E3S Web of Conferences 364, 04001 (2023) https://doi.org/10.

1051/e3sconf/202336404001
JOE3

CLASSIFICATION OF A QUICKBIRD SATELLITE IMAGE BY

MACHINE LEARNING TECHNIQUES: MAPPING AN URBAN
ENVIRONEMENT BY DECISION TREE METHOD.

Ameslek Oumaima1*, Hanan Latifi 2, Lamyae Haltou 2, Hafida Zahir2, Mostafa El Bachaoui1
1Remote Sensing and Geographic Information Systems Applied to Geosciences and Environment Laboratory, Faculty of Science and
Technology, Beni Mellal, Morocco
2 Bio-processes and Bio-interface Laboratory, Faculty of Science and Technology, Beni Mellal, Morocco

Abstract. Classification is a crucial stage in the processing of satellite images that influence
considerably the quality of the result. A variety of methods is proposed in the literature for the purposes
of image classification. They present many differences in their basic principles, thus in the quality of
the results obtained. Therefore, a study of different classification methods seems to be essential. The
classification of satellite images with conventional methods can be done in several ways using different
algorithms. These algorithms can be divided into two main categories: supervised and non-supervised.
Decision tree on the contrary is a machine learning tool. It is a plain model characterized by the
simplicity of understanding and interpretation. This work aims firstly, to classify a high resolution
Quickbird satellite image of an urban area by the decision tree method and compare it with the
conventional classification algorithms in order to evaluate its efficiency. The methodology consists of
two main stages: classification and evaluation of results. The second is based on the calculation of a
number of statistical indices derived from the confusion matrix: the statistical parameter "kappa" and
the overall coefficient of precision.

cover mapping. In literature, classification methods

range from unsupervised algorithms (i.e., ISODATA or
1 Introduction K-means) [11] to supervised algorithms (i.e., maximum
For proficient and sustainable management of urban likelihood) and machine learning algorithms such as k-
areas there is an urgent need for effective and Nearest Neighbors (KNN), decision trees (DT), support
successful monitoring of physical changes over time vector machines (SVM), and random forest (RF) [12].
[1,2,3], Satellite images can be decisive in helping
manage cities and infrastructure growth [4,5]. For the decision tree, it consists of a hierarchical series
They are considered as one of the most important data of decisions to be made in order to determine the correct
sources for land mapping due to their extensive class.
geographical coverage at an efficient cost while The decision tree consists of a number of decision
providing irreplaceable information on the earth’s nodes. Each node makes an assignment to a class or
surface [6,7,8]. group of classes [13].
However, the accuracy of the produced maps is The advantage of decision trees is that it is possible to
considerably affected by the accuracy and choice of the integrate different data sources and different types of
classification method [9]. attributes at each level of decision [12].
The choice of relevant decisions at each node is very
Choosing the right classification method is not an easy important to obtain an accurate classification.
task. The panoply of methods and algorithms used to Otherwise, there may be an accumulation of errors and
classify satellite images certainly leaves a wide choice difficulty in discriminating between subcategories of the
to the analyst, but complicates his task insofar as the same class [14].
basis changes completely from one method to another. In this paper, we will be using the decision tree method
According to Lu and Weng [10], it is not only the for land cover classification in an urban area presented
imagery appropriateness but also the right choice of by a high resolution Quickbird satellite image.
classification method that affects the results of land

* Corresponding author: o.ameslek@usms.ma

© The Authors, published by EDP Sciences. This is an open access article distributed under the terms of the Creative Commons Attribution License 4.0
(http://creativecommons.org/licenses/by/4.0/).
E3S Web of Conferences 364, 04001 (2023) https://doi.org/10.1051/e3sconf/202336404001
JOE3

The main purpose is to implement this machine learning 2.1 Study area and satellite image.
algorithm and evaluate its performance in a relatively
complex environment (urban zone) in order to optimize Our study area is Rabat city (figure 1), the
its usage and help researchers better choose between administrative capital of Morocco. It is located on the
different image classifiers. Atlantic coast at 33 ° 1', 31" North, and 6 ° 53' 10" Ouest.

2 Materiel et Methods.

Fig. 1. Study area.

