lOMoARcPSD|42949913

Sample 1 DL Internship Report Gtm

Computer Science Engineering (Ram Krishna Dharmarth Foundation University)

Scan to open on Studocu

Studocu is not sponsored or endorsed by any college or university

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)
 lOMoARcPSD|42949913

DEEP LEARNING
(To Identify Ships In A Satellite Image)
Summer Internship Report Submitted in partial fulfilment
of the requirement for undergraduate degree of

Bachelor of Technology

In

COMPUTER SCIENCE AND ENGINEERING

By

Kasarla Sahith Reddy

221710308025

Under the Guidance of

Mr. D. Srinivasa Rao
Assistant Professor

Department Of Computer Science and Engineering

GITAM School of Technology
GITAM (Deemed to be University)
Hyderabad-502329
July 2020

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

DECLARATION

I submit this industrial training work entitled “To Identify Ships In A Satellite Image”
to GITAM (Deemed To Be University), Hyderabad in partial fulfilment of the requirements for
the award of the degree of “Bachelor of Technology” in “Computer Science and
Engineering”. I declare that it was carried out independently by me under the guidance of Mr.
D. Srinivasa Rao, Asst. Professor, GITAM (Deemed To Be University), Hyderabad, India.

The results embodied in this report have not been submitted to any other University or
Institute for the award of any degree or diploma.

Place: HYDERABAD Name: Kasarla Sahith Reddy

Date: 21-07-2020 Student Roll No: 221710308025

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

GITAM (DEEMED TO BE UNIVERSITY)

Hyderabad-502329, India
Dated: 21 -07-2020

CERTIFICATE

This is to certify that the Industrial Training Report entitled “To Identify
Ships In A Satellite Image” is being submitted by Kasarla Sahith Reddy (221710308025) in
partial fulfilment of the requirement for the award of Bachelor of Technology in Computer
Science And Engineering at GITAM (Deemed to Be University), Hyderabad during the
academic year 2020-2021.

It is faithful record work carried out by him at the Computer Science And
Engineering Department, GITAM University Hyderabad Campus under my guidance and
supervision.

D. Srinavasa Rao Dr. S. Phani Kumar

Assistant Professor Professor and HOD

Department of CSE Department of CSE

II

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

ACKNOWLEDGEMENT

Apart from my effort, the success of this internship largely depends on the
encouragement and guidance of many others. I take this opportunity to express my gratitude to
the people who have helped me in the successful competition of this internship.

I would like to thank respected Dr. N. Siva Prasad, Pro Vice Chancellor, GITAM
Hyderabad and Dr. N. Seetharamaiah, Principal, GITAM Hyderabad.

I would like to thank respected S. Phani Kumar, Head of the Computer Science and
Engineering for giving me such a wonderful opportunity to expand my knowledge for my own
branch and giving me guidelines to present an internship report. It helped me a lot to realize of
what we study for.

I would like to thank the respected faculties Mr. D.Srinivasa Rao who helped me to
make this internship a successful accomplishment.

I would also like to thank my friends who helped me to make my work more organized
and well-stacked till the end.

Kasarla Sahith Reddy

221710308025

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

ABSTRACT

Deep learning has achieved great success in many fields, such as computer vision and
natural language processing. Compared to traditional machine learning methods, deep learning
has a strong learning ability and can make better use of datasets for feature extraction. Image
recognition is one of the most important fields of image processing and computer vision which
can be achieved by deep learning.

To Identifying Ships in Satellite Images is the main motive of this project. Ship
identification on satellite imagery can be used for fisheries management, monitoring of
smuggling activities, ship traffic services, and naval warfare. Real Dataset is collected from
website. I have adapted the view point of looking at features of the dataset, for deep
understanding of the problem. My primary objective of this case study was to to classify
whether the satellite image is consisting of ship or not. I have chosen Convolutional Neural
Network (CNN) method, which has the advantage of being able to extract features
automatically and produce reliable prediction for ship identification. Scaling is used for data as
it is used to normalize and reduces computational complexity Conclusion is made on the base
of achieve as higher accuracy as possible and correctly classifying unknown satellite images of
ships which are not present in training dataset.

III

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

Table of Contents

CHAPTER 1: MACHINE LEARNING 1

1.1. Introduction 1
1.2. Importance of Machine Learning 1
1.3. Uses of Machine Learning 2
1.4. Types of Learning Algorithms 2
1.4.1. Supervised Learning 3
1.4.2. Unsupervised Learning 3
1.4.3. Semi Supervised Learning 4

CHAPTER 2: DEEP LEARNING 6

2.1. Deep Learning and It’s Importance: 6
2.2. Uses of Deep Learning 7
2.3. Relation between Data Mining, Machine Learning and Deep
Learning 9

CHAPTER 3: PYTHON 11
3.1. Introduction To Python 11
3.1.1. History Of Python 11
3.2. How to Setup Python 11
3.2.1. Installation(using python IDLE) 11
3.2.2. Installation(using Anaconda) 12
3.3. Features Of Python 13
3.4. Python Variable Types 14
3.4.1. Python Numbers 14
IV

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

3.4.2. Python Strings 15

3.4.3. Python Lists 15
3.4.4. Python Tuples 15
3.4.5. Python Dictionary 16
3.5. Python Functions 16
3.5.1. Defining a Function 16
3.5.2. Calling a Function 17
3.6. Python Using OOPs Concepts 17
3.6.1. Class 17
3.6.2. __init__Method In Class: 18

CHAPTER 4: TO IDENTIFY SHIPS IN A SATELLITE IMAGE 19

4.1 Project Requirements 19
4.1.1. Packages used 19
4.1.2. Versions of the packages 20
4.1.3. Algorithms used 20
4.2. Problem Statement 21
4.3. Dataset Description 21
4.4. Objective of the Case Study 22

