Research

Theses/Dissertations:

  1. Ph.D. – Ant Colony Optimization Based Simulation of 3D Automatic Hose/Pipe Routing, Department of Electronic and Computer Engineering, College of Engineering, Design and Physical Sciences, Brunel University, UK (2009)
    Abstract: This thesis focuses on applying one of the rapidly growing non-deterministic optimization algorithms, the ant colony algorithm, for simulating automatic hose/pipe routing with several conflicting objectives. Within the thesis, methods have been developed and applied to single objective hose routing, multi-objective hose routing and multi-hose routing. The use of simulation and optimization in engineering design has been widely applied in all fields of engineering as the computational capabilities of computers has increased and improved. As a result of this, the application of non-deterministic optimization techniques such as genetic algorithms, simulated annealing algorithms, ant colony algorithms, etc. has increased dramatically resulting in vast improvements in the design process. Initially, two versions of ant colony algorithms have been developed based on, respectively, a random network and a grid network for a single objective (minimizing the length of the hoses) and avoiding obstacles in the CAD model. While applying ant colony algorithms for the simulation of hose routing, two modifications have been proposed for reducing the size of the search space and avoiding the stagnation problem. Hose routing problems often consist of several conflicting or trade-off objectives. In classical approaches, in many cases, multiple objectives are aggregated into one single objective function and optimization is then treated as a single-objective optimization problem. In this thesis two versions of ant colony algorithms are presented for multi-hose routing with two conflicting objectives: minimizing the total length of the hoses and maximizing the total shared length (bundle length). In this case the two objectives are aggregated into a single objective. The current state-of-the-art approach for handling multi-objective design problems is to employ the concept of Pareto optimality. Within this thesis a new Pareto-based general purpose ant colony algorithm (PSACO) is proposed and applied to a multi-objective hose routing problem that consists of the following objectives: total length of the hoses between the start and the end locations, number of bends, and angles of bends. The proposed method is capable of handling any number of objectives and uses a single pheromone matrix for all the objectives. The domination concept is used for updating the pheromone matrix. Among the currently available multi-objective ant colony optimization (MOACO) algorithms, P-ACO generates very good solutions in the central part of the Pareto front and hence the proposed algorithm is compared with P-ACO. A new term is added to the random proportional rule of both of the algorithms (PSACO and P-ACO) to attract ants towards edges that make angles close to the prespecified angles of bends. A refinement algorithm is also suggested for searching an acceptable solution after the completion of searching the entire search space. For all of the simulations, the STL format (tessellated format) for the obstacles is used in the algorithm instead of the original shapes of the obstacles. This STL format is passed to the C++ library RAPID for collision detection. As a result of using this format, the algorithms can handle free-form obstacles and the algorithms are not restricted to a particular software package.
  2. M.Sc. (Computer Science) – Designing Sinhala Characters using Bézier Curves and B-Spline Curves, Department of Computer Science & Information Management, Asian Institute of Technology (AIT), Bangkok, Thailand (2002)
    Abstract: In character design in the computer industry, the mathematical models like Bézier curves and B-Spline curves play a very important role. Earlier characters were stored in the bit maps, and the designer had to keep separate file for each size of each font. But after Adobe introduced the PostScript fonts into the market, the fonts were represented by Bézier curves and those fonts can be scalable to any size. Further, this needs only one or two files for storing the entire font type. Existing methods of representing characters and theoretical background about Bézier curves and B-Spline curves are reviewed in this study. Further, we have developed a Visual Basic program called “Fontica” for designing characters by using Bézier and B-Spline curves. With the help of this new program, we could design 10 Sinhala (my mother language) characters. In addition, we have proposed a new technique for approximating the degree reduction of Bézier curves. This approximation appears to perform satisfactorily.
