Journal of King Saud University – Computer and Information Sciences xxx (2018) xxx–xxx

Contents lists available at ScienceDirect

Journal of King Saud University –

Computer and Information Sciences
journal homepage: www.sciencedirect.com

Modernizing the multi-temporal multispectral remotely sensed image

change detection for global maxima through binary particle swarm
optimization
Anisha M. Lal ⇑, S. Margret Anouncia
Vellore Institute of Technology, SCOPE, Vellore, India

a r t i c l e i n f o a b s t r a c t

Article history: Change detection is the amount of changes that can guide to more concrete understandings into essential
Received 30 May 2018 method concerning land cover, land usage and ecological variations. This paper deals with an intelligent
Revised 4 October 2018 methodology to optimize the solution out of the solution space that improves the efficiency of the change
Accepted 16 October 2018
detection process. To support the theme, an integrated semi-supervised method is designed with a focus
Available online xxxx
on image fusion, semi supervised clustering and binary swarm based optimization. A new approach of
handling fusion using sparse coding is referred to expand the amount of information. Using the extracted
Keywords:
information, change detection process is carried out by a constrained clustering technique to provide a
Change detection
Binary swarm based optimization
solution reflecting the level of changes occurred in the region of investigation. To improve the accuracy
Semi-supervised by proposing a global optimum solution, still the result is refined through binary swarm based optimiza-
Sparse coding tion process and hence the results are accelerated towards the increased level of accuracy followed by
which the change map is reconstructed to show case changes prominently. To determine the accurate-
ness of the proposed methodology quantitative and qualitative analysis has been done with different
datasets. The proposed method has been evaluated with existing techniques such as k-means, AKM,
FCM, ECKM and ASCC to show the efficiency and proved to be the preeminent change detection method-
ology compared to the state-of-art methods.
Ó 2018 Production and hosting by Elsevier B.V. on behalf of King Saud University. This is an open access
article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

1. Introduction mental and urban changes, the issues are very difficult to address.
Irrespective of all difficulties, one of the popularly adapted tech-
Change detection is a process of detecting the changes in the nologies to address the issue is remote sensed image processing
state of an object or phenomenon by observing it at different times composed of enhancement, image comparison and image analysis.
(Singh, 1989; Bottomley, 1998). The changes in an image may be Many dynamic attempts had been carried out towards the change
due to shape, movement, occurrence or desertion of an object. detection of remotely sensed images. Most of the approaches fall
Especially, in the remotely sensed images, it is helpful in map under the category of supervised or unsupervised methods that
apprising, land use analysis, monitoring shifting cultivation, urban use multi-temporal datasets to qualitatively analyse the temporal
planning, deforestation assessment, disaster monitoring, thermal effects and measure the changes. The methodology is adopted
analysis of day & night, tracking urban and economic growth. either fully or partially on the sub processes such as enhancement,
The complexity involved in the applications leaves great challenges fusion, spatial difference identification, post classification and
in the change detection. Especially when dealing with environ- change map formulation. The general objectives of change detec-
tion in remote sensing include recognizing the environmental
locality and type of changes, measuring the changes, and calculat-
⇑ Corresponding author. ing the accurateness of change detection. Ultimately, the main mis-
E-mail addresses: [email protected] (A.M. Lal), [email protected] sion of change detection is to identify and recognize the changes in
(S.M. Anouncia).
multi-temporal multispectral Landsat images of the same geo-
Peer review under responsibility of King Saud University.
graphic area acquired at different timings.
Several evolutionary computing based optimization techniques
are used to optimize the solution out of solution space. One of the
most widely used techniques is the particle swarm optimization.
Production and hosting by Elsevier
Particle swarm optimization (PSO) is a computational method that

https://doi.org/10.1016/j.jksuci.2018.10.010
1319-1578/Ó 2018 Production and hosting by Elsevier B.V. on behalf of King Saud University.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
Please cite this article in press as: Lal, A.M., Anouncia, S.M. Modernizing the multi-temporal multispectral remotely sensed image change detection for
global maxima through binary particle swarm optimization. Journal of King Saud University – Computer and Information Sciences (2018), https://doi.
org/10.1016/j.jksuci.2018.10.010
2 A.M. Lal, S.M. Anouncia / Journal of King Saud University – Computer and Information Sciences xxx (2018) xxx–xxx

