Sunteți pe pagina 1din 6

See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/304674967

Automated Detection of fresh water Blue-Green


Algae (Cyanobacteria) Using Image Processing
Techniques: A Review

Article · July 2016

CITATIONS READS

0 87

2 authors, including:

Ahmed Atia
University of Tripoli
19 PUBLICATIONS 19 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

1st Libyan Medical Sciences Conference (2017) View project

Antibiotic Dispensing practice and bacterial resistance profile in Tripoli View project

All content following this page was uploaded by Ahmed Atia on 01 July 2016.

The user has requested enhancement of the downloaded file.


International Journal of Current Medical And Applied Sciences, 2016, June, 11(1),05-09.

REVIEW ARTICLE

Automated Detection of fresh water Blue-Green Algae


(Cyanobacteria) Using Image Processing Techniques: A
Review.
Awatef saad1& Ahmed Atia2
1Lecturer, Department of computer sciences, faculty of science, Alzawia University, Libya.
2Lecturer, Department of anaesthesia and intensive care, faculty of Medical technology, Tripoli University, Libya.
--------------------------------------------------------------------------------------------------------------------------------
Abstract:
Blue-green algae (cyanobacteria) can be used as indicators to offer relatively exclusive information pertaining to
ecosystem condition. Cyanobacteria react quickly and predictably to a broad range of pollutant. Thus provides
potentially constructive early caution signals of worsening environment and the possible causes. Previous studies
revealed that most automated analyses and identification of algae images were limited to only one type of algae.
Research on automated recognition system for tropical freshwater algae is still limited. In this review, we will illustrate
the image processing and pattern recognition methods used to detect and classify oscillatoria genus from
cyanobacteria.
Keywords: Automated system - Cyanobacteria –Algae - Image processing - Neural Network.
Introduction:
Algae are a wide range of plants heterogeneous in all competent algae taxonomists. This has led numerous
shapes, sizes, and physiological functions, except that scientists to discover different systems to automated
they contain chlorophyll pigments and others. Algae analyse and classify of algae images [2,3].
include members of the nucleus of a primitive prokaryote An automated computer-based recognition and
such as blue-green algae. Some algae are microscopic, assortment system for fast identification of
whereas others reach up to several meters in length. microorganisms such as various algae will obviously
Algae are abundant in salt water, fresh and stagnant pools decrease the load of routine identifications borne by
of water, lakes, and in humid places or on the rocks. Large taxonomist whose service are required in biological
numbers of algae are also found on soil [1]. diversity studies [4]. Artificial neural network (ANN)
The presence of algae in surface water has been a long- based automated algae recognition is beneficial due to its
standing issue all over the world because of their adverse learning ability from a given dataset, and it does not
effects on the treatment process and quality of drinking require a rule base to conclude outcome. ANN has also
water. The removal and control of algae in the water been utilized extensively to identify several types of algae
treatment industry are important global issues, especially species of lake-water samples, and microorganism. Many
in tropical and semi-tropical zones. However, scientist were extracted a set of convenient features of
identification of algae presents issues in their algae images such as Fourier descriptors, geometrical
classification and the application of the organisms in features, and features characterizing of grey level
environmental experiments. Many reports revealed that distribution in a region to use it for training process of
the usual identification of algae by utilizing microscopy ANN [5,6].
images is time consuming with the overall decrease in

Address for correspondence:


Dr. Ahmed Atia,
Access this Article Online
Lecturer,
Department Of Anaesthesia and Intensive Care, Website:
Faculty of Medical Technology,
Tripoli University, Libya. www.ijcmaas.com
Email:elbadri83@yahoo.com
How to cite this article:
Awatef Saad & Ahmed Atia: Automated Detection Of fresh water Blue-Green Subject:
Algae (Cyanobacteria) Using Image Processing Techniques: A Review: Medical Sciences
International Journal of current Medical and Applied sciences; 2016,
11(1), 05-09. Quick Response Code

IJCMAAS,E-ISSN:2321-9335,P-ISSN:2321-9327. Page | 05
Chetana P., KrishnakanthM., Sudha R., Gayathri R., MuruganS.&Mahalakshmi V.