CHAPTER 5: DATA PREPROCESSING/FEATURE ENGINEERING

AND EDA 23

5.1. Loading The Data: 23

5.2. Statistical Analysis 24
5.3. Visualisation Of Images: 25
5.4. Handling Missing Values: 28

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

CHAPTER 6: FEATURE SELECTION 29

6.1. Select Relevant Features For Analysis 29
6.2. Drop Irrelevant Features 29
6.3. Converting Data and Reshaping 30
6.4 Feature Scaling 31

CHAPTER 7: MODEL BUILDING AND EVALUATION 33

7.1. Brief About The Algorithms Used And Building Model 33
7.2.Train the Models 40
7.3.Validate the Models 42
7.4. Make Predictions 45

CONCLUSION 49

REFERENCES 50

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

List of Figures

FIG 1.2.1 The process flow 2

FIG 1.4.2.1 Unsupervised learning 4

FIG 1.4.3.1 Semi supervised learning 5

FIG 2.1.1 Deep neural network 7

FIG 2.2.1 The deep learning process 9

FIG 2.3.1 Relation between DM,ML,DL 9

FIG 2.3.2 Process in machine learning and deep learning 10

FIG 3.2.1.1 Python download 12

FIG 3.2.2.1 Anaconda download 13

FIG 3.2.2.2 Jupyter notebook 13

FIG 3.6.1.1 Defining a class 17

FIG 4.1.1.1 Packages used 20

FIG 4.1.2.1 Versions of packages 20

FIG 4.3.1 Sample ship images 21

FIG 4.3.2 Sample no ship images 22

FIG 5.1.1 Reading data from json file 23

FIG 5.1.2 Loading data set 24

FIG 5.2.1 Statistical data of labels 25

FIG 5.2.2 Checking for total number of ship and non ship images 25

FIG 5.3.1 Contents of file 26

FIG 5.3.2 Creating directories 26

FIG 5.3.3 Displaying 5 images names 26

FIG 5.3.4 Displaying sample image 27

FIG 5.3.5 Input for visualization of 15 images 27

FIG 5.3.6 Output for visualization of 15 images 28

FIG 5.4.1 No missing values in the dataset 28

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

FIG 6.2.1 Dropping columns 30

FIG 6.3.1 Converting of data into numpy array 30

FIG 6.3.2 Checking shape of data 31

FIG 6.3.3 Reshaping of data 31

FIG 6.4.1 Scaling of data 31

FIG 7.1.1 Illustration of convolution over feature map 34

FIG 7.1.2 A 5x5 input feature map and 3x3 convolution of depth1 34

FIG 7.1.3 Example of 3x3 convolution performed over 5x5 input feature map 35

FIG 7.1.4 Example of max pooling 36

FIG 7.1.5 Importing model and other parameters 37

FIG 7.1.6 Building model 38

FIG 7.1.7 Output of built model 39
FIG 7.1.8 Compiling the model 40
41
FIG 7.2.1 Training the model
FIG 7.2.2 Output of training the model 42
FIG 7.3.1 Train accuracy vs validation accuracy 43
FIG 7.3.2 Train loss vs validation loss 44

FIG 7.4.1 Unknown sample of ship image 45

FIG 7.4.2 Resizing and scaling unknown sample image 46
FIG 7.4.3 Predicting the ship image 47
FIG 7.4.4 Unknown sample of no ship image 47
FIG 7.4.5 Predicting the no ship image 48

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

CHAPTER 1: MACHINE LEARNING

1.1 INTRODUCTION:

Machine Learning(ML) is the scientific study of algorithms and statistical models that
computer systems use in order to perform a specific task effectively without using explicit
instructions, relying on patterns and inference instead. It is seen as a subset of Artificial
Intelligence(AI).

1.2 IMPORTANCE OF MACHINE LEARNING:

Consider some of the instances where machine learning is applied: the self-driving
Google car, cyber fraud detection, online recommendation engines—like friend suggestions on
Facebook, Netflix showcasing the movies and shows you might like, and “more items to
consider” and “get yourself a little something” on Amazon—are all examples of applied
machine learning. All these examples echo the vital role machine learning has begun to take in
today’s data-rich world.

Machines can aid in filtering useful pieces of information that help in major
advancements, and we are already seeing how this technology is being implemented in a wide

variety of industries.

With the constant evolution of the field, there has been a subsequent rise in the uses,
demands, and importance of machine learning. Big data has become quite a buzzword in the
last few years; that’s in part due to increased sophistication of machine learning, which helps
analyse those big chunks of big data. Machine learning has also changed the way data
extraction, and interpretation is done by involving automatic sets of generic methods that have
replaced traditional statistical techniques. The process flow depicted here represents how
machine learning works

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

Fig 1.2.1: The Process Flow

1.3 USES OF MACHINE LEARNING:

Earlier in this article, we mentioned some applications of machine learning. To

understand the concept of machine learning better, let’s consider some more examples: web
search results, real-time ads on web pages and mobile devices, email spam filtering, network
intrusion detection, and pattern and image recognition. All these are by-products of applying
machine learning to analyse huge volumes of data

Traditionally, data analysis was always being characterized by trial and error, an
approach that becomes impossible when data sets are large and heterogeneous. Machine
learning comes as the solution to all this chaos by proposing clever alternatives to analyzing
huge volumes of data.

By developing fast and efficient algorithms and data-driven models for real-time
processing of data, machine learning can produce accurate results and analysis.

1.4 TYPES OF LEARNING ALGORITHMS:

The types of machine learning algorithms differ in their approach, the type of data they
input and output, and the type of task or problem that they are intended to solve.

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

1.4.1 Supervised Learning :

When an algorithm learns from example data and associated target responses that
can consist of numeric values or string labels, such as classes or tags, in order to later predict
the correct response when posed with new examples comes under the category of supervised
learning.

Supervised machine learning algorithms uncover insights, patterns, and relationships

from a labelled training dataset – that is, a dataset that already contains a known value for the
target variable for each record. Because you provide the machine learning algorithm with the
correct answers for a problem during training, it is able to “learn” how the rest of the features
relate to the target, enabling you to uncover insights and make predictions about future
outcomes based on historical data.