Difference Image
LANDSAT Image Creation Binary
T1 Change Map Optimized Change Map
generated Using BPSO
Image Fusion using using ASCC
Enhanced Sparse

LANDSAT Image Change

T2 Detected
Semi-Supervised
Clustering Image

ASCC

Fig. 1. Overall block diagram of Change Detection Process.

enhances a problem by iteratively trying to develop an optimal tion (Lal and Anouncia, 2015b). Among these applications, with
solution with regard to a given degree of quality. It is a global opti- the advent of the high spatial resolution satellite images, the
mization method to provide a best solution when dealing with change detection using multispectral multi-temporal Landsat
problems which can be represented as a point or surface in an n- images has been an important field of research. Nielsen (2007) pro-
dimensional space. It is an effective and popular optimization posed a multivariate alteration detection (MAD) method for
method to explore the optimal solution in the search space prob-
lems. Keeping this scope, an efficient method and a cost function
to optimize the change detection process accurately is envisioned. Start

Thus, working with change detection of remote sensed images,

where a huge number of iterations are required to find the final
change detection mask or optimal result is alleviated with the Initial population is the
change mask created using
introduction of this efficient methodology. ASCC
The organisation of this paper is as follows. Section 2 presents
some of the related works and the gaps identified. Section 3 gives
an outline of the methodology adapted and describes the proposed Randomly generate other
system in detail. In Section 4, description about dataset, experi- particles based on initial
change mask
mental results and discussion are described. Performance evalua-
tion of results for change detection are analysed in Section 5.
Finally, Section ‘‘References” draws the conclusions of this work.
Assign random position
and velocity to the
particles
2. Related work

Remote sensing is the process of collecting information using

Calculate fitness value for
sensors which are not in physical contact with surfaces and phe- each particle
nomenon of interest, which provides details about earth surfaces
and phenomenon. Remote sensing includes the choice of sensors,
the reaction, recording and processing of signal data and finally
the analysis of the subsequent data. The varieties of remote sensing
Is current position
methods existing are such as aerial photography, multispectral, better than local best Yes
Set current position
as the local best
active and passive microwave. The information content of a single position

image is limited by the spatial and spectral resolution of the imag-

ing system. In spite of all the difficulties, the remote sensed images
No
are analysed frequently for a variety of applications ranging from
Land cover classification, Change Detection, Water Quality Moni- No
Assign best particles
toring, Measurement of sea surface temperature, Snow survey, local best value to
global best
Monitoring of Atmospheric Constituents to Geological Interpreta-

BPSO Update state and

velocity of each
Binary Change particle
Initial population binary
Map generated
change map
using ASCC

Targets or
Randomly generate maximum epochs
other populations reached

Yes
Change Detected
Swarm based optimizer
Image
End

Fig. 2. Optimization of Change Map using BPSO. Fig. 3. Flow Diagram for Binary PSO.
Please cite this article in press as: Lal, A.M., Anouncia, S.M. Modernizing the multi-temporal multispectral remotely sensed image change detection for
global maxima through binary particle swarm optimization. Journal of King Saud University – Computer and Information Sciences (2018), https://doi.
org/10.1016/j.jksuci.2018.10.010
 A.M. Lal, S.M. Anouncia / Journal of King Saud University – Computer and Information Sciences xxx (2018) xxx–xxx 3

