You are on page 1of 4

Proceedings of the 2nd International conference on Electronics, Communication and Aerospace Technology (ICECA 2018)

IEEE Conference Record # 42487; IEEE Xplore ISBN:978-1-5386-0965-1

Survey on Classification techniques used in remote


sensing for satellite images

Rahul Neware Prof. Amreen Khan


Computer Science and Engineering Department Computer Science and Engineering Department
GHRCE, Nagpur GHRCE, Nagpur
Nagpur, India Nagpur, India
rneware00@gmail.com Amreen.khan@raisoni.net

Abstract—Remote sensing deals with high resolution of In remote sensing there are two types of sensing,
geospatial data. It work on the various fields like agriculture,
water resource, environment, forests, land use or land cover, Passive sensing:
ocean applications, soil, geosciences, disaster warning and The energy which is generated by natural source that
management. To get required information from satellites images is sun, which is measured by remote sensing is called passive
we need to do some image processing on it. Remote sensing image sensor. Advantage of passive senses that is freely available and
processing consists of various steps in which classification and
limitless. Only limitation of passive sensor is it is available
segmentation of image into its land cover type is very essential
only at day time[3].
part. This paper gives classification technique used to find the
land cover classes in satellite images. In this paper three most
used classification ion technique is studied that is supervised
classification, unsupervised classification and Object oriented
classification.

Keywords— Remote sensing, Classification, Segmentation,


Sensors, Land cover and Use

I. INTRODUCTION
Remote sensing is the science (and to some extent, art) of Fig 1.1 Passive Sensing
acquiring information about the Earth's surface without
actually being in contact with it. This is done by sensing and (Source: http://www.nrcan.gc.ca/node/14639)
recording reflected or emitted energy and processing, Active sensing:
analyzing, and applying that information”[1]. Process of
remote sensing consist of sending radiation from satellite to The energy which is generated by satellites own
earth or on object which has to be studied. As the radiation is energy source like Electro Magnetic Radiation, Radar,
incident on object it reflect the energy. This reflected energy Microwave etc. then which is measured by remote sensing is
then captured by remote sensing satellites and sent to remote called passive sensors. Active sensors sends radiations to the
station for converting it into images[2]. earth surface object which has to be investigated and reflected
energy then collected for further processing. Advantage of
In remote sensing data is gathered from the earth passive sensor is getting object information at any time and
surface with the help of energy reflected by the object on earth disadvantage is need huge energy to acquitter object[3].
surface. Remote sensing deals with the collecting of data in
image from which is reflected energy. Collected image
information is then processed with GIS (Geographical
information system)then analyzed it and apply this information
to solve the real time problems. Remote sensing technique is
very simple and they collect senses data to further process it.
First, the energy or some radiations are emitted from sensors to
the earth surface to target object. These objects reflects the
energy which is again captured by the sensor and send it to the
remote sensing centers in the form of images. Then remote
sensing centers process image data and find out required Fig 1.2 Active Sensing
information like water bodies, agriculture, situation analysis,
land cover classification etc. (Source: http://www.nrcan.gc.ca/node/14639)

978-1-5386-0965-1/18/$31.00 ©2018 IEEE 1860


Proceedings of the 2nd International conference on Electronics, Communication and Aerospace Technology (ICECA 2018)
IEEE Conference Record # 42487; IEEE Xplore ISBN:978-1-5386-0965-1

