Remote Sensing

Creation of a land use map by

supervised classification of a Landsat image
In this lab, you will create a land use map using supervised classification.
a Landsat satellite image representing the commune Maarif– Casablanca and the implementation
on the page of the obtained map.

The major steps to follow for the implementation of a supervised classification are:
1. Observation of the image
1.1. Identification of land use classes
Visually analyze your image and identify the main types of land use:
Urban centers (city and village, buildings, train station, ...)
Routes (motorway, route, rail)
Wooded areas (forest, woods, ...)
Agricultural land (cultivated field and fallow field)
Watercourses (rivers, canals,...)
Decide on the classes you will try to identify through automatic image classification.
(between 5 and 8 classes).

1.2 Use of complementary data for the visual analysis of an image

Photo-interpretation (recognition of different land uses through analysis)
Visual interpretation of an image is sometimes difficult:
Either because the image does not present sufficient inequality (spatial and spectral resolution)
adequate,
Either due to a lack of experience of the photo interpreter or a lack of
Knowledge of the area.
Photo interpretation can be facilitated by using complementary data.
studied area, for example:
Visual analysis of the area in Google Earth. Google Earth has very high-resolution images.
high spatial resolution for an ever-growing part of the planet.
However, be careful to keep in mind that the images from Google Earth are dated.
probably not from the same period as the one that is the subject of your study.
Topographic maps
A field survey with GPS data collection
Various documents providing information on land use in the area
1.3 Contrast enhancement
an improvement in contrast facilitates the visual interpretation of the image
Contrast enhancement involves a virtual transformation of the data.
in order to use the entire range of color intensity available (256 levels)
to visualize the data on the screen. This improvement does not change your
source data (the value of pixels). It only assigns differently the
color palette available for your image to highlight certain aspects
elements more clearly.
2. Supervised image classification
Performing supervised classification of a satellite image implies that you
have a very good knowledge of the actual land use of the area
from the image to the time when the image was taken. This knowledge can come from
several sources. This knowledge can come from several sources such as
explained above "Use of complementary data for analysis
visual of an image.
2.1 Definition of the map legend (taxon)
From the examination of the colored composition of the area to be classified (also
with the improvement of contrast), and your knowledge of the studied region,
define the legend of the map, that is, the land use classes that
you want to see appear in the land use map that will be produced.
2.2 Definition of the number of spectral classes
Based on the examination of the colored composition of the area to be classified (also with
the improvement of contrast), also identify how many spectral signatures
different corresponds to each taxon of the legend. Indeed, a land occupation
such as 'culture' can present different spectral signatures at the time of
the recording of the image, according to the cultivated species and its vegetative stage. It is therefore about
define as many spectral classes as there are spectrally well-different situations
for each type of land use. The classification algorithm will thus be able to process each
independent spectral signature.
2.3 Selection of training areas or 'ROI - Region Of Interest' for
each class
It involves selecting a minimum of 30 spectrally representative pixels of each signature.
spectra previously identified. These samples, or 'training areas' will serve for the '
CALIBRATION" of your classification. They are defined on the screen by the scanning of one or
several polygons for each spectral class. Based on the polygons thus defined, the extraction
automatic extraction of pixel values contained within the polygons and the calculation of their
mean and standard deviation will produce the spectral signature of each class from the 3 channels of
the image.
Training samples must be representative of each spectral class.
considered. Here are some tips for selecting training areas:
It is often preferable for these training areas to be well distributed throughout
the image. Indeed, this will take into account:
The possible gradients of brightness between the different areas of the image.
(atmospheric conditions, relief/exposure,…)
A slight spectral difference between the same land use class in
different regions (gradient or soil classes, variation of plant species or
of their state of development, etc.).
o...
A minimum of 30 pixels per spectral class and a sufficient number of areas.
training samples so that they are representative of the class considered. The number
the necessary training areas will therefore depend on the spectral complexity of each
class and the complexity of the image. For example, an image that only includes as
Land occupation that the desert and the sea will be extremely easy to classify and this will be able to
Make do with little training. In contrast, an image of an agricultural area in Africa.
will present a much greater complexity of land use (plots with
different cultures at different stages of development, different states of soil and
different soils, trees in the plots, etc.) and will require the use of many
training areas so that they are well representative of all the variations of a
given spectral class.
Distribution curve of the number of pixels based on their reflectance (for a band
spectrale donnée) doit êtreuni-modale(ou s’en rapprocher). Si elle est multimodale, ceci constitue une
indication that these areas correspond to several spectrally distinguishable classes and that they should therefore be
separate into as many classes as there are modes in the curve.
Only consider 'pure pixels', that is to say pixels corresponding exclusively to the class
land use aimed at and not pixels covering several land use classes ("mixed": pixel
mixed). For example, for training areas of a river, do not consider pixels that are located
close to the bank because their reflectance may be influenced by the bank.
Be careful to include the term 'cal' in the name of these classes that will be used for
calibration ends (for example: 'forest_cal'), in order to easily distinguish between your classes
de calibration