change detection in multi-temporal remote sensing images. Xian rately an effective optimization method has to be applied. PSO per-
et al., (2009) used CVA for change detection with other techniques forms optimization based on a swarm of particles. Swarm based
such as MAD to normalize and classified the changes using tree optimization has been widely used in different problems such as
methods. Wu et al., (2017a) has concentrated on post classification image enhancement (Braik et al., 2007), face and iris recognition
using Iterative slow feature analysis and Bayesian soft fusion to (Ramadan and Abdel Kader, 2009), feature extraction (Xue et al.,
obtain reliable and accurate change detection maps. Wu et al., 2012), image coding (Kusetogullari, 2015) and remote sensed image
(2017b) has proposed a novel method using kernel new feature classification (Soliman et al., 2012). Change detection with less noise
analysis and a post classification fusion is done to identify the tran- using PSO has been proposed by Sagnika et al. (2015). The method
sition type. In order to improve the dependability for extracting has produced better accuracy with minimal noise. A Parallel Binary
remotely sensed information, and enhance the efficiency of change Particle Swarm Optimization (PBPSO) unsupervised approach for
detection, the process is carried out through common image pro- change detection in multi-temporal multispectral images has been
cessing tasks such as image enhancement followed by image com- proposed by Kusetogullari et al. (2015). The approach provides bet-
parison, region detection and analysis. In some cases, due to ter convergence and detection performance through parallel pro-
limited convergence towards accuracy, some of the optimization cessing. Change detection using particle swarm optimization in
techniques also had been employed along with image processing spherical coordinates has been proposed by Yavariabdi et al.
mechanisms to provide effective result. (2017) which provides less error than the existing methods. Thus
Generally the clustering approaches only locate the important an optimization process is attempted to reach the global optima
changes between two multispectral images and have deficiency of using Binary Particle Swarm Optimization (BPSO).
adaptableness and flexibility to attain feeble changes. Therefore, Based on the evaluations of the existing systems the change
computational methods are required to find optimum outcome by map formulated as a resultant of change map detection process
obtaining both feeble and robust changes on the difference image. has been less evaluated to provide the optimum solution. It was
To attain an optimal outcome, Celik (2010) proposed unsupervised also identified that the search over the spatial data of remote
change detection on the difference image by an optimization sensed images for change detection is tedious and time consuming
approach using Genetic Algorithm (GA). This method is based on a as it is difficult to interpret and label change information. In order
fitness function which minimizes according to the difference image. to overcome these issues an intelligent methodology to c which
GA was designed to create elements of binary change images which improves the efficiency of the change detection process.
are iteratively progressed to the final change detection mask. This
method successfully produced the change detection result on the 3. Methodology
difference image without a previous conventions. Particle Swarm
Optimization (PSO) is another effective computational based opti- An intelligent methodology is designed with a focus on image
mization method for searching the optimal solution in the search fusion, semi supervised clustering and binary swarm based opti-
space problems. A swarm based unsupervised change detection mization to improve the efficiency of the change detection process.
for remotely sensed images are proposed by Dai et al. (2010). In The overall block diagram of the change detection process adapted is
order to optimize the change detection problem efficiently and accu- shown in Fig. 1. The process starts with a pre-processing followed by

Fig. 4. (a) Landsat 5 image recorded on 12 October 1984 of Huelva, Spain. (b) Landsat 8 image recorded on 12 August 2014 of Huelva, Spain. (c) Landsat 5 Image Acquired on
October 24, 1984 of Dead Sea, Israel. (d) Landsat 8 Image Acquired on October 27, 2014 of Dead Sea, Israel.
Please cite this article in press as: Lal, A.M., Anouncia, S.M. Modernizing the multi-temporal multispectral remotely sensed image change detection for
global maxima through binary particle swarm optimization. Journal of King Saud University – Computer and Information Sciences (2018), https://doi.
org/10.1016/j.jksuci.2018.10.010
4 A.M. Lal, S.M. Anouncia / Journal of King Saud University – Computer and Information Sciences xxx (2018) xxx–xxx