II. SENSORS USED IN REMOTE SENSING for improving the satellites image classification accuracy using
Hrs. After using this new technique classification is very much
Following sensors are used in remote sensing, improved. J. Yuan, D. Wang [10] given 3D model for high
2.1 Airborne sensor: resolution image to detect car based on reference disruption.
This model detect only one vehicle at a time but it shows all the
In airborne sensors are kept on aircraft which gives features of car very clearly like car shadows, color and other
high spatial resolution. When we use airborne sensors it features. Q. Yu, P. Gong [11] proposed segmentation tool
perform only one time operation. Only the problem while using based on the technique of semi-automatic seamen nation. This
airborne sensor is that it covers very less area[4]. technique is the fusion of automatic manual segmentation.
2.2 Space borne sensor: Where automatic based on bimodal technique and manual
based on dropping and filling splines. This technique gives
In space borne sensors are kept on satellite or shuttles. very good segmentation of selected area but the problem is it
It covers more earth surface as compared with air borne sensor. involves manual validation process. M. Celenk [12] given new
When we use space borne sensors it perform continuous technique for temporal change detection in land cover. They
monitoring of earth surface. Only problem is that it gives used three techniques image segmentation, Gamma map filter
images with low spatial resolution. and block average. For segmentation they used thresholding
2.3 Radar: technique but while using thresholding efficiency is
automatically less.
Radar sensors used to measure distance between
devices. In radar Electro-Magnetic radiation is send to earth E. Berthier, E. Schiefer [13] proposed a technique
surface object which incident on it in right angle and reflected which automatically segment and classify the satellites image.
energy is recorded back. It is used for calculating distance and They used data of Rostockuty, Germany form of QuickBird
also to detect moving objects like tracking ships, wind and LiDAR images. They find k-coefficient of images and then
direction, land cover classification ion etc.[5]. that coefficients is compared with other algorithm like SVM,
TF, CTF and MRF. Accuracy of segmentation and
classification by this technique is very high. G. Baudat and F.
III. CLASSIFICATION TECHNIQUES Anouar [14] given a technique for 3D train change detection
In remote sensing doing classification ion is very important and investigation using DEMs (Digital elevation models). They
step. By doing classification of remote sensing images we can studied deformation in earth surface by earthquake using
analyzed and find patterns of required area. Mainly stereoscopic broom push technique. C. S. Fraser and H. B.
classification is applied to find out land cover classification, Hanley [15] given technique for Dasy metric mapping using
vegetation, water bodies, cloud extraction etc. object oriented classification for remote sensing imagery. This
method eliminates ecological error and also various problems
occurred while using other techniques. L. Zhang, X. Huang
3.1 Object Oriented Classification: [16] given technique for analyzing damages in earth surface
with depth and volume. For experiment they used very high
Object oriented classification is alternative for pixel resolution images of SPOT5.
based classification object oriented classification specially use
for high resolution imagery. Mainly object-oriented
classification segment satellite image into its objects so that is 3.2 Supervised Classification:
called object oriented classification. In object oriented
classification knowledge base is created for segmented object Supervised classification is the type of machine
like, agency, length, shape and size etc. Object oriented learning in which training input is given then according
classification gives very accurate classified result of high training inputs classification is applied. Supervised
resolution satellites image. classification is most used classification algorithm in remote
sensing. After applying this technique output is classified
J. Goldberger, S. Gordon [6] invented technique of image according to each class called training. In supervised
multi-thresholding for classification. Work on the principal of classification user has to select region of interest which is act as
thresholding. The threshold value selected on the moments in classifier and pixels of whole image is classified according to
gray level. Segmentation and classifications accuracy is very region of interest. Maximum likelihood and Spectral mapping
high. A. Ramachandra, S. Abhilash [7] given a combined classification algorithm is mostly used while performing
technique feature enhancement and object oriented supervised classification in remote sensing[17].
classification. Studied the output results and checked the
accuracy of object oriented classed result. R. R. Muskett, C. S. C. Xiao, M. Liu [18] studied spectral and spatial
Lingle [8] developed new algorithm which used for change property of remote sensing images. They used polynomial fit
detection in SAR images annual basis. Thresholding technique classification ion teen to classify land cover and objects on
of cumulative history and 5X5 mean filter is used for the earth surface. But by using this method minute required details
element ion of noise. The accrued of classification ion by this of object is not get after classification. N. Gillis, D. Kuang [19]
algorithm is very high and also validated. used concept of graph theory for the classification of various
area especially urban area from satellites image. Then after
Karoui, R. Fablet [9] given a technique which is the classification ion the accuracy they achieved is very very high.
combination of object classification and feature enhancement They used images which has 1m resolution. T. Mei, L. An [20]