  3. M.Sc. (Industrial Mathematics) – Improved Newton’s Method for Solving Nonlinear Equations, Department of Mathematics, University of Sri Jayewardenepura, Sri Lanka (1998)
    Abstract: An iterative scheme is introduced improving Newton’s method which is widely used for solving nonlinear equations. The method is developed for both functions of one variable and two variables. Proposed scheme replaces the rectangular approximation of the indefinite integral involved in Newton’s Method by a trapezium. It is shown that the order of convergence of the new method is at least three for functions of one variable. Computational results overwhelmingly support this theory and the computational order of convergence is even more than three for certain functions. Algorithms constructed were implemented by using the high-level computer language Turbo Pascal (Ver. 7)PDF: https://tgifernando.files.wordpress.com/2015/12/inm.pdf

On-going Undergraduate Research Projects:

  1. Piriyankan Kirupaharan, A mobile application to identify fish species using Convolutional Neural Networks, B.Sc. (Special) Degree in Computer Science (2017).
    Supervisor(s): TGI Fernando

On-going Postgraduate Research Projects:

  1. MKA Ariyaratne, Designing and Developing Nature-Inspired Algorithms to solve nonlinear equations,
    Ph.D. Degree in Computer Science (2013-present).
    Supervisor(s): TGI Fernando and Sunethra Weerakoon
  2. WPJ Pemarathne, Electrical cable optimization system (guiding path) for single story building using ant colony optimization,
    M.Phil. Degree in Computer Science (2015-present).
    Supervisor(s): TGI Fernando
  3. WMKS Ilmini, Computational Personality Assessment using Machine Learning Algorithms, M.Phil. Degree in Computer Science (2017-present).
    Supervisor(s): TGI Fernando

Research Projects Supervised

Undergraduate Research Projects Supervised

  1. PM Jayanka, A computer based system to identify the Sinhala ayurvedic herbal plant leaves in Sri Lanka,
    B.Sc. (Special) Degree in Computer Science (2016).
    Supervisor(s): TGI Fernando
    Abstract: In this research, we have developed a computer system to recognize Sinhala ayurvedic plant leaves based on deep learning. Convolutional neural networks with RGB and grayscaled images and multilayer neural network with RGB images were used to identify the ayurvedic plant leaves. In this work, we were able to achieve an accuracy of 97.71% for recognizing Sinhala Ayurvedic plant leaves using a Convolutional neural network with RGB images.
  2. WDNS Wijesuriya, Predicting successfulness of radio advertisements using Recurrent Neural Networks,
    B.Sc. (Special) Degree in Computer Science (2016).
    Supervisor(s): TGI Fernando
    Abstract: Advertisers spend a lot of money on creating advertisements which may be successful or may fail. Successful advertisements will increase the profit of the advertiser whereas failed advertisements will be a heavy loss. Therefore, it is required to find a method of classifying an advertisement before putting an advertisement on the air. This research proposes a solution using Recurrent Neural Networks to classify advertisements. In the research collection of 100 advertisements were used in order to train and validate the models. Altogether there were 4 different models implemented in the research. Each model was trained six times with a different number of epochs and batch sizes in order to optimize the final output. Deep Neural network with two LSTM layers was selected as the best network to be used for classification. Therefore, by this research, we have managed to prove that RNN can be used for classifying radio advertisements successfully.
  3. TC Matharage, Training Convolutional Neural Networks Using the Fire y Algorithm,
    B.Sc. (Special) Degree in Computer Science (2016).
    Supervisor(s): TGI Fernando
    Abstract:
    Nowadays in all fields, Neural networks are very popular in solving practical problems. Convolutional Neural Network (CNN) is a kind of Deep Neural Network that has wide applications in fields such as image recognition, hand written character recognition, speech recognition and natural language processing, etc. When considering solving these problems, the most important thing is that, training the neural network as quickly as possible with an optimal solution. Although backpropagation algorithm is the most common and traditional method of training the neural networks, there are some limitations with this method. Therefore, researchers tended to find new algorithms to train neural networks. As a result of that many research studies have been conducted to observe the usability of nature inspired algorithms in neural network training. Nature inspired algorithms are created by mimicking the patterns and processes in nature. In this research, a nature inspired algorithm called Firefly algorithm is implemented to test as training algorithm for Convolutional Neural Networks. To train the implemented algorithm a benchmark dataset, MNIST is used and results are compared to select the best training algorithm for Convolutional Neural Network training using negative log likelihood error and CPU time.