integrated sparse and semi-supervised technique known as ASCC the convergence rate and minimizes the error in finding the final
(Lal and Margret Anouncia, 2017) for finding the change detection change detection mask. The method operates by considering the
mask. In order to obtain a global optimum solution an intelligent created initial solution for searching the optimal solution. The
methodology has been adapted with binary particle swarm opti- other populations are created randomly using the initial solution.
mization to further optimize the change detection mask. Thus the particle is assumed to move towards the gbest and once
A new approach of handling fusion using sparse coding is devised the specified convergence is met, it is assumed that the gbest is
to expand the amount of information (Lal and Anouncia, 2016). achieved.
Using the extracted information, change detection process is carried Thus, the change map formulated after applying BPSO will have
out by a constrained clustering technique to provide a solution changes projected prominently to result in improved visual effect
reflecting the level of changes occurred in the region of investiga- on the change map.
tion. Thus, the analysis towards the suitability of region detection The process followed is shown in Fig. 3.
methods pushed to a decision of choosing ASCC as a better method- The procedure adopted in the case of optimizing the solution
ology to perform change detection in the multi-temporal multispec- space of c as follows:
tral images. However, the solution obtained through these
clustering is the minimization of the objective function, so that it Input: Initial particle is the binary change map formulated using
is clear that it would reach only local maxima. In case of change ASCC
detection in remote sensed images, this may not provide the Output: Final Change detected image
required accuracy as the minor changes in the region may not be Choose binary change map formulated using ASCC as
revealed. Hence, it is required to optimize the solution so that all the initial population
the possible changes are reflected in the change map to make change Generate other populations randomly
detection effective. One of the feasible approaches to carry such pro- Initialize the populations with random positions and
cess is applying the binary change map obtained as result of region velocities
detection for optimization so that the global maximum is achieved. For each population
Thus the process demands to use evolutionary approaches that lead Evaluate the fitness value for each
to fulfil the objective of reaching global maxima. The popular meth- population as,
ods that converges solution to global maxima is PSO. 2
X 2
F i ¼ 4 ðN0  ADM ði; jÞ  meanðS0 ÞÞ
3.1. Particle swarm optimization 8ði;jÞ2S0

X 2
PSO is a computational optimization technique which performs þ ðN 1  ADM ði; jÞ  meanðS1 ÞÞ =ðM  NÞ
based on the association and shrewdness of swarms. It involves a 8ði;jÞ2S1
number of agents (particles) that create a swarm, moving around
in the search space, looking for the global best solution. In this case,
Fitness function = minimize (Fi)
each particle can communicate to each of the other particle and
Compare the current position of the
thereby a fully connected network is formulated. Of course, each
particle to the pbest position
particle is attracted to the best particle out of entire swarm. Hence,
If
this kind of approach is well suited for discrete optimization. One
Current position is better.
of the areas where the discrete optimization takes a prime role in
Set current position ? pbest
the remotely sensed images is the change detection process. Espe-
Else continue with the old pbest.
cially, in the case of detecting changes happened in a region between
Choose the particle having best fitness
time stamps T1 and T2, it is required to determine the changes hap-
value and set as gbest.
pened in the whole area in terms of discrete regions if present.
Update the state ‘sta’ and velocity ‘vel’ of
Hence, an appropriate method is needed to find such changes which
each particle as,
may be considered to be the global best with respect to that region of
interest. Since the change map formulated is in the binary format, v el ¼ v i þ a1  rand1  ðpbest  si Þ
the process is carried out using binary particle swarm optimization.
The optimization of the change map obtained using binary particle þa2  rand2  ðgbest  si Þ
swarm optimization is shown in Fig. 2. sta ¼ si þ v el
Check for stopping criteria
3.2. Binary particle swarm optimization If stopping criteria is achieved
Stop
BPSO is a variant of PSO that is designed to handle discrete opti- Else
mization problem. According to this method, each particle is Continue
assumed to be present in binary format. It differs from PSO in Until convergence
updating the particle using current velocity than by combining
velocity and current position as in PSO. Due to this, it is always
noted that the search space is based on the current velocity only.
In the designed BPSO method N0 and N1 are the number of elements
Hence position is not an issue, yet the method updates the velocity
in the set S0 and S1 respectively. S0 denotes the number of elements
towards the pbest and gbest. Considering the accelerating concept,
in the current mask representing 0 and S1 denotes the number of
the change detection is approached.
elements in the current mask representing 1. The mean of sets S1
Initially, a binary change detection mask is obtained by using
and S2 will be are represented in Eqs. (1) and (2) ,
the Adapted Sparse Constrained Clustering (ASCC) method (Lal  
and Margret Anouncia, 2017). The ASCC method would generate 1 X
meanðS0 Þ ¼  ADMði; jÞ ð1Þ
a change map with significant changes in the multi-temporal N0 8ði;jÞ2S0
images. The post processing using the BPSO is expected to increase