The choice of this area is due on the one hand to the In this study, we used a Quickbird image (figure 2),
availability of images and on the other hand to the previously corrected, in a merged panchromatic and
diversity of the existing details: buildings of different multispectral mode, dating from the year 2007.
configurations, a dense road network and a vegetation
cover.

Fig. 2. The Quickbird image used.

2
E3S Web of Conferences 364, 04001 (2023) https://doi.org/10.1051/e3sconf/202336404001
JOE3

This image has four spectral bands for which Table1

presents the corresponding wavelengths.

In panchromatic mode, the spatial resolution is 0.6 m, in

multispectral mode, the spatial resolution is 2.4 m.

Table 1. Spectral bands of the Quickbird image used.

Band Wavelenght

Blue 0 .446 - 0.500 μm

Green 0 .500 - 0.578 μm

Red 0 .620 - 0.700 μm

Near infrared 0 .800 - 1.500 μm

2.2 Software.

To accomplish this study, we will use the ENVI

software (Environment for Visualizing Images) which is
a professional software from the company "EXELIS"
for processing remote sensing, optical and radar images.
All image processing methods for geometric and
radiometric corrections, classification and cartographic
layout are present. Other tools related to the
visualization and modelling of topographic data are also
available.

The ENVI software is designed in IDL (Interactive Data

Language) and therefore offers advanced programming Fig. 3. Steps in constructing the decision tree.
resources. This software is much more specialized for
multi-spectral images than for the more cartographic 2.3.1 Identification of thematic classes
aspects of other GIS software. It allows easy integration
of raster and vector data. It offers some classification The determination of these classes requires a very good
algorithms. On the other hand, it has the ability to easily knowledge of the actual land use of the image area. Land
add modules and external programs in various ways can cover exhibits different spectral signatures at the time of
use its algorithms. image registration. It is therefore a question of defining
as many spectral classes as there are very different
2.3 The decision tree workflow. situations for each land use.
The classification algorithm will thus be able to process
The decision tree workflow takes place according to the each spectral signature independently. In this study, six
six following steps (figure 3): thematic classes have been identified on the image
(Table 2):

Table 2. The thematic classes used.

The classes Image Image Description

includes buildings that appear with a dark gray,

light gray, or white color, with well-
Buildings
distinguished geometric shapes.

3
E3S Web of Conferences 364, 04001 (2023) https://doi.org/10.1051/e3sconf/202336404001
JOE3

constructed of materials that appear with a dark

Roads
color having an elongated shape

Bare ground These are bare land, with a yellowish color.

Is characterized by the density of the vegetation

and the saturation of the green color which
Trees appears red when viewing the image in false
colors.

Ground covered with little less dense vegetation

Pastures that appears with a lighter red color.

This is the darkest part of the image. Usually

Shadow black.

2.3.2 Identification of attributes and indices (NIR) channels. The normalized vegetation index
highlights the difference between the visible red band
We mainly used the spectral attributes based on the and the near infrared band. It is widely used for the
spectral values of the pixels in the different bands of the discrimination of vegetation type objects. It was chosen
satellite image. These spectral values are used either to spectrally characterize the pixels of vegetation and
directly or to calculate other indices that will facilitate bare soil.
the classification.
 − 
• Spectral response in the bands  =
In our case, it is adopted to characterize the shadow  + 
This index is sensitive to the vigor and quantity of
pixels. Indeed, the shadow has low radiometric values vegetation. The value of the NDVI ranges between -1
in the 4 bands of the image used. and +1, the greater the biomass, the more the index tends
towards 1. Generally, bare or plowed soils have values
• Normalized Vegetation Index (NDVI) between 0 and 0.3, areas with little covering vegetation
The normalized difference vegetation index, also called between 0.3 and 0.6 and vegetation cover zones between
NDVI, is constructed from the red (R) and near infrared 0.6 and 1. It is the most widely used vegetation index