978-1-5386-0965-1/18/$31.00 ©2018 IEEE 1861


Proceedings of the 2nd International conference on Electronics, Communication and Aerospace Technology (ICECA 2018)
IEEE Conference Record # 42487; IEEE Xplore ISBN:978-1-5386-0965-1

they invented a new technique for showing different type of segmentation and fuzzy algorithm. They classify quick bird
land cover segments by combining tabour filter and variable images for getting grass, forest and vegetation land. J. Feng, L.
means shift. Jiao [28] proposed DTM (Digital Terrain Models) technique
which is the extension of DSM (Digital Surface Models) this
L. Gueguen [21] invented technique for the method is used to enhance and eliminate the disadvantage of
supervised classification named pixel based system. Data DSM. It takes input from DSM and filter ground point and fits
I’mages used in this experiment are very high spatial resolution that ground point to get accurate results.
images. Output provided by this technique is accurate map and
information of spatial context. M. Espinola, J. A. Piedra [22] T. Daniels, W. L. Smith [29] they make some changes
proposed classification algorithm based on POK. Supervised in segmentation and fuzzy classification techniques to get k-
classification is done on the panchromatic data but the accuracy coefficient, which gives very accurate land cover classification
of the classified output is very low. S. Kraft, U. Del Bello [23] ion for vegetation, grassland and forest land. L. Bruzzone and
proposed a TS-MRF technique for supervised classification. It L. Carlin [30] they given comparison between various
gives very good accuracy after classification but non stationary unsupervised clustering algorithms (DBSCAN, K-mean and
modeling property is not did by this technique. F. Ling, Y. SOM). For non-convex cluster they said out of DBSCAN gives
Zhang [24] classify satellites images according to their land very good results than k-mean and SOM. K-mean and SOM
cover class and land use. For that they proposed GRFc and algorithms are used when the data size is high and cluster shape
used fuzzy system and compared their results. Finally they got is convex. A. Baraldi and F. Parmiggiani [31] invented
high accuracy of classification using GRFc algorithm than unsupervised classification algorithm using cellular automata
fuzzy system. technique. The output of this classification ion is much better
than other algorithm that is RBF, multilayer perceptron, kNN
and Naive Bayes. They perform classification of all the bands
of image that is 1to 6 and get very good accuracy of
classification but this classification ion does not work for
texture data classification. I. Sola, M. Gonzalez [32] proposed
Mahalanobis classifier. They used it for the land cover
classification ion and noted the performance of this classifier.
The performance and occupy of this classifier is very high.
This classifier shows the importance of classifier and data set
relationship for classification.

IV. CONCLUSION
After This paper gives the detailed study of classification
technique used in remote sensing for satellite images. Types of
sensors used and active and passive sensing is important part of
remote sensing. Object-based, Supervised and Unsupervised
type of classification are explained with their recent work.
Supervised classification technique is mostly used in remote
sensing because its accuracy of classification is very high as
compared to other technique.

REFERENCES
Fig 3.2.1 Supervised Classification Flowchart
[1] Schowengerdt, Robert A. (2007). Remote sensing: models and methods
for image processing (3rd ed.). Academic Press. p. 2. ISBN 978-0-12-
3.3 Unsupervised Classification: 369407-2.
[2] Liu, Jian Guo & Mason, Philippa J. (2009). Essential Image Processing
S. Bharathi, V. Shreyas [25] Proposed DSM (Digital for GIS and Remote Sensing. Wiley-Blackwell. p. 4. ISBN 978-0-470-
Surface models) technique which maps the object points of 51032-2.
surface in satellites image. They proposed this technique and [3] http://www.nrcan.gc.ca/node/14639
chicken accrual and robustness of technique in agriculture [4] Y. Han, H. Kim, J. Choi, and Y. kim, ―A Shape Size Index Extraction
grounded land and hill area. Accuracy and robustness of this for Classification of High Resolution Multispectral Satellite Images,‖
technique is very good. L. Bruzzone and L. Carlin [26] International Journal of Remote Sensing, vol. 33, no. 6, pp. 1682–1700,
proposed technique which prefer segmentation on spectral and 2012.
texture of image. This technique reduced representative of [5] B. Yektakhah and K. Sarabandi, ―All-Directions Through-the-Wall
known and unknown features. In this technique no need to Radar Imaging using a Small Number of Moving Transceivers,‖ IEEE
Transactions on Geoscience and Remote Sensing, vol. 54, no. 11, pp.
select scales every time, it selects scale automatically which 6415–6428, 2016.
does not require segmentation. Due to the projection problem [6] J. Goldberger, S. Gordon, and H. Greenspan, ―Unsupervised Image-Set
finer details are not proper. Y. Wang, A. I. Lyapustin [27] Clustering using an Information Theoretic Framework,‖ IEEE
given unsupervised classification technique by using Transactions on Image Processing, vol. 15, no. 2, pp. 449–458, 2012.