Please cite this article in press as: Lal, A.M., Anouncia, S.M. Modernizing the multi-temporal multispectral remotely sensed image change detection for
global maxima through binary particle swarm optimization. Journal of King Saud University – Computer and Information Sciences (2018), https://doi.
org/10.1016/j.jksuci.2018.10.010
 A.M. Lal, S.M. Anouncia / Journal of King Saud University – Computer and Information Sciences xxx (2018) xxx–xxx 5

  X
1 4. Experimental results and analysis
meanðS1 Þ ¼  ADMði; jÞ ð2Þ
N1 8ði;jÞ2S1
4.1. Datasets
where, ADM is the difference image created using Absolute differ-
ence method. The updation of state and velocity is based on the ran- The developed framework is evaluated using two different data
dom variables ‘rand1’, ‘rand2’ and learning factors ‘a1’ and ‘a2’ of sets. Data Set 1 is the Original Landsat 5 image recorded on 12
local and global information. October 1984 from Huelva, Spain and Original Landsat 8 image

(a) (b)

(c) (d)

(e) (f)
Fig. 5. Change detected via various methods with fusion for Huelva data set. (a) Change Detection using K-means. (b) Change Detection using AKM. (c) Change Detection
using FCM. (d) Change Detection using ECKM. (e) Change Detection using ASCC. (f) Change detected using EASCC.

Please cite this article in press as: Lal, A.M., Anouncia, S.M. Modernizing the multi-temporal multispectral remotely sensed image change detection for
global maxima through binary particle swarm optimization. Journal of King Saud University – Computer and Information Sciences (2018), https://doi.
org/10.1016/j.jksuci.2018.10.010
6 A.M. Lal, S.M. Anouncia / Journal of King Saud University – Computer and Information Sciences xxx (2018) xxx–xxx

recorded on 12 August 2014 from Huelva, Spain of size 512  512. 4.2. Results and discussions
Data set 2 is Original Landsat 5 image acquired on October 24,
1984 form Dead Sea and Landsat 8 image acquired on October The performance of the designed intelligent approach (EASCC)
27, 2014 from Dead Sea of size 512  512. is qualitatively and quantitatively analysed. In order to assess the
Having identified the two study areas, Two images of the study performance of the developed methodology, a comparative analy-
area 1 say Huelva, Spain acquired as Landsat 5 image recorded on sis is performed with k means, AKM, FCM, ECKM and ASCC.
12 October 1984 and Landsat 8 image recorded on 12 August 2014 The change detection maps obtained for Huelva and Dead Sea
as shown in Fig. 4 (a) and (b) are considered. Two images of the data set by k means, AKM, FCM, ECKM with sparse fusion, ASCC
study area 2 say Dead Sea acquired as Landsat 5 image acquired and EASCC are shown in Figs. 5 and 6. The visual effects obtained
on October 24, 1984 and Landsat 8 image acquired on October out of the designed intelligent methodology shows the clarity of
27, 2014 as shown in Fig. 4(c) and (d) are considered. changes in the images which is quite evident from the results that

(a) (b)

(c) (d)

(e) (f)
Fig. 6. Change detected via various methods with fusion for Dead Sea data set. (a) Change detected using k means. (b) Change detected using AKM. (c) Change detected using
FCM. (d) Change detected using ECKM. (e) Change detected using ASCC. (f) Change detected using EASCC.
Please cite this article in press as: Lal, A.M., Anouncia, S.M. Modernizing the multi-temporal multispectral remotely sensed image change detection for
global maxima through binary particle swarm optimization. Journal of King Saud University – Computer and Information Sciences (2018), https://doi.
org/10.1016/j.jksuci.2018.10.010
 A.M. Lal, S.M. Anouncia / Journal of King Saud University – Computer and Information Sciences xxx (2018) xxx–xxx 7

Error obtained for various change detection methods Fp – the number of unaltered pixels recognised as altered
pixels.
35
30 The quantities given above are evaluated by a confusion matrix
% OF ERROR

25 and various metrics can be obtained using the above derived quan-
20 tities to assess the performance of an algorithm. In this paper, the
15 following metrics are adopted:
10
5 1. Overall Error (OE)
0
OE (%) CE (%)
ERROR PARAMETERS Overall error deals with the probability that a altered pixel
incorrectly recognised as unaltered pixels.
K Means AKM FCM ECKM ASCC EASCC Fn
OE ¼ ð3Þ
Fig. 7. Percentage of error obtained using various change detection techniques for Fn þ Tp
Huelva Dataset.