  4. SL Heenatigala, Search for an efficient algorithm without the derivatives to numerically solve nonlinear equations, B.Sc. (Special) Degree in Mathematics (2016).
    Supervisor(s): S Weerakoon and TGI Fernando
    Abstract: The research was mainly conducted to explore the possibility of formulating an efficient algorithm to find roots of nonlinear equations without using the derivative of the function. We know that the Newton’s method also carries the derivative part in the formula. But the Secant method can be used to overcome this difficulty. In Secant method, it uses forward difference method to replace the derivative part. But the order of convergence of the Secant method is only 1.68. Which is lower than the Newton’s method and the Improved Newton’s method. Comparing Newton’s method over Improved Newton’s method, Improved Newton’s method is better than Newton’s method because of its higher order of convergence. Therefore the Improved Newton’s method was used in this project to find a new method without the derivative. We call our method Finite Difference Improved Newtons Method (FDINM). The FDINM we have derived has given 2.4 or more as the computational order of convergence for all examples we have tested. This higher order of convergence was retained even when the FDINM was implemented on nonlinear equations with complex roots and also on systems of non-linear equations. So the computational order of convergence of the FDINM is not only higher than secant method but it exceeds the computational order of convergence of the most popular Newton’s method as well. We followed the procedure in Broyden’s method to solve systems of nonlinear equations using the FDINM due to the involvement of the Jacobian. For almost all of the test functions, the FDINM returned a computer order of convergence higher than that of the Newton’s method, the secant method and also of the Broyden’s method. The computational order of convergence of the FNINM is in fact closed to 2.5
  5. UR Weeratne, Recurrent neural network based Approach for Sinhala speech recognition, B.Sc. (Special) Degree in Computer Science (2015).
    Supervisor(s): TGI Fernando
    Abstract: Speech is the most powerful mode of communication among human beings. Speech recognition involves the conversion of acoustic signals into text format which is readable. It has been an active research area over the past years, due to its applicability in different fields. Most of the researches are based on few popular languages like English, Mandarin and French. The number of approaches tested on Sinhala speech is limited. Therefore, the research has been conducted to test the applicability of one of the latest research approaches on Sinhala speech recognition. Different models of selected approach have been used to implement a speech recognizer with the highest rate of recognition accuracy. Recurrent Neural Networks, which is proven to be an effective neural network architecture when processing temporal data, has been selected as the core concept of building the recognizer. `Keras’ – a Theano based Deep Learning Library developed to facilitate neural network based experiments, is used as the main library in implementing the speech recognizer. Mel Frequency Cepstral Coefficients (MFCC) derived from Python speech features library are used as the feature vectors. Training data have been collected from both male and female speakers. Considerably large database with noiseless discrete speech utterances is created with the support of many speakers. The performance evaluation carried out on the accuracy of the recognizer shows around 98% accuracy over speaker independent scope. Since this is a primary approach on applying neural networks on speech recognition, it has been tested on noiseless speech samples. But, future work will be carried out related to continuous speech recognition in both noisy and noiseless environments.
  6. ARAS Weerathunga, Convolutional neural network based Approach for Sinhala speech recognition,
    B.Sc. (Special) Degree in Computer Science (2015).