978-1-5386-0965-1/18/$31.00 ©2018 IEEE 1862


Proceedings of the 2nd International conference on Electronics, Communication and Aerospace Technology (ICECA 2018)
IEEE Conference Record # 42487; IEEE Xplore ISBN:978-1-5386-0965-1

[7] A. Ramachandra, S. Abhilash, R. KB, and K. R. Venugopal, ―Feature [20] T. Mei, L. An, and Q. Li, ―Supervised Segmentation of Remote
Level Fusion based Bimodal Biometric using Transformation Domine Sensing Image using Reference Descriptor,” IEEE Geoscience and
Techniques,‖ IOSR Journal of Computer Engineering (IOSRJCE), vol. 3, Remote Sensing Letters, vol. 12, no. 5, pp. 938–942, 2015.
no. 3, pp. 39–46, 2012. [21] L. Gueguen, ―Classifying Compound Structures in Satellite Images: A
[8] R. R. Muskett, C. S. Lingle, J. M. Sauber, A. S. Post, W. V. Tangborn, Compressed Representation for Fast Queries,‖ IEEE Transactions on
B. T. Rabus, and K. A. Echelmeyer, ―Airborne and Spaceborne DEM Geoscience and Remote Sensing, vol. 53, no. 4, pp. 1803–1818, 2015.
and Laser Altimetry- Derived Surface Elevation and Volume Changes of [22] M. Espinola, J. A. Piedra-Fernandez, R. Ayala, L. Iribarne, and J. Z.
the Bering Glacier System, Alaska, USA, and Yukon, Canada, 1972– Wang, ―Contextual and Hierarchical Classification of Satellite Images
2006‖, Journal of Glaciology, vol. 55, no. 190, pp. 316–326, 2013. based on Cellular Automata,‖ IEEE Transactions on Geoscience and
[9] I. Karoui, R. Fablet, J.-M. Boucher, and J.-M. Augustin, ―Variational Remote Sensing, vol. 53, no. 2, pp. 795–809, 2015.
Region-based Segmentation using Multiple Texture Statistics,‖ IEEE [23] S. Kraft, U. Del Bello, M. Bouvet, M. Drusch, and J.Moreno, ―FLEX:
Transactions on Image Processing, vol. 19, no. 12, pp. 3146–3156, 2013. ESA’s Earth Explorer 8 Candidate Mission,‖ IEEE Transactions on
[10] J. Yuan, D. Wang, and R. Li, ―Remote Sensing Image Segmentation by Geoscience and Remote Sensing, vol. 53, no. 7, pp. 7125–7128, 2016.
Combining Spectral and Texture Features,‖ IEEE Transactions on [24] F. Ling, Y. Zhang, G. M. Foody, X. Li, X. Zhang, S. Fang, W. Li, and
Geoscience and Remote Sensing, vol. 52, no. 1, pp. 16–24, 2014. Y. Du, ―Learning-based Super resolution Land Cover Mapping,‖ IEEE
[11] Q. Yu, P. Gong, N. Clinton, G. Biging, M. Kelly, and D. Schirokauer, Transactions on Geoscience and Remote Sensing, vol. 54, no. 7, pp.
―Object-Based Detailed Vegetation Classification with Airborne High 3794–3810, 2016.
Spatial Resolution Remote Sensing Imagery,‖ Photogrammetric [25] S. Bharathi, V. Shreyas, R. Anirudh, S. Sanketh, P. D. Shenoy, K.R
Engineering Remote Sensing, vol. 72, no. 7, pp. 799– 811, 2014. Venugopal, and L.M Patnaik, ―Performance Analysis of Segmentation
[12] M. Celenk, ―A Color Clustering Technique for Image Segmentation,‖ Techniques for Land Cover Types using Remote Sensing Images,‖ 2012
Computer Vision, Graphics, and Image Processing, vol. 52, no. 2, pp. Annual IEEE India Conference (INDICON), pp. 775– 780, 2012.
145–170, 2014. [26] L. Bruzzone and L. Carlin, ―A Multilevel Context-Based System for