However, work on automated analysis and recognition food, whereas some are useful for nitrogen fixation.
of algae images were restricted to some species of algae. Moreover, some Oscillatoria species are known to
This is because of the difficulties in implementation of produce hepatotoxins, which damage the liver, and
an application that can inspect all types of algae division neurotoxins, which affect the nerve cells. However, no
due to the differences in algae shapes, properties, and other useful function has been found for
colours. Until recently, only a few work exist on Oscillatoriacurviceps [12].
automated identification of tropical freshwater algae Experimental results found many evidences prove that
[7]. This review will discuss the automated at least one of molecular mechanisms action consistent
identification system for tropical freshwater algae. certain known carcinogens, which led researchers to
suspect that these toxins are liver poisons and could
Blue-Green Algae (Oscillatoria species): prove significance to humans after continuous low-level
Cyanobacteria, representing some of the most ancient exposure. Given the advantages and disadvantages to
life forms on earth, are one of the most studied the environment and to living organisms, algae would
organisms worldwide [8]. Margulis et al. characterized be easier to manage if identification and classification
them as unusual prokaryotic microorganisms that can can be done in a more automated and convenient
perform oxygenic photosynthesis. They can also manner [1].
synthesize chlorophyll a. Similar to eukaryotic algae and
plants, cyanobacteria use H2O as an electron donor for MATLAB Software:
the production of oxygen. The oldest fossil findings of MATLAB is considering as an object oriented graphical
cyanobacteria can be dated back to approximately 3500 language, and has an interactive environment for
million years ago [9]. On the last decades, cyanobacteria designing interface and development algorithm to solve
blooms had given special consideration over worldwide complex engineering problems such as data analysis,
because of it is highly effects on water environments by visualization, and numerical computation. Research
increasing anthropogenic input of nitrogen and reported that using MATLAB to solve the technical
phosphorus. To date, cyanobacteria are still largely computing problems is faster than using the traditional
present in oceans, freshwaters, and soils. Most programing language such as FORTRAN, and C. MATLAB
cyanobacteria live in water as phytoplankton. included a variety of built in function and procedure
Cyanobacteria are often called “blue-green algae” and that can be used in for different applications, including;
are named after the blue pigment phycocyanin, which, control design, computational biology, measurements,
together with chlorophyll a and other pigments, is used image processing, communications, and financial
to capture light for photosynthesis [10,11]. modelling. The add-on toolboxes which are a collection
Oscillatoriasp is one of the simplest filamentous blue- of scientific purpose functions was improve MATLAB
green algae, where each thread is surrounded by environments for solving technical problem in wide
gelatinous scabbard, and do not contain different variety application tasks. MATLAB have been enhanced
vesicles. One of the initial signs resulting from the with several advantages such as features of documents
presence of moderate pollution where it appears in and sharing projects and research, and it is ability to
large numbers, leading to change the colour of the integrate the code with other programming language
water. Oscillatoria uses photosynthesis to survive and which facilitates the distribution process for MATLAB
reproduce. Each filament of Oscillatoria consists of algorithms and applications [13].
trichomes made up of rows of cells. Filaments MATLAB is included a set of Image processing tools and
commonly exhibit oscillating, sporadic flexing, or gliding function to provide engineers and scientists with a
movements under the microscope, especially near the widespread suite of reliable functions for digital image
anterior ends. The filaments are composed of disk- processing analysis. Image processing tools integrated
shaped cells that are wider than they are long. Their end with MATLAB is developed by scientific researcher and
cells are usually rounded or tapered. Oscillatoria, with professional technical to reduce the time consuming of
an average size of 7 µm, is an organism that reproduces coding and debugging for the essential image processing
by fragmentation and forms long filaments of cells that tasks. Using built in function of image processing tools is
can break into fragments called hormogonia, which can more reliable and faster than building application from
grow into new and longer filaments. Breaks in the scratch, it is illustrated a significant results in saving
filament usually occur where dead cells (necridia) are time and operation cost which enable users to spend
present. Oscillatoriais categorized under the domain of less time in design and coding algorithms. MATLAB also
bacteria because it does not possess membrane-bound help user by allowing them to learn and apply
organelles, a nucleus, or an endoplasmic reticulum. In specialized technology in varying area of image
addition, metabolically active bacteria include those that processing approaches [14].
are heterotrophic, chemoautotrophic, and
photosynthetic. Oscillatoria is a photosynthetic bacteria Artificial Neural Networks (ANNs):
grouped under the kingdom of cyanobacteria. Some Artificial Neural Network (ANN) is abbreviations for
Oscillatoria species, such as Oscillatoriaplantesis, are artificial neural networks which is the scientific field
consumed for their nutritional contribution and sold as that model information-processing structures as