    Supervisor(s): TGI Fernando
    Abstract: Speech is one of the most powerful ways of communicating between humans and they are finding ways to use the speech for communicating with machines as well. This has been a major research area for more than 5 decades and there are state of art speech recognition systems available today as a result of that. But the main limitation is that there are a lot of differences between languages of all around the world. Most of the state of art Automatic Speech Recognition (ASR) systems available today is based on the English language. The number of studies has been carried out based on the Sinhala language is very limited. Therefore this study was carried out based on Sinhala speech recognition. Although there are lots of traditional approaches available for speech recognition we have decided to carry out this study based on Convolutional Neural Network (CNN) approach which is a novel area of speech recognition. Two open source deep learning libraries which are Keras and NVidia Digits were used for the implementation of speech recognition classifier based on CNN. Since CNN have shown state of art result on image classifications, we used spectrograms of speech data as the inputs for the CNN architecture. This study was carried out using Sinhala digits dataset which is a discrete speech dataset. The dataset was created by getting speech samples from 90 different speakers of different ages. The performance was measured based on the accuracy rates of the CNN classifier for speech samples and the results were surprisingly well. We were able to achieve 96% accuracy rate for this implementation under noiseless environment.
  7. HTM Perera, Deep learning approach for Sinhala hand-written character recognition,
    B.Sc. (Special) Degree in Computer Science (2015).
    Supervisor(s): TGI Fernando
    Abstract:
    Handwritten character recognition can be considered as one of the main sub fields in computer vision and machine intelligence. With the recent development of deep learning based computer vision methods, most of the complex image recognition became much easier and accurate. Some deep learning methods have performed better in other classical approaches that have been used for recognition of handwritten characters. Sinhala handwritten characters have considerable variations due to the unique shapes involved with Sinhala characters. Up to now, some studies have been carried out for recognition of Sinhala handwritten character recognition using only the classical image processing methods and basic machine learning methods. The main objective of this study was to identify and implement an efficient Sinhala handwritten character recognition method based on deep learning. In this study, we were able to obtain top 1 and top 5 error rate of 2.74% and 0.07% respectively using convolutional neural networks for Sinhala handwritten character recognition.
  8. K Nanayakkara, Independent EEG enabled affective human computer interface,
    B.Sc. (Special) Degree in Computer Science (2015).
    Supervisor(s): TGI Fernando
    Abstract: In recent years, human emotion detection using Electroencephalogram (EEG) started playing a crucial role in developing a smarter Brain-Computer Interface (BCI). In this research, we are using DEAP physiological emotional database to identify emotions using the arousal valance model. Using wavelet transformation the EEG signal was decomposed into four frequency bands (theta, alpha beta and gamma). The Daubechies order 4 wavelet function (db4) was utilized to do the processing. From these frequency bands, linear and nonlinear statistical features were extracted to be used in the classification.The classification is performed using an artificial neural network (ANN). Different network architectures and features were used in the experiment. The document finally presents the evaluations and results obtained in the classification stage.
  9. HN Gunasinghe, System for choosing the right eyeglasses based on face shape using deep learning algorithms,
    B.Sc. (Special) Degree in Computer Science (2014).
    Supervisor(s): TGI Fernando
    Abstract: Identifying one’s face shape is important for various purposes including choosing the matching hairstyle, haircut and eyewear. Face shapes are distinguished by the formation and composition of lines, edges and curves at the outline of one’s face. There are seven basic shapes to be compared in order to detect the accurate shape. The identification process of face shapes is highly technical and time-consuming. The naked eye observation by experts is the main approach adopted for detection and identification of face shapes. A computer aided system can provide clues and technological support to confirm the prediction of the expert. The work described in this thesis is based on an attempt to implement a system to identify face shapes by classifying the images of human frontal faces. The underlying concept of building the system was designed by taking the advantage of the Image Processing techniques and Machine learning. Image processing was used to obtain a new and accurate dataset. Four types of machine learning architectures were created including single layer neural network, support vector machine, multilayer perceptron and convolutional neural network. Each network was trained and obtained the accuracy of classification for seven face shapes. The convolutional neural network with the best architecture could identify the face shape with an overall accuracy of 67%. The experimental results indicate that the proposed approach is a valuable approach, which can significantly support an accurate detection of face shapes with a little computational effort.
