORTING THE DEPENDENCIES
3) DATASET INFORMATION
CHAPTER 4
REFLECTION NOTES
Interning at Svasti Innovation labs for AIML has been an incredibly rewarding and motivating experience.
With empathetic, compassionate, and supportive mentors, I was able to complete my internship on Deep
Learning while facing a range of challenges, including:
Here are some of the skills I acquired during my time at Svasti Innovation labs:
Throughout my internship, I was able to develop and foster a positive and compassionate learning environment
with the support and mentorship of my cooperating teacher. With a focus on time management, organization,
discipline, and consistent practice, my development skills improved greatly.
Along with development skills, critical and analytical thinking directly impacted my knowledge gain. Through
engaging activities and supportive learning methodologies, I was able to make significant gains in both reading
comprehension and fluency. I also made substantial progress in my speaking skills, including pronunciation
and confidence.
Some key advantages of interning at Svasti Innovation labs for AIML are:
Overall, my experience at Svasti Innovation labs was excellent. Tasks were clearly defined both in terms of
implementation and learning, resources for learning were well-structured, and mentors were always available.
Structured and well-documented notes were provided, and I learned various approaches for converting a
trained model to an end product. Deployment of AIML products and workloads was also taught well on
multiple cloud platforms.
CONCLUSION
In conclusion, this project has delved into the intricacies of Convolutional Neural Networks (CNNs) and their
applications in image classification and live detection using deep learning techniques. By elucidating the
architecture of CNNs and their ability to automatically learn hierarchical representations of features from raw
pixel data, the project has showcased their versatility and effectiveness in addressing diverse computer vision
tasks. Through the integration of transfer learning for image classification and real-time object detection
algorithms such as You Only Look Once (YOLO) or Single Shot MultiBox Detector (SSD) for live detection,
CNNs have proven indispensable in modern computer vision pipelines. Looking forward, the continual
evolution of CNN architectures and training methodologies promises further advancements, paving the way
for intelligent systems capable of understanding and interacting with the visual world in increasingly
sophisticated ways.
REFERENCES
[1] Krizhevsky, Alex, Ilya Sutskever, and Geoffrey E. Hinton. "ImageNet classification with deep
convolutional neural networks." Communications of the ACM 60.6 (2017): 84-90. - This seminal paper
introduced the AlexNet architecture, which demonstrated the power of CNNs for image classification by
achieving a significant improvement in accuracy on the ImageNet dataset.
[2] Simonyan, Karen, and Andrew Zisserman. "Very deep convolutional networks for large-scale image
recognition." arXiv preprint arXiv:1409.1556 (2014). - This paper proposed the VGGNet architecture,
consisting of very deep convolutional layers, which achieved state-of-the-art performance on the ImageNet
dataset and provided insights into the importance of network depth in CNNs.
[3] Redmon, Joseph, and Santosh Divvala. "You only look once: Unified, real-time object detection."
Proceedings of the IEEE conference on computer vision and pattern recognition. 2016. - This paper
introduced the YOLO (You Only Look Once) algorithm, a real-time object detection system that processes
images in a single pass, achieving high accuracy and speed.
[4] Liu, Wei, et al. "SSD: Single shot multibox detector." European conference on computer vision.
Springer, Cham, 2016. - This paper introduced the Single Shot MultiBox Detector (SSD) algorithm,
another real-time object detection method that combines multiple detection layers to efficiently detect
objects at various scales and aspect ratios.
[5] Szegedy, Christian, et al. "Going deeper with convolutions." Proceedings of the IEEE conference on
computer vision and pattern recognition. 2015. - This paper introduced the Inception architecture (also
known as GoogLeNet), which introduced the concept of inception modules for efficient utilization of
computational resources in deep CNNs.