IJCMAAS, E-ISSN: 2321-9335,P-ISSN:2321-9327. Page | 6


Logic Publications @ 2016, IJCMAAS, E-ISSN: 2321-9335,P-ISSN:2321-9327.

massively parallel structure similar to human brain. a representative sample of data patterns for each taxon
ANN is usually consisting of several nodes connected to be identified with related information identification
together through adjusted functions to construct the by using the training data [16].
neural networks. ANN is constructed from three ANN used to generate special mapping form between
different layers often which are the input layer, hidden input and output variables and produced a complex
layer, and output layer. ANN is considering as a arbitrarily for nonlinear decision boundaries. Mostly
powerful data modeling tool that create a neural network complexity increases with the number
representation form for complex input / output of nodes and layers of net. Neural networks usually use
relationships [15]. the sequential training process to observe data, and
ANN has become increasingly significantly in solving distinguish between patterns of interest that makes
complex problems which required special consideration them an efficient pattern recognizer [17].
of thinking and decision in many disciplines. ANN is ANN is developed to perform specific information
designed to be used for solving particularly complex computation which totally different from conventional
problems related to highly non-linear relationships. methods. Neural networks construct from nodes with
Particularly, ANN has been widely used as scientific numerous simple elements that process individually
tools in many disciplines to solve different types of several aspects of a large problem. A processing element
problems, such as identification and control, forecasting, (PE) is used to multiply the input elements by a set of
classification, and optimization because the weights, to transfer the results into nonlinear outputs
heterogeneous systems are extremely difficult to model values. Neural computation is represented a massive
in mathematic models. Flexible structure of ANN is interconnection among PEs to share the load of overall
providing simple and reasonable solutions to various processing task, and adaptive nature of the parameters
types of problems. In addition, ANNs were commonly (weights) with PEs. ANN usually has several different
employed in computerized pattern recognition tool for layers of PEs as shown in figure 1. Neural network
automated taxonomic identification based on several works to find optimal solution based on a measurement
types of features such as morphological features, shape of its performance by adapting its weights during
features, etc. ANN is simple has two phases training training process. ANN performance for a supervised
phase and decision phase, in training phase learning learning is measured in terms of error criteria, a desired
process of specific information is performed, and in signal, and iteration cycles or epoch, whereas the
decision phase the classification or recognition results is performance for unsupervised ANN is measured based
produced based on the learning phase. In training phase on learning law, and topology constraints [18].
the internal structure of ANN is adjusted in response to

Figure 1: A simple multilayer MLP neural networks. The circles are represents PEs which arranged in
different layers. The input is the left node, hidden layer are the two columns in the middle, and
outputs is represented by the right nodes in last column. The lines that connected between PEs is
represented the weighted or sometimes known as a scaling factor.

MLP Neural Networks: achieve classification tasks with a great accuracy.


MLP is one type of feed-forward ANN model which used Increasing number of hidden layers is considered a
to map the input sets of data into a set of appropriate disadvantageous because the training process will need
output or desired results. MLP designed with multiple extra times which can effects network performance. The
node layers in a directed graph where each layer fully input layer in MPL ANN are denoted the set of
connected to the next layer. Each node in MPL ANN is information or data need to be processed by multiplied
represented a neuron, or processing element with a it with the weight of hidden layers of network. This
nonlinear activation function that control the process of procedure is repeated many times until the desired
weight adjusted except input nodes. Layers of MPL ANN results obtain which considered the learning process for
are important in solution, and optimization results, the network. A few learning algorithms are exist
where one or two hidden layers are nearly sufficient to recently which available to obtain the correct values for

International Journal of Current Medical And Applied Sciences [IJCMAAS], Volume: 11, Issue: 1.
Awatef Saad & Ahmed Atia

the weights, the supervised and unsupervised learning


paradigm is the common methods employed to adjust Conclusion:
the networks weight values [18]. It has been demonstrated in this review that the
MLP is used a back propagation technique as supervised automated detection of Blue-Green Algae is both
learning for training the ANN network. MLP is substantial and reasonably practicable. Mostly, by using
considered as one of the most neural network the ANN, a high rate of identification and classification
topologies used widely for general application purpose, of cyanobacteria and algae was achieved. Algae play a
for example MLP with two hidden layers can be used as substantial part in the determination of water quality in
general purpose for statistic pattern classifier. In other freshwater lakes. Algae are found to affect the water
words, discriminant functions of MPL are used to properties and chemical compounds which may cause
cluster a set of input data which can be taken any shape conceivable hazards for human health. Many studies
[19]. have been done on cyanobacteria to explore more about
Furthermore, researcher found that MLP is usually its features, category, and effect in the environments,
achieves better performance in computing the but there are limited studies on many aspects of tropical
maximum posteriori receiver which optimize the freshwater algae. Recently, the pre-processing image
classification tasks especially when weights are technique are increased the appearance of
normalized properly including output classes which cyanobacteria significantly and the process of automatic
usually normalize into 0 or 1 values. MLP has the ability detection is currently being enhanced to detect all
also of approximating arbitrary functions, this capability common Blue-Green Algae.
is important on design nonlinear dynamics system [20].
MLPs are trained normally by using the back Acknowledgments:
propagation algorithm which depends on propagation We thank all our present and previous collaborators
the rules to optimize errors through the network by and colleagues.
automated for weight of hidden PEs. MLP neural
network mostly trains by using error correction of
learning procedure, where perceptron elements of a
References:
given layer connected with all other elements of next
layer, and the desired values must be initializing at early 1. Crayton, M.A. Toxic Cyanobacteria Blooms, Office of
Environment health Assessments Washington State.
stages. Each input data sets need to be associated with
1993, pp 1-17.
desired outputs data set which considers the heart of 2. Culverhouse PF, Williams R, Reguera B, Herry V,
pattern recognition process because the ANN is usually Gonzalez-Gil S: Do experts make mistakes? A
used to transfer the input data sets into appropriate comparison of human and machine identification of
values during training phase [21]. dinoflagellates. Marine Ecology Progress Series 2003,
MPL ANN is first start off by defining the initial 247:17-25.
estimates of the correct output, then the process of 3. Culverhouse PF, Williams R, Benfield M, Flood PR, Sell
weight randomize for the input pattern presented. The AF, Mazzocchi MG: Automatic image analysis of
main difference between the actual network output and plankton: future perspectives. Marine Ecology
Progress Series 2006, 312:297-309.
desired output is evaluated during training process [22].
4. Weeks PJ, et al: Automating the identification of
The average squared error between actual output and insects: a new solution to an old problem. Bull Entomol
target value is minimized as much as possible by Res 1997, 87:203-211.
network. This minimization process of networks is used 5. Simpson R, Culverhouse PF, Ellis R, Williams R:
a gradient descent in MLP ANN, which obtains the Classification of Ceratium and Dinophyceae plankton
common back propagation algorithm for training neural species. In Proc IntConf on Artificial Neural Networks,
networks. Gradient descent is used mostly to find the ICANN ‘94, Sorrento Marinaro M, Morasso PG 1994,
minimum values of a function by taking steps to the 843-846.
negative function of the gradient at current point [23]. 6. Embleton KV, Gibson CE, Heaney SI: Automated
counting of phytoplankton by pattern recognition: A
Back propagation method is a common method used for
comparison with a manual counting method. Journal
learning the MPL ANN. This technique of calculating the of Plankton Research 2003, 25(6):669-681.
derivatives in a computationally efficient manner has a 7. Yao Z, et al: Recognition of blue-green algae in lakes
graphical interpretation and operates by propagating using distributive genetic algorithm-based neural
error backwards from output nodes to inner nodes. networks. J of Neuro comp 2007,70:641-647.
Momentum learning is commonly applied in this 8. Schopf, J.W., Barghoorn, E.S., Maser, M.D., and Gordon,
process. This type of learning is used to prevent the R.O. 1965. Electron microscopy of fossil bacteria two
network and the system from converging to a local billion years old. Science. 149: 1365-1367.
minimum or saddle point. It also allows changes to 9. Margulis, L. : Symbiotic theory of the origin of
eukaryotic organelles; criteria for proof. Symp. Soc.
persist for a number of adjustment cycles and takes into
Exp. Biol. 1997,29: 21-38.
account the previous weight change [24]. 10. Paerl, H.W., Fulton 3rd, R.S., Moisander, P.H., Dyble, J.
Harmful freshwater algal blooms, with an emphasis on
cyanobacteria. Sci. World J. 2001, 1, 76–113.
Logic Publications @ 2016, IJCMAAS, E-ISSN: 2321-9335,P-ISSN:2321-9327. Page | 08
Logic Publications @ 2016, IJCMAAS, E-ISSN: 2321-9335,P-ISSN:2321-9327.

11. Khan, F., Ansari, A. Eutrophication: an ecological http://www.cs.sun.ac.za/~kroon/courses/machine_le


vision. Bot. Rev.2005, 71, 449– 482. arning/lecture5/mlp.pdf
12. Delpeuch F, Joseph A, Cavelier C. Ann Nutr Aliment 19. Albu RD. Investigating the Effect of Hidden Layers
(French): Consumption and nutritional contribution of Number on Web Services Response Time Prediction,
the blue algae (Oscillatoria platensis) among some Nonconventional Technologies Review, 2013, 7(1):4-9.
populations ofKanem (Tchad).1975,29(6), 497-516. 20. Chang, K.T. & Liu, J.K.: Landslide features interpreted
13. Lidierth M. sigTOOL: A MATLAB-based environment by neural network method using a high-resolution
for sharing laboratory-developed software to analyze satellite image and digital topographic data. In Geo-
biological signals. J Neurosci Methods. 2009 Mar Imagery Bridging Continents; Proc. 0f 2Qth ISPRS
30;178(1):188-96. Congress, Istanbul, 12- 23 July 2004.
14. Tahir H. H; Pareja T. F, MATLAB Pckage and Science 21. Belciug S, Gorunescu F. Error-correction learning for
Subjects for Undergraduate Studies. International artificial neural networks using the Bayesian
Journal for Cross-Disciplinary Subject in Education. paradigm. Application to automated medical
2010;1,38-42. diagnosis. J Biomed Inform. 2014. pii: S1532-
15. Gaston K. J; O'Neill M. A.; Automated Species 0464(14)00160-9.
Identification: Why Not?. The Royal Society. 2004,: 22. Akram A. Performance evaluation of artificial neural
655-667. networks for spatial data analysis. WSEAS
16. Misra B. B; Dehuri S,: Functional Link Artificial Neural Transactions on Computers 2011; 10(4):115-124.
Network for Classification Task in Data Mining. 23. Ghaffari A, Abdollahi H, Khoshayand MR, Bozchalooi
Journal of Computer Science, 2007,3(12),: 948-955. IS, Dadgar A, Rafiee-Tehrani M. Performance
17. Jayanta K.B., Debnath B., Kim T; Use of Artificial comparison of neural network training algorithms in
Neural Network in Pattern Recognition ; International modeling of bimodal drug delivery. Int J Pharm. 2006
Journal of Software Engineering and Its Dec 11;327(1-2):126-38.
Applications:.2010;4,2,23-34. 24. Chabaa S, Zeroual A, Antan J. Identification and
18. Noriega,L.(2005).Multilayer perception Tutorial. prediction of internet traffic using artificial neural
Received from networks. J. Intelligent Learning systems &
Applications, 2010, 2, 147-155.
--------------------------------------------------------------------------

Conflict of interest: None declared


No source of funding.

International Journal of Current Medical And Applied Sciences [IJCMAAS], Volume: 11, Issue: 1.

View publication stats

S-ar putea să vă placă și