2. Commission Error (CE)

Error obtained for various change detection methods Commission error deals with the probability that an unaltered
pixel incorrectly recognised as altered pixels.
25 Fp
CE ¼ ð4Þ
% OF ERROR

20 Tn þ Fp
15
10
5 3. Percentage Correct Classification (PCC)
0
OE (%) CE (%) It identifies the overall accuracy of the proposed method by
ERROR PARAMETERS means of detecting the altered pixels as altered and unaltered pix-
els as unaltered.
K Means FCM AKM ECKM ASCC EASCC
ðTp þ TnÞ
PCC ¼ ð5Þ
Fig. 8. Percentage of error obtained using various change detection techniques for ðTp þ Tn þ Fp þ FnÞ
Dead Sea Dataset.

4. Recall
the designed method outperformed than the other methods (see
Figs. 7 and 8). Recall is referred to the fraction of altered pixels recognised as
For the quantitative assessment of the proposed technique, the unaltered.
following metrics have been used for each change map with respect
Tp
to the reference map (Rosenfield and Fitzpatrick-Lins, 1986). Recall ¼ ð6Þ
Tp þ Fn
Tp – the number of altered pixels recognised correctly. From the Tables 1, 2, and Figs. 9 and 10 it’s very clear that
Tn – the number of pixels correctly recognised as unaltered. designed approach EASCC has obtained the PCC of 99.63% for Dead
Fn – the number of altered pixels incorrectly recognised as Sea Dataset and PCC of 97.65 for Huelva Dataset which is higher
unaltered pixels. than existing methods such as k means, AKM (Lal et al., 2015),

Table 1
Result of Various Parameters Used for Quantitative Comparison with existing methods for Huelva Dataset.

Method Tp (Pixels) Tn (Pixels) Fp (Pixels) Fn (Pixels) OE (%) CE (%) PCC (%) Recall
K Means 4390 214,543 41,992 2228 33.67 16.37 83.2 0.66
AKM 4590 226,843 29,592 2128 31.68 11.54 87.95 0.68
FCM 5395 231,223 25,175 1360 20.13 9.82 89.92 0.8
ECKM 6058 240,100 16,054 941 13.44 6.27 93.54 0.87
ASCC 7998 244,100 10,545 510 5.99 4.14 95.8 0.94
EASCC 8250 248,725 5805 373 4.33 2.28 97.65 0.96

Table 2
Result of Various Parameters Used for Quantitative Comparison with existing methods for Dead Sea Dataset.

Method Tp (Pixels) Tn (Pixels) Fp (Pixels) Fn (Pixels) OE (%) CE (%) PCC (%) Recall
K Means 2390 232,730 26,413 611 20.36 10.19 89.69 0.8
FCM 2791 255,204 3939 210 7 1.52 98.42 0.93
AKM 2869 255,265 3878 132 4.4 1.5 98.47 0.96
ECKM 3740 254,512 3622 270 6.73 1.4 98.52 0.93
ASCC 3689 257,095 1260 100 2.64 0.49 99.48 0.97
EASCC 3840 257,324 900 80 2.04 0.35 99.63 0.98

Please cite this article in press as: Lal, A.M., Anouncia, S.M. Modernizing the multi-temporal multispectral remotely sensed image change detection for
global maxima through binary particle swarm optimization. Journal of King Saud University – Computer and Information Sciences (2018), https://doi.
org/10.1016/j.jksuci.2018.10.010
8 A.M. Lal, S.M. Anouncia / Journal of King Saud University – Computer and Information Sciences xxx (2018) xxx–xxx

FCM, ECKM (Lal and Anouncia, 2015a) and ASCC (Lal and Margret From the overall analysis it is felt that while solving the change
Anouncia, 2017). An analysis has been carried out in terms of OE detection problem, incorporating fusion of the images with semi-
and CE. It was observed that the lesser the value of OE and CE, supervised clustering the output is improved compared to the
the better is the technique. The overall error (OE) 4.33% for data- existing ones. Further in order to attain global maxima output opti-
set1 and 2.04% for dataset 2 obtained in the designed intelligent mization is carried out to produce more accurate results for change
approach EASCC indicates that changed pixels have been almost detection. The existing techniques require either the assumption of
identified accurately with less misclassification errors. The CE distributions of clusters and are very time consuming (k means,
obtained by the designed approach is also lesser than the existing AKM, FCM, ECKM and ASCC). On the other hand the designed
techniques. The Recall measures obtained by EASCC is 0.96 and approach concentrates more on the changed areas to further pre-
0.98 which is more than k means, AKM, FCM, ECKM and ASCC dict the changes in the near future years.
(see Fig. 11).