4
E3S Web of Conferences 364, 04001 (2023) https://doi.org/10.1051/e3sconf/202336404001
JOE3

and therefore constitutes a reference value when looking Table 3. The thresholds used in the classification.
at plant cover. Index Threshold Property
• The route extraction report Spectral
NDVI > 0,6 response of
It is calculated from the blue channel (B) and the near pastures
infrared channel (PIR) of the image according to the Spectral
following formula: NDVI 0,3<NDVI<0,6 response of
trees
 −  Ground
 +  0.1 <NDVI<0,3 spectral
This report is used to discriminate between the class of
response
roads and the class of buildings. Indeed, the roads have
low values of this ratio which facilitates their (Bleu < 225) and
Shadow
classification. Band (Vert <330) and
Spectral
signatures (Rouge < 22)
Response
and (PIR<250)
2.3.3 Analysis of the discriminating power of
attributes
Spectral
(B-PIR) (B-PIR)
response of
/(B+PIR) /(B+PIR) < 1
This analysis aims to study the discriminating power of buildings
the different quantifiable attributes. It seeks to associate Spectral
with each attribute a threshold with respect to which the (B-PIR) (B-PIR)
response of
attribute characterizes the abstract property that it /(B+PIR) /(B+PIR) > 1
roads
describes. For this, the analysis consists in first studying
the mathematical formulation and the variability of each
attribute[14]. In this study, the thresholds were set on the 2.3.4 Construction of the tree and classification
basis of their use in the literature and by testing. Thus,
for the NDVI vegetation index, the thresholds have been The construction of the decision tree will be done per
set considering the values given in the literature by [15]. node, each node will include a test on an index or a given
For the other indices and spectral values, the thresholds spectral value, with the threshold which has been judged
were set after performing several tests until satisfactory the most discriminating between the two classes or
results were obtained. groups of resulting classes. We will follow the following
architecture to build the decision tree (figure 4):

Fig. 4. The decision tree built on Envi.

5
E3S Web of Conferences 364, 04001 (2023) https://doi.org/10.1051/e3sconf/202336404001
JOE3

Table 4 presents the evaluation of the results of the

2.3.5 Accuracy assessment classification according to the value of the Kappa
coefficient.
No result - even if quantitatively proven - will convince,
if it does not satisfy the human eye [16]. The result will
Table 4. Classification categories according to “Landis and
be evaluated first visually; by measuring its similarity Coch” [17]
with the ground truth. The image used in our application
covers an urban area. The comparison is based Kappa 0-0.20 0 .21-0.40
essentially on the criteria of shape and delimitation of
buildings, roads, trees and pastures. Bare ground and The classification result Very weak Weak
shade are also taken into consideration. We will use for
the evaluation the error report which will contain the
confusion matrix and the indices which are derived from 3 Results and discussion
it namely: the global accuracy and the kappa coefficient. The execution of the decision tree built on Envi gives
The value of the kappa coefficient varies from 0 to 1,
the following result (figure 5).

Fig. 5. Image classified by the decision tree.

the same Quickbird image we find that the decision tree

According to the visual interpretation of the result by the gave better global precision than the K-means,
classification carried out by the decision tree, it can be ISODATA, minimal distance, Mahalanobis distance as
seen that the image obtained is very satisfactory and shown in Figure 6. The value of the Kappa Coefficient
very faithful to the reality on the ground. All classes are also exceeded the values obtained by the other
present and the geometric shapes have been well algorithms.
respected. The qualitative parameters that evaluate the
result obtained during the classification by the decision
100 Global precision
tree are presented in the following table (table 5):
80
Table 5. Evaluation of classification accuracy by the decision 60
tree.
40
Kappa
Classification Global precision
Coefficient 20
Tree decision 90.22% 88.36% 0

The classification by the decision tree gave an "overall

precision" of 90.22% and a "kappa" coefficient of
88.36% and, hence the result is considered very good
and very satisfactory. If we compare the result of
classification by the decision tree with the results Fig. 6. Diagram of comparison between classification
obtained by some traditional algorithms used to classify methods.