Examples of Supervised Machine Learning Techniques are Regression, in which the

algorithm returns a numerical target for each example, such as how much revenue will be
generated from a new marketing campaign.

Classification, in which the algorithm attempts to label each example by choosing

between two or more different classes. Choosing between two classes is called binary
classification, such as determining whether or not someone will default on a loan. Choosing
between more than two classes is referred to as multiclass classification.

1.4.2 Unsupervised Learning:

When an algorithm learns from plain examples without any associated response,
leaving to the algorithm to determine the data patterns on its own. This type of algorithm tends
to restructure the data into something else, such as new features that may represent a class or a
new series of uncorrelated values. They are quite useful in providing humans with insights into
the meaning of data and new useful inputs to supervised machine learning algorithms.

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

Fig 1.4.2.1: Unsupervised Learning

Popular techniques where unsupervised learning is used also include self-organizing

maps, nearest neighbour mapping, singular value decomposition, and k-means clustering.
Basically, online recommendations, identification of data outliers, and segment text topics are
all examples of unsupervised learning.

1.4.3 Semi Supervised Learning:

As the name suggests, semi-supervised learning is a bit of both supervised and

unsupervised learning and uses both labelled and unlabelled data for training. In a typical
scenario, the algorithm would use a small amount of labelled data with a large amount of
unlabelled data.

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

Fig 1.4.3.1: Semi Supervised Learning

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

CHAPTER 2: DEEP LEARNING

2.1 DEEP LEARNING AND IT’S IMPORTANCE:

Deep learning algorithms run data through several “layers” of neural network
algorithms, each of which passes a simplified representation of the data to the next layer. Most
machine learning algorithms work well on datasets that have up to a few hundred features, or
columns.

Basically deep learning is itself a subset of machine learning but in this case the machine
learns in a way in which humans are supposed to learn. The structure of deep learning model
is highly similar to a human brain with large number of neurons and nodes like neurons in
human brain thus resulting in artificial neural network. In applying traditional machine learning
algorithms we have to manually select input features from complex data set and then train them
which becomes a very tedious job for ML scientist but in neural networks we don’t have to
manually select useful input features, there are various layers of neural networks for handling
complexity of the data set and algorithm as well.

In one of the recent project on human activity recognition , when applied traditional
machine learning algorithm like K-NN then we have to separately detect human and its activity
also had to select impactful input parameters manually which became a very tedious task as
data set was way too complex but the complexity dramatically reduced on applying artificial
neural network, such is the power of deep learning. Yes it’s correct that deep learning
algorithms take lots of time for training sometimes even weeks as well but its execution on new
data is so fast that its not even comparable with traditional ML algorithms.

Deep learning has enabled Industrial Experts to overcome challenges which were
impossible, a decades ago like Speech and Image recognition and Natural Language
Processing. Majority of the Industries are currently depending on it , be it Journalism,
Entertainment, Online Retail Store, Automobile, Banking and Finance, Healthcare,
Manufacturing or even Digital Sector. Video recommendations, Mail Services, Self Driving
cars, Intelligent Chat bots, Voice Assistants are just trending achievements of Deep Learning.

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

Furthermore, Deep learning can most profoundly be considered as future of Artificial

Intelligence due to constant rapid increase in amount of data as well as the gradual development
in hardware field as well, resulting in better computational power.

Fig 2.1.1: Deep Neural network

2.2 USES OF DEEP LEARNING:

i. Translations:

Although automatic machine translation isn’t new, deep learning is helping enhance
automatic translation of text by using stacked networks of neural networks and allowing
translations from images.

ii. Adding colour to black-and-white images and videos:

It is used to be a very time-consuming process where humans had to add colour to

black-and-white images and videos by hand can now be automatically done with deep-learning
models.
iii. Language recognition:

Deep learning machines are beginning to differentiate dialects of a language. A machine

decides that someone is speaking English and then engages an AI that is learning to tell the
differences between dialects.
Once the dialect is determined, another AI will step in that specializes in that particular
dialect. All of this happens without involvement from a human.

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

iv. Autonomous vehicles:

There's not just one AI model at work as an autonomous vehicle drives down the street.
Some deep-learning models specialize in streets signs while others are trained to recognize
pedestrians. As a car navigates down the road, it can be informed by up to millions of individual
AI models that allow the car to act.

v. Computer vision:

Deep learning has delivered super-human accuracy for image classification, object
detection, image restoration and image segmentation—even handwritten digits can be
recognized. Deep learning using enormous neural networks is teaching machines to automate
the tasks performed by human visual systems.

vi. Text generation:

The machines learn the punctuation, grammar and style of a piece of text and can use
the model it developed to automatically create entirely new text with the proper spelling,
grammar and style of the example text. Everything from Shakespeare to Wikipedia entries have
been created.

vii. Deep-learning robots:

Deep-learning applications for robots are plentiful and powerful from an impressive
deep-learning system that can teach a robot just by observing the actions of a human
completing a task to a housekeeping robot that’s provided with input from several other AIs
in order to take action.
Just like how a human brain processes input from past experiences, current input from senses
and any additional data that is provided, deep-learning models will help robots execute tasks
based on the input of many different AI opinions.

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

Fig 2.2.1: The Deep Learning Process

2.3 RELATION BETWEEN DATA MINING, MACHINE LEARNING

AND DEEP LEARNING:

Fig 2.3.1: Relation between DM,ML, DL

The deep learning, data mining and machine learning share a foundation in data science,
and there certainly is overlap between the two. Data mining can use machine learning
algorithms to improve the accuracy and depth of analysis, and vice-versa; machine learning
can use mined data as its foundation, refining the dataset to achieve better results.

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

You could also argue that data mining and machine learning are similar in that they
both seek to address the question of how we can learn from data. However, the way in which
they achieve this end, and their applications, form the basis of some significant differences.

Fig 2.3.2: Process in machine learning and deep learning

Machine Learning comprises of the ability of the machine to learn from trained data set
and predict the outcome automatically. It I a subset of artificial intelligence.

Deep Learning is a subset of machine learning. It works in the same way on the machine
just like how the human brain processes information. Like a brain can identify the patterns by
comparing it with previously memorized patterns, deep learning also uses this concept.

Deep learning can automatically find out the attributes from raw data while machine
learning selects these features manually which further needs processing. It also employs
artificial neural networks with many hidden layers, big data, and high computer resources.

Data Mining is a process of discovering hidden patterns and rules from the existing
data. It uses relatively simple rules such as association, correlation rules for the decision-
making process, etc. Deep Learning is used for complex problem processing such as voice
recognition etc. It uses Artificial Neural Networks with many hidden layers for processing. At
times data mining also uses deep learning algorithms for processing the data.

10

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

CHAPTER 3 : PYTHON

Basic programming language used for machine learning is : PYTHON

3.1 INTRODUCTION TO PYTHON:

 Python is a high-level, interpreted, interactive and object-oriented scripting language.

 Python is a general purpose programming language that is often applied in scripting

roles
 Python is Interpreted: Python is processed at runtime by the interpreter. You do not
need to compile your program before executing it. This is like PERL and PHP.
 Python is Interactive: You can sit at a Python prompt and interact with the interpreter
directly to write your programs.
 Python is Object-Oriented: Python supports the Object-Oriented style or technique of
programming that encapsulates code within objects.

3.1.1 History of python:

 Python was developed by GUIDO VAN ROSSUM in early 1990’s

 Its latest version is 3.7 , it is generally called as python3

3.2 HOW TO SETUP PYTHON:

 Python is available on a wide variety of platforms including Linux and Mac OS X. Let's
understand how to set up our Python environment.
 The most up-to-date and current source code, binaries, documentation, news, etc., is
available on the official website of Python.

3.2.1 Installation(using python IDLE):

 Installing python is generally easy, and nowadays many Linux and Mac OS
distributions include a recent python.

11

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

 Download python from www.python.org

 When the download is completed, double click the file and follow the instructions to
install it.
 When python is installed, a program called IDLE is also installed along with it. It
provides a graphical user interface to work with python.

Fig 3.2.1.1: Python download

3.2.2 Installation(using Anaconda):

 Python programs are also executed using Anaconda.

 Anaconda is a free open source distribution of python for large scale data processing,
predictive analytics and scientific computing.
 Conda is a package manager quickly installs and manages packages.
 In WINDOWS:
 Step 1: Open Anaconda.com/downloads in web browser.
 Step 2: Download python 3.4 version for (32-bitgraphic installer/64 -bit graphic
installer)
 Step 3: select installation type( all users)
 Step 4: Select path(i.e. add anaconda to path & register anaconda as default python 3.4)
next click install and next click finish
 Step 5: Open jupyter notebook ( it opens in default browser)

12

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

Fig 3.2.2.1: Anaconda download

Fig 3.2.2.2: Jupyter notebook

3.3 FEATURES OF PYTHON:

 Easy-to-learn: Python has few keywords, simple structure, and a clearly defined syntax,
This allows the student to pick up the language quickly.
 Easy-to-read: Python code is more clearly defined and visible to the eyes.
 Easy-to-maintain: Python's source code is fairly easy-to-maintaining.
 A broad standard library: Python's bulk of the library is very portable and cross-
platform compatible on UNIX, Windows, and Macintosh.

13

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

 Portable: Python can run on a wide variety of hardware platforms and has the same
interface on all platforms.
 Extendable: You can add low-level modules to the Python interpreter. These modules
enable programmers to add to or customize their tools to be more efficient.
 Databases: Python provides interfaces to all major commercial databases.
 GUI Programming: Python supports GUI applications that can be created and ported to
many system calls, libraries and windows systems, such as Windows MFC, Macintosh,
and the X Window system of Unix.

3.4 PYTHON VARIABLE TYPES:

Variables are nothing but reserved memory locations to store values. This means that when
you create a variable you reserve some space in memory. Based on the data type of a variable,
the interpreter allocates memory and decides what can be stored in the reserved memory.
Therefore, by assigning different data types to variables, you can store integers, decimals or
characters in these variables.

Python has five standard data types –

 Numbers
 Strings
 Lists
 Tuples
 Dictionary

3.4.1 Python Numbers:

 Number data types store numeric values. Number objects are created when you assign
a value to them.
 Python supports four different numerical types − int (signed integers) long (long
integers, they can also be represented in octal and hexadecimal) float (floating point
real values) complex (complex numbers).

14

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

3.4.2 Python Strings:

 Strings in Python are identified as a contiguous set of characters represented in the

quotation marks.
 Python allows for either pairs of single or double quotes.
 Subsets of strings can be taken using the slice operator ([ ] and [:] ) with indexes starting
at 0 in the beginning of the string and working their way from -1 at the end.
 The plus (+) sign is the string concatenation operator and the asterisk (*) is the repetition
operator.

3.4.3 Python Lists:

 Lists are the most versatile of Python's compound data types.

 A list contains items separated by commas and enclosed within square brackets ([]).
 To some extent, lists are similar to arrays in C. One difference between them is that all
the items belonging to a list can be of different data type.
 The values stored in a list can be accessed using the slice operator ([ ] and [:]) with
indexes starting at 0 in the beginning of the list and working their way to end -1.
 The plus (+) sign is the list concatenation operator, and the asterisk (*) is the repetition
operator.

3.4.4 Python Tuples:

 A tuple is another sequence data type that is similar to the list.

 A tuple consists of a number of values separated by commas. Unlike lists, however,

tuples are enclosed within parentheses.
 The main differences between lists and tuples are: Lists are enclosed in brackets ( [ ] )
and their elements and size can be changed, while tuples are enclosed in parentheses (
( ) ) and cannot be updated.
 Tuples can be thought of as read-only lists.
 For example − Tuples are fixed size in nature whereas lists are dynamic. In other words,
a tuple is immutable whereas a list is mutable. You can't add elements to a tuple.

15

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

 Tuples have no append or extend method. You can't remove elements from a tuple.
Tuples have no remove or pop method.