  10. CD Athuraliya, Deep Learning Approach for Logo Recognition in Images of Complex Environments,
    B.Sc. (Special) Degree in Computer Science (2014).
    Supervisor(s): TGI Fernando
    Abstract: Logo recognition can be considered as a sub-field of object recognition which is found useful in many day-to-day applications such as enterprise identification and product recognition. Despite the fact that there are many success stories in handling object recognition using tools which are solely based on computer vision, utilizing machine learning on these problems has resulted in better outcomes. Deep learning is a sub-field of machine learning which attempts to model high-level abstractions in data by using architectures, composed of multiple non-linear transformations. In this study, we are looking for a more efficient approach for logo recognition using deep architectures which will replace a considerable amount of image processing tasks from recognition phase with novel approaches. The varying conditions that logos can appear and the ways photographs of them can be taken render generic object detection methods, such as SIFT, useless. This problem was taken into consideration by comparing and trying to improve current methods or as a more direct application of machine learning, developing a working logo recognition system. This study aims to conduct a comprehensive analysis on research domain and to implement a usable logo recognition system utilizing deep learning approaches. Logo recognition mainly involves two phases; namely object localization and object classification in a given image. It has been identified that a special type of neural networks, convolutional neural networks are more efficient in computer vision problems. In this study we are trying to implement a network architecture to extract low-level features from input images and gradually recognize more abstract features for complete logo recognition. The solution will be developed and implemented by applying programming language tools along with deep learning libraries. Finally, the performance and efficiency of the solution will be evaluated with regard to the objectives of the problem.
  11. RMEJ Rathnayaka, Development of a Tool to find the Astrological Effects of a Name,
    B.Sc. (Special) Computer Science Special Degree (2014).
    Supervisor(s): TGI Fernando
    Abstract:
     Sri Lankans, especially Sinhala Buddhists and Hindus follow customs and rituals introduced by astrology. These rituals start with the birth of a new born and end with the time of his death. First of these rituals is naming a baby. At present many parents want to name their children with astrologically beneficial names. Although they spend money on this matter and there is no reliable way to find out whether a given name by an astrologer is according to the rules of astrology. A common situation is that names given by them are not according to the rules of astrology. Naming a baby is not only connected with astrology but also with the language. In this research, Sinhala Language and its concepts are taken into consideration. Many concepts in astrology have been absorbed by the Sinhala language. Sinhalese have practised astrology for centuries and Buddhist priests were the pioneers of preserving both astrology and language in ancient Sri Lanka. This might be the reason for the close relationship between the Sinhala language and the astrology. Astrology has some concepts of categorizing words using the pronunciation pattern of them. Using these concepts of astrology, the effect can be given when a word is pronounced or placed as the first word of a poem. In this research, both astrological concepts and linguistic concepts have been studied to develop a web-based application to implement the concepts used in predicting the effect of a given name. The algorithms and concepts introduced and studied through the research may give new ways of thinking about the pronunciation of a word, especially the concept of වර්ණ (warna) in the Sinhala language may help to find more reliable grapheme to phoneme conversions.
  12. KS Ilmini, Recognize Person’s Characteristics using Articial Neural Networks,
    B.Sc. (Special) Computer Science Special Degree (2013).
    Supervisor(s): TGI Fernando
    Abstract: The context of this work is the development of persons’ personality recognition system using machine learning techniques. Identifying personality traits from a face image includes three separate algorithms; they are Artificial Neural Networks (ANN) with backpropagation learning algorithm, Support Vector Machine (SVM) and Deep Learning. Face area in an image is identified by a colour segmentation algorithm. Then that extracted image is input to personality recognition process. Features of the face are identified manually in ANN and SVM. The main research area of this project is the development of a Multi-class recognition system using an artificial neural network which is used to recognize personality traits using extracted face image.