3. Validation of classification results

4. Production of the land use map derived from classification
 Supervised classification
(With the help of QGIS tools)
* Qgis Opening ----New project titled 'Supervised Classification'----coordinate system
projected EPSG:32629
Added data (ImageLandsat "[Link]" in the folder "Supervised Classification")

8
Addition of the shapefile "Maarif" in the same folder
We cut the raster image "[Link]" using the shapefile "Maarif" as a mask.
The obtained raster will be named 'MaarifSat'
Property of the obtained raster layer 'MaarifSat'----section 'style' then make the changes as
suit

10
 Menu «SCP»------«Semi-Automatic Classification Plugin»
Then refresh

Then in 'Input image' select: 'MaarifSat'.

1
4 3 2

For image contrast enhancement ---Show

 Activate the automatic projection on the fly
1

Add the topographic map of Casablanca georeferenced in the 'TP1 remote sensing'
The superimposition of topographic maps with the satellite image allows us better recognition.
of the different occupations of the soil
Creation of Shapefile of training areas (ROI)

2 3
The created Shapefile appears in the list of layers in Qgis.
it is an empty layer but contains field names of
Shapefile "ROI"
2

1
 Creation of spectral signature file

The spectral signature represents the spectral variations, that is to say the
variations in the reflectance of elements as a function of wavelength
Recording taken in electromagnetic spectrum usable areas
in remote sensing.

2 3
 Start the scanning of the shapefile "ROI"

4
5

The record "Bati" appears in "Training shapefile" of the shapefile "ROI".

Choose the following classes:
We merge the two areas that represent the same class 'Vegetation':
We obtain the merged class 'MergedVegetation' and then we delete the
two old classes
 Change the settings as follows, then press '+' and then on the image

We obtain a similar image.

to this one:
Then, we begin the classification:
We obtain a Raster image and a
vector layer

The shapefile contains several entities of

the same class
To calculate the areas by class, let's group the entities belonging to
the same category. All entities having identical values for this field are
combined to form a single entity.

22
Remove the entity "0" from the field "C_ID"
Then add a field containing the names of the classes:

3 4
1
Then we enter the names of the different classes as indicated below:

To calculate the area of each class, we need to create a field.

"Surface" in the attribute table of the layer "QgisClassification"
Use the following symbology:
Cartographic edition
The map is a particularly effective medium for communicating the results of
your research. It indeed allows for synthesis and provides an overview
on a sometimes considerable amount of information at a glance. It is very
It is important to ensure that it is as readable, attractive, and 'talking' as possible.
so that the person reading your card can easily understand what you
have decided to represent there.

Essential elements on a map

Every map must include the following elements:
A title
A graphic scale
A legend
An arrow indicating the North
The information related to the coordinate system used
The author (person and/or 'Company')
The creation date of the map
You can add a situation map

Represent your results on a map that you create in QGIS (see TPQgis)
"tuto10". Layout: the print composer (page:12).
Then export your layout in 'Jpeg' format according to the model below:
 Supervised classification
(With the help of Erdas Imagine tools)
 Addition of data (Landsat Image
"[Link]" in the folder
Supervised classification

To bring the image "[Link]"

to be visible on the screen:
Right-click on the layer, then 'fit layer to
window
Addition of the shapefile of the municipality 'Maarif' in the same folder
Selection of the Maarif shapefile (1 then 2)
Click on 'Paste from selected object' (3)
Then click on the newly created layer "AOI" (4)
1

2
We cut the raster image '[Link]' according to the created layer.
represent the boundaries of the municipality 'Maarif'. The obtained raster will be named
"MaarifSat"
1
1 2

2 3

6 3
In order to perform supervised classification using Erdas tools, one must first
create a signature file from the training samples
Start the scanning of samples:

1
2

3
4
Choose the following classes:

To group the two areas that represent the same class 'Vegetation'
Use the following menu:
Save the signature file under the name "Signature":
We can now start the classification: the classified raster will be
named "maarifclass"
We obtain a layer that looks like the image below:
 To change the colors of the classes: right-click on the
layer the classes and do as indicated below:

Use the following palette: