Professional Documents
Culture Documents
To be filled by Student
Answer the following questions (Plagiarism will be checked, please avoid copying and
pasting).
Q.No.1. Download Landsat 8 and Sentinel 2 image of your hometown and subset it to the size of
10 by 10 km. Perform following tasks on it (you are supposed to take screenshot of your every
single step like a tutorial (if two students live in the same home, then one of them should change
study area by taking their grandparents’ home as study area in order to keep results different from
each other):
1) Annotate important places like your own house, hospital, school, playgrounds, croplands (for
villages)/parks/gardens (for cities).
2) Apply change detection on both satellite images from 2017 to 2020 and discuss what changes
happened there (quantitatively, means calculate change in area of places) and compare any
differences in Sentinel 2 and Landsat 8 results (if any). (Marks 04.)
Answer:
Annotate important places like your own house, hospital, school, playgrounds, croplands (for
villages)/parks/gardens (for cities).
Annotation map :
I have applied annotation of my home Tehsil villages.
Part II
2) Apply change detection on both satellite images from 2017 to 2020 and discuss what changes
happened there (quantitatively, means calculate change in area of places) and compare any
differences in Sentinel 2 and Landsat 8 results (if any).
Change Detection Results:
I have applied change detection.
Analysis on First Image 2010:
I classify image in 5 classes,
Water
Barren land
Urban land
Tree
Agriculture greenry
Attribute Table :
Analysis on 2nd image:
Attribute Table:
Q.No.2. a) Make a list of high resolution satellites and low resolution satellites who are in orbits
and provide following information in tabular form:
Spatial resolution
Spectral resolution,
launching year
Country who launched
Area of research it can be used for.
Answer:
year n launched
Cartosat-3 2019 MX 4 1.14 India Cartosat-3 is an advanced satellite, its
Cartosat-3 2019 PAN 1 0.28 India
imagery used in earth observation
Cartosat-2F 2018 HRMX 4 2.0 India
Cartosat-2F 2018 PAN 1 0.65 India purposes, surverys s, uban planning,
military and weather mapping.
b) Make a list of all satellites in geo-stationary/ non-synchronous orbits and provide following
information in tabular form:
Spatial resolution
Spectral resolution,
launching year
Country who launched
Area of research it can be used for.
(Marks 04)
Answer:
Geo-Stationary Satellites
Name Spectral Spatial Launching Country Field of use
(m) launched
HYLAS-1 0.23-12.5 30 2010 USA Used for the coast line observation
MICROMETER
AMC-4 100nm – 100 30 1999 Dutch It is used for weather monitoring and
micrometer weather mapping.
HIMAWARI- 8 bands 644 500-2000 2015 Japan It is used for aviation and surveying.
8 nm to2260
nm
SYNCOM-2 12 bands 500 1963 USA It is used for ocean study and also used
o.51 to 0.91 for weather monitoring
micrometer
GALAXY-25 28 kubands 25 1997 USA It is used for communication.
GOES-16 16 bands 1000 2016 USA Used for weather mapping
0.64 meter
METEOSAT- 12 bands 3000 2008 ESA It is used for weather monitoring and
8 disaster control
Echostar 12 band 500 2017 USA Weather monitoring and communication
Astra-1 14 band 2.5 1998 UK Telecommunication purposess
Not car
Q.No.3. Discuss all existing resources available in doing research on image classification by using
Machine Learning and Deep Learning practices in the context of remote sensing. Especially focus
what Google Company provides in this domain in comparison with its competitors like Facebook.
(Marks 04)
Answer:
Machine Learning:
Machine learning is the study that gives ability to the computers or machines to learn without being
explicitly programmed”. A Machine Learning project run a piece of software that produced output
from B from given input A.
⛰⛪
input feature extraction classification output
Deep Learning:
It is the subset fo ML and type of artificial intelligence (AI), Deep learning makes the network
capable to learn unsupervised from the data which is unlabeled or unstructured. It works on the
principle of Deep Neural Networks.
Image classification is done by using supervised learning method and in which we define a set of
target objects to identify the image of interest and after that train the model to recognize the
objects using labeled photos. It involves extraction of useful information and features from the
image.
Tools and libraries:
Following are the tools and programing libraries used for the image classification in remote sensing:
model.
Keras: It is library used for the implementation of Neural Network architecture. Keras
Numpy: it used for mathematical computation and algorithem for operations of data
structures.
KNN is used for identification of objects based on the closest training model in the feature
space. .
SSVM is used for non-linear classification and it is done by using kernel trick.
In MLP there can be more than one non-linear layer can be hidden between input and output.
CNNs have is used to manage the to achieve superhuman performance by increasing the
Image Classification
It means to label the objects and features in the image or photograph
Object Detection
Object detection is the most challenging and important part of the image classification, it
involves recognition of object.
Object Segmentation
It involves to segmentation of objects in an image on the basis of pixels, also involves
segmentation of pixel in which a line draw around the object.
Image Colorization
It involves to giving color to the simple grey style image.
Image Reconstruction
This part involves filling of damage or missing part in the image.
Image Super-Resolution
In this task, high resolution given to an image.
Image Synthesis
It is the task in which model is train on the basis of object style. For example the physical
orientation of zebra and horses are same, but the color pattern is different, so this is
classified using this model.
Deep learning and machine learning and widely used in remote sensing for image
classification, object recognition, feature extraction from the image.
Google provides built in algorithms of image classification of machine learning and deep
learning. User from all over the world can write and implement code on Google Cloud.
Google Cloud
Google provides Google cloud at there developer can write and run code and built in
function. No need of any other software, so that’s why it is very popular in public.
Deployment of Trained model:
Cloud AutoML is a suite provide by the google for image classification and object
recognition.
ML Kit
This kit is capable of detecting face, text and identifying landmarks and other features from
the object.
It can be run in API or cloud
Provides easy ways to deployment of the model.
Security
Google AI hub provides platform where user can share their model and project from end to
end pile line of ML.
TenserFlow
Google provides research products about ML in that. It also provides Jupyter notebook in
which we have a large package of all ML tools.
Facebook developers has made the algorithms that ranks ads and feeds. Facebook has strong object
recognition and image recognition algorithms i.e if any user of two billions user of facebook upload
picture of another person, Facebook notify the person about this by detecting face.
PyTorch:
PyTorch is created by facebook to facilitate their users and developers. It is easy to use and
debug as compared to TenserFlow
ONNX
It is open source frame work by facebook to facilitate the developers to make model and
deployment of that model.
Detectron
Google is providing more services in machine learning image processing and classification as
compared to its competitors like facebook and Microsoft.