  13. TYT Totagamuwa, A Comparative Study on weight optimization of neural networks using Nature-inspired Algorithms with Cuckoo search Algorithm,
    B.Sc. (Special) Degree in Computer Science (2013).
    Supervisor(s): TGI Fernando
    Abstract: Nowadays neural networks are very popular in solving practical problems, especially in solving optimization problems. When solving these problems, training the neural network quickly with an optimal solution is very important. Common and traditional methods to train the neural networks are using the backpropagation algorithm. But due to the limitations in the backpropagation people tend to find new algorithms for training neural networks. As a result of that many research studies have been conducted to observe the usability of nature inspired algorithms in neural network training. Nature inspired algorithms are created by mimicking the patterns and processes in nature. Nature has unlimited patterns and processes, therefore still new nature inspired algorithms are being developed. To choose a better algorithm for neural network training, it has to check the performance of nature inspired algorithms in neural network training context. In this project two common nature inspired algorithms; Genetic algorithm, Particle Swarm Optimization and two newly implemented nature inspired algorithms; Cuckoo Search algorithm, Firefly algorithm are implemented to test as training algorithms for neural networks. To test the implemented algorithms two benchmark problems, Iris flower classification and Wisconsin breast cancer classification were used and results are compared to select the best training algorithm for neural network training using mean square error and CPU time. According to the results, it has been found that the Cuckoo search algorithm has better performances among those four algorithms.
  14. TMTM Perera, Multi-agent system for Doctor, Patient, Hospital and Pharmacy,
    B.Sc. (Special) Degree in Computer Science (2013).
    Supervisor(s): TGI Fernando and B Hettige
    Abstract: Computer science caters solutions for different types of problems relevant to the medical industry. For example, online channelling applications are a result of an attempt at solving real life, day to day problems. These channelling applications are available worldwide. They have great value and high popularity since they save time and money – two very important facts in our busy lives. Buying medicine is a basic necessity. Every person has gone through this experience at least once in their life. People prefer to buy medicine at a pharmacy if it has the cheapest medicine but finding a pharmacy with the lowest priced medicine cannot be done by guessing. The simple fact is that the buyer may have to visit several pharmacies to find the cheapest medicine and this simply is not practical for busy lives. Making appointments and channelling doctors online by using basic knowledge like symptoms, is convenient for people. And searching online for best pharmacies with low priced medicines can be a safeguard against inadvertently wasting money on high priced pharmaceuticals. So an application which contains these two features is another valuable product of computer science. This application uses knowledge of expert systems and multi agent systems to achieve the goal of producing a system that is capable of covering the above two features within one application.
  15. MKA Ariyaratne, A Comparative Study on Nature-Inspired Algorithms with Firefly Algorithm,
    B.Sc. (Special) Degree in Computer Science (2012).