3.4.5 Python Dictionary:

 Python's dictionaries are kind of hash table type.

 They work like associative arrays or hashes found in Perl and consist of key-value pairs.
 A dictionary key can be almost any Python type, but are usually numbers or strings.
Values, on the other hand, can be any arbitrary Python object.
 Dictionaries are enclosed by curly braces ({ }) and values can be assigned and accessed
using square braces ([]).
 You can use numbers to "index" into a list, meaning you can use numbers to find out
what's in lists. You should know this about lists by now, but make sure you understand
that you can only use numbers to get items out of a list.
 What a dict does is let you use anything, not just numbers. Yes, a dict associates one
thing to another, no matter what it is.

3.5 PYTHON FUNCTION:

3.5.1 Defining a Function:

You can define functions to provide the required functionality. Here are simple rules
to define a function in Python. Function blocks begin with the keyword def followed by the
function name and parentheses (i.e.()).

Any input parameters or arguments should be placed within these parentheses. You can
also define parameters inside these parentheses

The code block within every function starts with a colon (:) and is indented. The
statement returns [expression] exits a function, optionally passing back an expression to the
caller. A return statement with no arguments is the same as return None.

16

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

3.5.2 Calling a Function:

Defining a function only gives it a name, specifies the parameters that are to be
included in the function and structures the blocks of code. Once the basic structure of a function
is finalized, you can execute it by calling it from another function or directly from the Python
prompt.

3.6 PYTHON USING OOP’s CONCEPTS:

3.6.1 Class:

A user-defined prototype for an object that defines a set of attributes that characterize any object
of the class. The attributes are data members (class variables and instance variables) and
methods, accessed via dot notation.

● Class variable: A variable that is shared by all instances of a class. Class variables are
defined within a class but outside any of the class's methods. Class variables are not used
as frequently as instance variables are.
● Data member: A class variable or instance variable that holds data associated with a
class and its objects.
● Instance variable: A variable that is defined inside a method and belongs only to the
current instance of a class.
● Defining a Class:
● We define a class in a very similar way how we define a function.
● Just like a function ,we use parentheses and a colon after the class name(i.e. ():).

Fig 3.6.1.1: Defining a Class

17

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

3.6.2 __init__ method in Class:

● The init method — also called a constructor — is a special method that runs when an
instance is created so we can perform any tasks to set up the instance.
● The init method has a special name that starts and ends with two underscores:__init__().

18

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

CHAPTER 4: TO IDENTIFY SHIPS IN A SATELLITE IMAGE

4.1. PROJECT REQUIREMENTS:

4.1.1. Packages used:

Numpy: In Python we have lists that serve the purpose of arrays, but they are slow to process.
Numpy aims to provide an array object that is up to 50x faster that traditional Python lists. The
array object in Numpy is called ndata, it provides a lot of supporting functions that make
working with ndarray very easy. Arrays are very frequently used in data science, where speed
and resources are very important.

Pandas: Pandas is an open-source Python Library providing high-performance data

manipulation and analysis tool using its powerful data structures. The name Pandas is derived
from the word Panel Data – an Econometrics from Multidimensional data. Python with Pandas
is used in a wide range of fields including academic and commercial domains including
finance, economics, Statistics, analytics, etc.

Matplotlib: Matplotlib is one of the most popular Python packages used for data visualization.
It is a cross-platform library for making 2D plots from data in arrays. Matplotlib is written in
Python and makes use of Numpy, the numerical mathematics extension of Python. It provides
an object-oriented API that helps in embedding plots in applications using Python GUI toolkits
such as PyQt, WxPythonotTkinter. It can be used in Python and IPython shells, Jupyter
notebook and web application servers also Matplotlib has a procedural interface named the
Pylab, which is designed to resemble MATLAB, a proprietary programming language
developed by MathWorks. Matplotlib along with Numpy can be considered as the open source
equivalent of MATLAB.

TensorFlow: TensorFlow is an end-to-end open source platform for machine learning. It has
a comprehensive, flexible ecosystem of tools, libraries and community resources that lets
researchers push the state-of-the-art in ML and developers easily build and deploy ML powered
applications.

19

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

OS: The OS module in python provides functions for interacting with the operating
system. OS, comes under Python's standard utility modules. This module provides a portable
way of using operating system dependent functionality.

JSON: Python has a built-in package called json, which can be used to work with JSON data.

Fig 4.1.1.1: Packages used

4.1.2. Versions of the packages:

The versions of the packages are found by following command

Fig 4.1.2.1: Versions of packages

4.1.3. Algorithms used:

Convolutional Neural Network (CNN): The CNN algorithm is efficient at recognition and
highly adaptable. It’s also easy to train because there are fewer training parameters, and is
scalable when coupled with backpropagation.

20

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

4.2 PROBLEM STATEMENT:

The aim of this dataset is to help difficult task of detecting the ships in satellite images.

4.3 DATASET DESCRIPTION:

The dataset is also distributed as a JSON formatted text file shipsnet.json. The loaded object
contains data, label, scene_ ids, and location lists.

Data: It contains image data in the form of array

labels: Valued 1 or 0, representing the “ship” class and “no-ship” class, respectively.

Scene id: The unique identifier of the PlanetScope visual scene the image chip was extracted
from. The scene id can be used with the Planet API to discover and download the entire scene.

Longitude_latitude: The longitude and latitude coordinates of the image centre point, with
values separated by a single underscore.

The pixel value data for each 80x80 RGB image is stored as a list of 19200 integers within
the data list. The first 6400 entries contain the red channel values, the next 6400 the green, and
the final 6400 the blue. The image is stored in row-major order, so that the first 80 entries of
the array are the red channel values of the first row of the image.

The “ship” class includes 1000 images. Images in this class are near-centred on the body of a
single ship. Ships of different sizes, orientations, and atmospheric collection conditions are
included.