[13] E. Berthier, E. Schiefer, G. K. Clarke, B. Menounos, and F. Remy, Classification of Very High Spatial Resolution Images,” IEEE
―Contribution of Alaskan Glaciers to Sea-Level Rise Derived from transactions on Geoscience and Remote Sensing, vol. 44, no. 9, pp.
Satellite Imagery,‖ Nature Geoscience, vol. 3, no. 2, pp. 92–95, 2014. 2587–2600, 2012.
[14] G. Baudat and F. Anouar, ―Generalized Discriminant Analysis using a [27] Y. Wang, A. I. Lyapustin, J. L. Privette, J. T. Morisette, and B. Holben,
Kernel Approach,‖ Neural computation, vol. 12, no. 10, pp. 2385–2404, ―Atmospheric Correction at AERONET Locations: A New Science and
2015. Validation Data Set,‖ IEEE Transactions on Geoscience and Remote
[15] C. S. Fraser and H. B. Hanley, ―Bias-Compensated RPCs for Sensor Sensing, vol. 47, no. 8, pp. 2450–2466, 2012.
Orientation of High-Resolution Satellite Imagery,‖ Photogrammetric [28] J. Feng, L. Jiao, X. Zhang, and D. Yang, ―Bag-of- Visual-Words based
Engineering and Remote Sensing, vol. 71, no. 8, pp. 909–915, 2015. on Clonal Selection Algorithm for SAR Image Classification,‖ IEEE
[16] L. Zhang, X. Huang, B. Huang, and P. Li, ―A Pixel Shape Index Geoscience and Remote Sensing Letters, vol. 8, no. 4, pp. 691–695,
Coupled with Spectral Information for Classification of High Spatial 2013.
Resolution Remotely Sensed Imagery,‖ IEEE Transactions on [29] T. Daniels, W. L. Smith, and S. Kireev, ―Simulation of Airborne
Geoscience and Remote Sensing, vol. 44, no. 10, p. 2950, 2016. Radiometric Detection of Wake Vortices,‖ IEEE Transactions on
[17] S. Jia, G. Tang, J. Zhu, and Q. Li, ―A Novel Ranking- Based Clustering Geoscience and Remote Sensing, vol. 53, no. 12, pp. 6336–6343, 2015.
Approach for Hyperspectral Band Selection,‖ IEEE Transactions on [30] L. Bruzzone and L. Carlin, ―A Multilevel Context-based System for
Geoscience and Remote Sensing, vol. 54, no. 1, pp. 88–102, 2016. Classification of Very High Spatial Resolution Images,‖ IEEE
[18] C. Xiao, M. Liu, D. Xiao, Z. Dong, and K.-L. Ma, ―Fast Closed-Form transactions on Geoscience and Remote Sensing, vol. 44, no. 9, pp.
Matting using a Hierarchical Data Structure,” IEEE Transactions on 2587–2600, 2014
Circuits and Systems for Video Technology, vol. 24, no. 1, pp. 49–62, [31] A. Baraldi and F. Parmiggiani, ―A Refined Gamma MAP SAR Speckle
2014. Filter with Improved Geometrical Adaptivity,‖ IEEE Transactions on
[19] N. Gillis, D. Kuang, and H. Park, ―Hierarchical Clustering of Geoscience and Remote Sensing, vol. 33, no. 5, pp. 1245–1257, 2015.
Hyperspectral Images using Rank-Two Nonnegative Matrix [32] I. Sola, M. Gonzalez-Audicana, J. Alvarez-Mozos, and J. L. Torres,
Factorization,‖ IEEE Transactions on Geoscience and Remote Sensing, ―Synthetic Images for Evaluating Topographic Correction
vol. 53, no. 4, pp. 2066–2078, 2015. Algorithms,” IEEE Transactions on Geoscience and Remote Sensing,
vol. 52, no. 3, pp. 1799–1810, 2014

978-1-5386-0965-1/18/$31.00 ©2018 IEEE 1863

You might also like