5. Conclusion
Accuracy
95.8 97.65 In this paper, an intelligent change detection approach EASCC
89.92 93.54
100 83.2 87.95 combining ASCC and BPSO for multi-temporal remote sensing
% OF ACCURACY

80 images has been designed. The method aimed in detecting changes

in remote sensed multispectral, multi temporal images. Though
60
the results appear to be acceptable, as the followed methods tend
40 to converge at local minima the solution obtained would result in
20 local optimum. Hence, to further seek good response in terms of
0 global optimum, a binary PSO is applied. The optimization of solu-
K Means AKM FCM ECKM ASCC EASCC tion achieved using BPSO resulted in a notable improvement with
VARIOUS CHANGE DETECTION METHODS high recall and good precision rate. The increase in the precision
rate of the designed process helped in drifting the visual quality
PCC (%) of the detected change and thereby the change is highlighted
Fig. 9. Accuracy obtained using various change detection techniques for Huelva
prominently. The designed EASCC approach provided better results
Dataset. in terms of accuracy and effectiveness.
Investigations are carried out on multi-temporal and multi-
spectral datasets to authorise the effectiveness of the designed
Accuracy intelligent approach. The methodology is demonstrated and evalu-
ated using the measures recall, OE, CE and PCC. The comparison of
98.04 98.47 98.52 99.48 99.6 outcome with other existing methods showed a notable enhance-
100 89.69
ment on the designed methods. In future, more images can be used
% OF ACCURACY

80 for change detection study to further promptly figure out the

60 changes and ca also focus on the spectral characteristics of the
image and the category of scene used for the change detection
40
process.
20

0
K Means AKM FCM ECKM ASCC EASCC References
VARIOUS CHANGE DETECTION METHODS
Bottomley, B.R., 1998. Mapping rural land use & land cover change in carroll county,
Arkansas utilizing multi-temporal Landsat Thematic Mapper satellite imagery
PCC (%) Diss. University of Arkansas, Arkansas..
Braik, M., Sheta, A.F., Ayesh, A., 2007. Image enhancement using particle swarm
Fig. 10. Accuracy obtained using various change detection techniques for Dead Sea optimization. In: World Congress on Engineering, pp. 978–988. No. 1.
Dataset. Celik, T., 2010. Change detection in satellite images using a genetic algorithm
approach. IEEE Trans. Geosci. Remote Sens. Lett. 7 (2), 386–390.
Dai, Q., Liu, J., Liu, S., 2010. Remote sensing image change detection based on swarm
intelligent algorithm. In: International Conference on Multimedia Technology
Performance Evaluation (ICMT), pp. 1–3, China.
Kusetogullari, H., 2015. Dynamic multiple description wavelet based image coding
1 using enhanced particle swarm optimization. IEEE J. Sel. Top. Signal Process. 9
0.9 (1), 125–138.
Kusetogullari, H., Yavariabdi, A., Celik, T., 2015. Unsupervised change detection in
0.8
multitemporal multispectral satellite images using parallel particle swarm
0.7 optimization. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 8 (5), 2151–2164.
F1 SCORE