Fig 4.3.1: Sample ship images

21

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

The "no-ship" class includes 3000 images. A third of these are a random sampling of different
land cover features - water, vegetation, bare earth, buildings, etc. - that do not include any
portion of an ship. The next third are "partial ships" that contain only a portion of an ship, but
not enough to meet the full definition of the "ship" class. The last third are images that have
previously been mislabelled by machine learning models, typically caused by bright pixels or
strong linear features.

Fig 4.3.2: Sample no-ship images

4.4. OBJECTIVE OF THE CASE STUDY:

Objective of the problem is to help address the difficult task of identifying the presence of large
ships in satellite images. Automating this process can be applied to many issues including
monitoring port activity levels and supply chain analysis. Goal is to identify ships in satellite
images. To classify whether ships are present or not present.

22

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

CHAPTER 5: DATA PREPROCESSING/FEATURE

ENGINEERING AND EDA

5.1 LOADING THE DATA:

Python has a built-in open() function to open a file. This function returns a file object, also
called a handle, as it used to read or modify the file accordingly.

We can specify the mode while opening a file. In mode, we specify whether we want to read,
write or append to the file. We can also specify if we want to open the file in text mode or
binary mode.

The default is reading in text mode. In this mode, we get strings when reading from the file.
Python has built-in package called json, while can be used to work JSON data. Contents of json
file are stored in dataset variable

Fig 5.1.1: Reading data from json file

Pandas DataFrame to which all the operations can be performed which helps us to access each
and every row as well as columns and each and every value can be access using the dataframe.
Any missing value or NaN value have to be cleaned.

23

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

Fig 5.1.2: Loading data set

5.2 STATISTICAL ANALYSIS:

Descriptive statistics include those that summarize the central tendency, dispersion and
shape of a dataset’s distribution, excluding nan values. Analyses both numeric and object
series, as well as DataFrame column sets of mixed data types. The output will vary depending
on what is provided.

For numeric data, the result’s index will include count, mean, std, min, max as well as
lower, 50 and upper percentiles. By default the lower percentile is 25 and the upper percentile
is 75.The 50 percentile is the same as the median.

For object data (e.g. strings or timestamps), the result’s index will include count,
unique, top and freq. The top is the most common value. The freq is the most common value’s
frequency. Timestamps also include the first and last items.

If multiple object values have the highest count, then the count and top results will be
arbitrarily chosen from among those with the highest count.

For mixed data types provided via a DataFrame , the default is to return only an analysis
of numeric columns. If the dataframe consists only of object and categorical data without any
numeric columns, the default is to return an analysis of both the object and categorical columns.
If include='all' is provided as an option, the result will include a union of attributes of each
type.

24

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

Fig 5.2.1: Statistical data of labels

Fig 5.2.2: Checking for total number of ship and non ship images

Observations:-

1. There are total 4000 images.

2. By observing the mean of the labels there are 1000 ship containing images and 3000 no
ship images.

5.3 VISUALISATION OF IMAGES

To know the image names from the main folder these following steps are done accordingly.

1. Checking for the contents in main data containing folder

25

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

Fig 5.3.1: Contents in file

2. Here base directory is created so that it makes joining the other paths easy.

Here ship_imgs contains all the images

Fig 5.3.2: Creating directories

3. Displaying 5 images names

Fig 5.3.3: Displaying 5 images names

26

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

4. By, using one of the image name and displaying a sample image using matplotlib package.

Fig 5.3.4: Displaying Sample image

5. Visualization of 15 images with title as the image shapes

Fig 5.3.5: Input for visualization of 15 images

27

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

Fig 5.3.6: Output for visualization of 15 images

Observation: Images are of size 80 x 80 x 3.

5.4 HANDLING MISSING VALUES:

There are a number of schemes that have been developed to indicate the presence of
missing data in a table or DataFrame. Generally, they revolve around one of two strategies:
using a mask that globally indicates missing values, or choosing a sentinel value that indicates
a missing entry.

Fig 5.4.1: No Missing Values in the data set

28

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

CHAPTER 6: FEATURE SELECTION

6.1 SELECT RELEVANT FEATURES FOR THE ANALYSIS:

Feature Selection is the process where you automatically or manually select those
features which contribute most to your prediction variable or output in which you are interested
in. Having irrelevant features in your data can decrease the accuracy of the models and make
your model learn based on irrelevant features. In Classification of images input should be of
numpy array or image.

Fig 5.1.2: Loading data set

Here data column can be considered as input and labels column can be considered as output.
Remaining columns are irrelevant.

6.2 DROP IRRELEVANT FEATURES:

In the dataset locations, scene_ids columns are irrelevant so these columns are to be
dropped.

29

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

Fig 6.2.1: Dropping columns

6.3 CONVERTING DATA AND RESHAPING:

In CNN image is defined in the form of array based on that it will classify the images.
So, as the project data is not directly involving the images so data should be in the array format.
As the data is in not array format it should be converted to array format of building the model.

Fig 6.3.1: Conversion of data into numpy array

Here, the data column which is in the numpy array format is named as items and labels column
which is in the numpy array format is named as output_labels.

30

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

When doing image classification, it is important to make sure you are using the correct size of
images; otherwise you may get unexpected results or errors.

Fig 6.3.2: Checking shape of data

Here, the shape of the items data is not in correct size format so reshaping should be done for
correct format to proceed further.

Fig 6.3.3: Reshaping of data

Now, data is of 80 x 80 x 3 size.

6.4 FEATURE SCALING:

It is a step of Data Pre Processing which is applied to independent variables or features

of data. It basically helps to normalise the data within a particular range. Sometimes, it also
helps in speeding up the calculations in an algorithm. Real world dataset contains features that
highly vary in magnitudes, units, and range. Normalisation should be performed when the scale
of a feature is irrelevant or misleading and not should Normalise when the scale is meaningful.

Fig 6.4.1: Scaling of data

31

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

Scaling is done only for items column where image data is represented in numpy
array format. The items column is divided by 255 because every digital image is formed by
pixel having value in range 0~255. 0 is black and 255 is white. For colourful image, it
contains three maps: Red, Green and Blue, and all the pixel still in the range 0~255. Since
255 is the maximin pixel value. Rescale 1./255 is to transform every pixel value from range
[0,255] -> [0,1].which helps treating all images in the same manner.