6
E3S Web of Conferences 364, 04001 (2023) https://doi.org/10.1051/e3sconf/202336404001
JOE3

Indeed, products based on decision trees are easy to use, 8. C. Zhang, A. Marzougui, and S. Sankaran,
very visual and their implementation is very intuitive. It Comput. Electron. Agric. 175, 105584 (2020),
is a very understandable and easily interpretable white https://doi.org/10.1016/j.compag.2020.105584
box model thanks to its meaningful and informative 9. C. Yang, G. Wu, K. Ding, T. Shi, Q. Li, and J.
graphical representation[18]. It also makes it possible to Wang, Remote Sens. 9(12), 1222 (2017),
manage and use data from different sources and types. https://doi.org/10.3390/rs9121222
10. D. Lu, Q. Weng, Int. J. Remote Sens. 28(5), 823–
4 Conclusion 870 (2007),
https://doi.org/10.1080/01431160600746456
In this paper, Decision tree technique is applied to 11. S. N. MohanRajan, A. Loganathan, and P.
classify a very high resolution satellite image in order to Manoharan, Environ. Sci. Pollut. Res. 27(24),
get the land cover presentation. The main objective of 29900–29926 (2020),
this paper is to test a machine learning technique and https://doi.org/10.1007/s11356-020-09091-7
evaluate it performance but also to help researchers
choose between the different available image 12. A. E. Maxwell, T. A. Warner, F. Fang, Int. J.
classification techniques. The decision tree has proven Remote Sens. 39(9), 2784–2817 (2018),
its efficiency and performance with a 90.22% for global https://doi.org/10.1080/01431161.2018.1433343
precision and 88.36% for Kappa coefficient. If 13. F. F. Camargo, E. E. Sano, C. M. Almeida, J. C.
compared with traditional classification algorithms, Mura, and T. Almeida, Remote Sens. 11(13), 1600
these results outperformed them all. (2019), https://doi.org/10.3390/rs11131600
These findings provide insights into the selection of 14. B. Charbuty and A. Abdulazeez, J. Appl. Sci.
classifiers and highlights the importance of the decision Technol. Trends. 2(01), 20–28 (2021),
tree method as a machine learning classification tool. It https://doi.org/10.38094/jastt20165
is a clear model characterized by simplicity of 15. A. Bégué, V. Lebourgeois, E. Bappel, P. Todoroff,
understanding and interpretation that has proven its A. Pellegrino, F. Baillarin, B. Siegmund, Int. J.
efficiency and performance in the classification of high Remote Sens. 31(20), 5391–5407 (2010),
resolution satellite images. Finally, for a possible similar https://doi.org/10.1080/01431160903349057
study, it is recommended to.
• Include texture in the classification process. 16. M. Baatz, A. Schape, Multiresolution
• Use other indices and try other thresholds to build the Segmentation—An Optimization Approach for
decision tree. High Quality Multi-Scale Image Segmentation.
• Use an object-based approach in classification. Proceedings of the Angewandte Geographische
Informations Verarbeitung XII, Karlsruhe, 12-23
(2000).
References 17. J. R. Landis, G. G. Koch, Biometrics, 33(2), 363
(1977), https://doi.org/10.2307/2529786
1. I. Ezeomedo, J. Igbokwe, J. Wildl. Manage.
104(1), 218–221 (2013). 18. S. Ray, International Conference on Machine
Learning, Big Data, Cloud and Parallel Computing
2. A. Bose and I. R. Chowdhury, Model. Earth Syst.
(Com-IT-Con), 35–39 (2019),
Environ. 6(4), 2235–2249 (2020).
https://doi.org/10.1109/COMITCon.2019.8862451
https://doi.org/10.1007/s40808-020-00842-6
3. R. Avtar, P. Kumar, A. Oono, C. Saraswat, S.
Dorji, Z. Hlaing, Geocarto Int. 32(8), 874–885,
(2017),
https://doi.org/10.1080/10106049.2016.1206974
4. G. Warth, A. Braun, O. Assmann, K. Fleckenstein,
V. Hochschild, Remote Sens. 12(11) (2020),
https://doi.org/10.3390/rs12111730
5. Z. Shao, N. S. Sumari, A. Portnov, F. Ujoh, W.
Musakwa, and P. J. Mandela, Geo-Spatial Inf. Sci.
24(2), 241–255 (2021),
https://doi.org/10.1080/10095020.2020.1787800
6. J. Hoffmann, Hydrogeol. J. 13(1), 247–250
(2005), https://doi.org/10.1007/s10040-004-0409-
2
7. I. Nurwauziyah, D. S. Umroh, I. Gede, B. Putra,
and M. I. Firdaus, Satellite Image Classification
using Decision Tree, SVM and k-Nearest
Neighbor, July, (2018).