0.6 Lal, A.M., Anouncia, S.M., 2015a. Semi-supervised change detection approach
0.5 combining sparse fusion and constrained k means for multi-temporal remote
sensing images. The Egypt. J. Remote Sens. Space Sci. 18 (2), 279–288. https://
0.4
doi.org/10.1016/j.ejrs.2015.10.002.
0.3 Lal, A.M., Anouncia, S.M., 2015b. Detection of boundaries by fusing the topographic
0.2 sheets and multi-spectral images for geographic landscapes. Int. J. Ecol. Dev. 30
0.1 (3), 11–25.
0 Lal, A.M., Anouncia, S.M., 2016. Enhanced Dictionary based sparse representation
K Means AKM FCM ECKM ASCC EASCC fusion for multi-temporal remote sensing images. Europ. J. Remote Sens. 49 (1),
317–336. https://doi.org/10.5721/EuJRS20164918.
VARIOUS QUANTITATIVE MEASURES
Lal, Anisha M., Margret Anouncia, S., Gopalakrishnan, M., 2015. Multispectral image
change revealing using translation invariant wavelet transform fusion and
Recall for Huelva Dataset Recall for Dead Sea Dataset adaptive K-means clustering. Int. J. Appl. Environ. Sci. 10 (2), 681–693.
Lal, Anisha M., Margret Anouncia, S., 2017. Adapted sparse fusion with constrained
Fig. 11. Recall value of various change detection techniques for Huelva Dataset and clustering for semisupervised change detection in remotely sensed images. J.
Dead Sea Dataset. Appl. Remote Sens. 11, (1). https://doi.org/10.1117/1.JRS.11.016013 016013.
Please cite this article in press as: Lal, A.M., Anouncia, S.M. Modernizing the multi-temporal multispectral remotely sensed image change detection for
global maxima through binary particle swarm optimization. Journal of King Saud University – Computer and Information Sciences (2018), https://doi.
org/10.1016/j.jksuci.2018.10.010
 A.M. Lal, S.M. Anouncia / Journal of King Saud University – Computer and Information Sciences xxx (2018) xxx–xxx 9

Nielsen, A.A., 2007. The regularized iteratively reweighted MAD method for change Wu, C., Du, B., Cui, X., Zhang, L., 2017a. A post-classification change detection
detection in multi-and hyperspectral data. IEEE Trans. Image Process. 16 (2), method based on iterative slow feature analysis and Bayesian soft fusion.
463–478. Remote Sens. Environ. 199, 241–255.
Ramadan, R.M., Abdel Kader, R.F., 2009. Face recognition using particle swarm Wu, C., Zhang, L., Du, B., 2017b. Kernel slow feature analysis for scene change
optimization-based selected features. Int. J. Signal Process. Image Process. detection. IEEE Trans. Geosci. Remote Sens. 55 (4), 2367–2384.
Pattern Recogn. 2 (2), 51–65. Xian, G., Homer, C., Fry, J., 2009. Updating the 2001 national land cover database
Rosenfield, G.H., Fitzpatrick-Lins, K., 1986. A coefficient of agreement as a measure land cover classification to 2006 by using Landsat imagery change detection
of thematic classification accuracy. Photogramm. Eng. Remote Sens. 52 (2), methods. Remote Sens. Environ. 113 (6), 1133–1147.
223–227. Xue, B., Zhang, M., Browne, W.N., 2012. Multi-objective particle swarm optimisation
Sagnika, S., Bilgaiyan, S., Mishra, B.S.P., 2015. Image change detection using particle (PSO) for feature selection. In: Proc. of the ACM Annual Conference on Genetic
swarm optimization. Intell. Comput. Commun. Devices, 795–803. and Evolutionary Computation, pp. 81–88. Italy.
Singh, A., 1989. Review article digital change detection techniques using remotely- Yavariabdi, A., Kusetogullari, H., Usakli, A.B., 2017. Unsupervised satellite change
sensed data. Int. J. Remote Sens. 10 (6), 989–1003. detection using particle swarm optimisation in spherical coordinates. Int. J. Sci.
Soliman, O.S., Mahmoud, A.S., Hassan, S.M., 2012. Remote sensing satellite images Res. Inf. Syst. Eng. (IJSRISE) 1 (2).
classification using support vector machine and particle swarm optimization.
In: IEEE International Conference on Innovations in Bio-Inspired Computing and
Applications (IBICA), pp. 280–285. India.

Please cite this article in press as: Lal, A.M., Anouncia, S.M. Modernizing the multi-temporal multispectral remotely sensed image change detection for
global maxima through binary particle swarm optimization. Journal of King Saud University – Computer and Information Sciences (2018), https://doi.
org/10.1016/j.jksuci.2018.10.010