32

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

CHAPTER 7: MODEL BUILDING AND EVALUATION

7.1 BRIEF ABOUT THE ALGORITHMS USED AND BUILDING

MODEL

Convolutional Neural Networks:

A breakthrough in building models for image classification came with the discovery
that a convolutional neural network (CNN) could be used to progressively extract higher- and
higher-level representations of the image content. Instead of pre-processing the data to derive
features like textures and shapes, a CNN takes just the image's raw pixel data as input and
"learns" how to extract these features, and ultimately infer what object they constitute. CNN
receives an input feature map: a three-dimensional matrix where the size of the first two
dimensions corresponds to the length and width of the images in pixels. The size of the third
dimension is 3 (corresponding to the 3 channels of a colour image: red, green, and blue). The
CNN comprises a stack of modules, each of which performs three operations.

1. Convolution:

A convolution extracts tiles of the input feature map, and applies filters to them to
compute new features, producing an output feature map, or convolved feature (which may have
a different size and depth than the input feature map).Convolutions are defined by two
parameters:

 Size of the tiles that are extracted (typically 3x3 or 5x5 pixels).
 The depth of the output feature map, which corresponds to the number of filters that are
applied.

During a convolution, the filters (matrices the same size as the tile size) effectively slide
over the input feature map's grid horizontally and vertically, one pixel at a time, extracting each
corresponding tile.

33

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

Fig 7.1.1: Illustration of convolution over feature map

In fig 7.1.1 a 3x3 convolution of depth 1 performed over a 5x5 input feature map, also
of depth 1. There are nine possible 3x3 locations to extract tiles from the 5x5 feature map, so
this convolution produces a 3x3 output feature map.
In Figure 7.1.1, the output feature map (3x3) is smaller than the input feature map (5x5).
If you instead want the output feature map to have the same dimensions as the input feature
map, 2you can add padding (blank rows/columns with all-zero values) to each side of the input
feature map, producing a 7x7 matrix with 5x5 possible locations to extract a 3x3 tile.
For each filter-tile pair, the CNN performs element-wise multiplication of the filter
matrix and the tile matrix, and then sums all the elements of the resulting matrix to get a single
value. Each of these resulting values for every filter-tile pair is then output in the convolved
feature matrix (see Figures 7.1.2 and 7.1.3).

Fig 7.1.2: A 5x5 input feature map and 3x3 convolution of deph1

34

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

Fig 7.1.3: Example of 3 x 3 convolution performed over 5 x 5 input feature map

In Figure:

Left: The 3x3 convolution is performed on the 5x5 input feature map.

Right: the resulting convolved feature.

For each filter-tile pair, the CNN performs element-wise multiplication of the filter
matrix and the tile matrix, and then sums all the elements of the resulting matrix to get a single
value. Each of these resulting values for every filter-tile pair is then output in the convolved
feature matrix.

During training, the CNN "learns" the optimal values for the filter matrices that enable
it to extract meaningful features (textures, edges, shapes) from the input feature map. As the
number of filters (output feature map depth) applied to the input increases, so does the number
of features the CNN can extract. However, the trade-off is that filters compose the majority of
resources expended by the CNN, so training time also increases as more filters are added.
Additionally, each filter added to the network provides less incremental value than the previous
one, so engineers aim to construct networks that use the minimum number of filters needed to
extract the features necessary for accurate image classification.

2. ReLU

Following each convolution operation, the CNN applies a Rectified Linear Unit (ReLU)
transformation to the convolved feature, in order to introduce nonlinearity into the model. The
ReLU function.

35

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

F(x)=max(0,x) returns x for all values of x > 0, and returns 0 for all values of x ≤ 0.

3. Pooling

After ReLU comes a pooling step, in which the CNN down samples the convolved feature (to
save on processing time), reducing the number of dimensions of the feature map, while still
preserving the most critical feature information. A common algorithm used for this process is
called max pooling.

Max pooling operates in a similar fashion to convolution. We slide over the feature map and
extract tiles of a specified size. For each tile, the maximum value is output to a new feature
map, and all other values are discarded. Max pooling operations take two parameters:

Size of the max-pooling filter (typically 2x2 pixels)

Stride: the distance, in pixels, separating each extracted tile. Unlike with convolution, where
filters slide over the feature map pixel by pixel, in max pooling, the stride determines the
locations where each tile is extracted. For a 2x2 filter, a stride of 2 specifies that the max
pooling operation will extract all non-overlapping 2x2 tiles from the feature map (see Figure
7.1.4).

Fig 7.1.4: Example of Max Pooling

36

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

In Figure 7.1.4.

Left: Max pooling performed over a 4x4 feature map with a 2x2 filter and stride of 2.

Right: the output of the max pooling operation. Note the resulting feature map is now 2x2,
preserving only the maximum values from each tile.

4. Fully Connected Layers

At the end of a convolutional neural network are one or more fully connected layers
(when two layers are "fully connected," every node in the first layer is connected to every node
in the second layer). Their job is to perform classification based on the features extracted by
the convolutions. Typically, the final fully connected layer contains a softmax activation
function, which outputs a probability value from 0 to 1 for each of the classification labels the
model is trying to predict.

Preventing Overfitting

As with any machine learning model, a key concern when training a convolutional neural
network is overfitting: a model so tuned to the specifics of the training data that it is unable to
generalize to new examples.

Two techniques to prevent overfitting when building a CNN are:

Data augmentation: artificially boosting the diversity and number of training examples by
performing random transformations to existing images to create a set of new variants. Data
augmentation is especially useful when the original training data set is relatively small.

Dropout regularization: Randomly removing units from the neural network during a training
gradient step.

Fig 7.1.5: Importing model and other parameters

 CNN Parameters are implemented in sequential model. Here I have added 4

convolution layers followed by max pooling layers.

37

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

 As the model is binary classification the output dense value is 1 and using sigmoid as
activation function for the last dense layer. Sigmoid activation function gives output for
binary classification in range 0 to 1 whereas for the value less than 0.5 its classifies to
0 and for the value greater than 0.5 its classifies to 1.

 Flatten is used to convert data into 1D array.

 Same padding: It applies padding to the input image so that input image gets fully
covered by the filter and specified stride. It is same because, for stride 1,the output is
same as the input.

Fig 7.1.6: Building model

38

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

Fig 7.1.7: Output of Built model

Here, total 2,700,009 parameters to trained in model. After building compiling should be
done.

Compiling the Model:

Here we need to define how to calculate the loss or error. Since we are using a binary
classification, we can use binary_crossentropy. With the optimizer parameter, we pass how to
adjust the weights in the network such that the loss gets reduced. There are many options that
can be used and here I used Adam Optimizer method. Finally, the metrics parameter will be
used to estimate how good our model is here we use the accuracy.

Adam optimizer:

Adam optimization is a stochastic gradient descent method that is based on adaptive estimation
of first-order and second-order moments.

39

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

Binary crossentropy:

Binary crossentropy is a loss function that is used in binary classification tasks. These are tasks
that answer a question with only two choices (yes or no, A or B, 0 or 1, left or right).

Accuracy metric:

This metric creates two local variables, total and count that are used to compute the frequency
with which y_pred matches y_true. This frequency is ultimately returned as binary accuracy:
an idempotent operation that simply divides total by count.

If sample_weight is None, weights default to 1. Use sample_weight of 0 to mask values.

Arguments

 name: (Optional) string name of the metric instance.

 dtype: (Optional) data type of the metric result.

Fig 7.1.8: Compiling the model

After compiling the model, it should be trained.

7.2 TRAIN THE MODELS

Splitting the data : after the pre-processing is done then the data is split into train and Validation
sets

 In deep learning in order to access the performance of the classifier. You train the
classifier using 'training set' and then test the performance of your classifier on unseen
'validation set'. An important point to note is that during training the classifier only uses
the training set. The validation set must not be used during training the classifier. The
validation set will only be available during testing the classifier.
 Training set - a subset to train a model.(Model learns patterns between Input and
Output)

40

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

 Validation set - a subset to test the trained model.(To test whether the model has
correctly learnt )
 The amount or percentage of Splitting can be taken as specified.
 First we need to identify the input and output variables and we need to separate the input
set and output set.
 An epoch is a term used in machine learning and indicates the number of passes of the
entire training dataset the machine learning algorithm has completed. Datasets are
usually grouped into batches (especially when the amount of data is very large). Some
people use the term iteration loosely and refer to putting one batch through the model
as an iteration.. This helps in attaining good accuracy to the model.
 Here, the model has been split into 80% for trained data and rest 20% of the data is of
validation data by using valididation_split parameter.

Fig 7.2.1: Training the model

41

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

Fig 7.2.2: Output of training the model

7.3 VALIDATE THE MODELS:

Model validation is the process of evaluating a trained model on test data set. This provides the
generalization ability of a trained model. Here I provide a step by step approach to complete
first iteration of model validation in minutes.

 The model are validate after completion of training and testing the model.
 Checking the accuracy scores as metrics to validate the models.

42

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

Fig 7.3.1: Train accuracy vs Validation accuracy

43

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

Fig 7.3.2: Train loss vs Validation loss

1. Here, training accuracy is 100%.

2. Validation accuracy is 94.75%

By observing from the above graph test accuracy and validation accuracy doesn't have much
variation.

So, the model doesn't over fit or under fit.

44

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

7.4 MAKE PREDICTIONS:

 We have a method called predict , using this method we need to predict the output for
given input and we need to compare that the model is predicting correctly or not. If the
model is predicting correctly then the model is good otherwise some changes should be
done.
 Now, I am trying to predict an unknown image which is not present in the training or
validation set.

Below image is an unknown image with a different size. Now it should predict the image is
consisting of ship or not.

Fig 7.4.1: Unknown sample of ship image

In order to predict the new image we need to resize and scale the new image. So, that image is
classified without any errors.

45

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

Fig 7.4.2: Resizing and scaling unknown sample image

 Firstly, we import image from TensorFlow, then we load the image and paste the URL
of image we are predicting.
 After loading we convert the image to array format and then check the shape and type
of image. Now we will be using this during resizing the image we are predicting on to
convert it into required shape.
 Next, we will be applying scaling on the image and to get the dimensions right, expand
dims enters the process, finally we check the shape and observe that we got the desired
result.

46

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

 Predicting the scaled image

Fig 7.4.3: Predicting the ship image

 The output array([1.]], datatype=float32) represents ship is present in the image.

 Similarly, Predicting for no ship image from unknown satellite image where image
does not consists of ship.

Fig 7.4.4: Unknown sample of no ship image

47

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

Fig 7.4.5: Predicting the no ship image

 The output array([[1.7355088e-10]], dtype=float32)this values is less than 0.5 and it is

almost equal to 0.
 From this it can be confirmed that the built model classifies images correctly.

48

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

CONCLUSION

It is concluded after performing thorough building model using CNN gives 94.75%
accuracy. The model is computed to get good accuracy for unknown data that detects ship
images and non-ship images correctly which leads to clear understanding of the how the model
parameters should be defined to get best accuracy and not leading to under fit or over fit. Thus,
it leads to point of getting the solution for the problem statement of identification of ships
through satellite images.

49

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)

 lOMoARcPSD|42949913

REFERENCES

● https://en.wikipedia.org/wiki/Deep_learning

● https://en.wikipedia.org/wiki/Machine_learning

● https://www.kaggle.com/rhammell/ships-in-satellite-imagery

● https://www.tensorflow.org/tutorials/images/cnn

● https://towardsdatascience.com/building-a-convolutional-neural-network-cnn-in-keras

● GITHUB LINK: https://github.com/sahithreddy567/Project-AIML

50

Downloaded by Pratham Sharma (prathamsharmatrigger2002@gmail.com)