    Supervisor(s): TGI Fernando
    Abstract: The processes of optimization can be defined simply as an attempt of making something better or finding the best solution for a maximization or minimization problem. The basic two approaches of optimization are classical and natural where in some problems classical approach works better and for some other problems, natural methods are good. Natural optimizing techniques, which are extracted from the behaviour of the natural world, are known as Nature Inspired optimization techniques. Ant colonies which mimic the natural food finding behaviour of ants, particle swarm optimizations algorithms which take the advantage of schooling behaviour of fish or flocking behaviour of birds are some examples for them. In this research, the main purpose is to measure the optimizing performance of such nature inspired algorithms with one new nature inspired algorithm known as Firefly inspired algorithm, which came to the stage, extracting the flashing behaviour of fireflies. Two major areas of nature inspired algorithms are evolutionary strategies and swarm intelligence. An evolutionary algorithm (EA) is a generic population-based metaheuristic optimization algorithm. An EA uses tools motivated by biological evolution: reproduction, mutation, recombination, and selection. Swarm intelligence is another problem-solving behaviour, inspired by the nature that emerges from the interaction of individual agents (e.g., bacteria, ants, termites, bees, spiders, fish, and birds) which communicate with other agents by acting on their local environments. For this research, genetic algorithms are taken as an algorithm from evolutionary strategies and Ant colonies, particle swarm optimization from swarm intelligence to make the comparison with the firefly inspired algorithm, which is also an algorithm that belongs to swarm intelligence. Traveling salesman problem, which is a representative of NP hard problems, was taken as the benchmark problem to employ all these algorithms. Each algorithm was used to solve four TSP instances with 16, 29, 51 and 100 cities that were taken from the TSPLIB library and statistics were taken appropriately. Another 5 instances of 29 cities were generated randomly and results were calculated for all four algorithms. The results of the study were manipulated using MATLAB 2008. For all 9 TSP instances, firefly algorithm gave the best results and sometimes ant colony systems too. Particle swarm optimization algorithm always scores the third place and Genetic algorithm performs last. With the results obtained, it can be clearly said that the firefly algorithm is remarkably successful and better than other three algorithms in its discrete version solving TSP.
  16. Janaki Wanigasooriya, Multi-Vehicle Passenger Allocation and Route Optimization for Employee Transportation using Genetic Algorithms,
    B.Sc. (Special) Degree in Computer Science (2011).
    Supervisor(s): TGI Fernando
    Abstract: Design of optimization of real world problems are quite complicated and optimizing vehicle routing is most important to the today’s world. Vehicle routing problems are combinatorial and NP hard. The research discusses the employee transportation optimization which uses split deliveries when the employee demand of a city greater than the vehicle capacity where vehicle capacities may be homogeneous or heterogeneous. The problem is purely multi-objective and the objectives conceded in the problem are minimizing travel time, minimizing total distance, and minimizing no of vehicles which are most concerned by companies and the employees. We successfully engaged with the popular meta-heuristic method, Genetic algorithms in the research work and we addressed the geography of the problem by designing and developing two new initialization methods and by proposing two algorithms for the employee transportation problem. The first algorithm uses the dominance relation between individual routing solutions and the second approach uses scalar weight mechanism. The algorithms implemented with C#.Net and the developed graphical user interface allows to tune the genetic parameters and also to take the routing decisions to the decision maker. The proposed algorithms for the employee transport optimization run efficiently and give invaluable support to the decision maker for taking right routing decisions.
  17. KAG Udeshani, Lung Cancer Detection System Using Neural Networks and Image Processing Techniques,
    B.Sc. (Special) Degree in Computer Science (2010).
    Supervisor(s): TGI Fernando
    Abstract: Lung cancer is tumours arising from cells lining the airways of the respiratory system. Chest X-rays are used for lung cancer detection in early stages. This system uses digital images of chest X-rays as the inputs to the system. It categorizes a given suspicious area into two categories: nodule or non-nodule. Two approaches have been used as the Methodology of this system. Those are the Neural Networks and the Image Processing Techniques. Two methods have been used to train the Neural Network: the first one is a feature based method and the other one is a pixel based method. The concept of Connected Component Analysis has been introduced and using that technique the roundness of lung nodules of a chest X-ray are identified. According to the roundness of those nodules, we can classify suspicious areas into those two categories. As the database, images of 154 lung nodules (100 malignant cases, 54 benign cases), and 93 non-nodules were collected from the Digital Image Database developed by the Japanese Society of Radiological Technology (JSRT). To consider about Sri Lankan patients, digital images of chest X-rays were collected from the National Cancer Institute, Maharagama. This dissertation discusses this research with the background study. And the system has been implemented using MATLAB R2008a. This system is a flexible and practical system that can be used to develop and add more features later on.
  18. PCP Peiris, Review of the Citations of the Research Paper: A Variant of Newton’s Method with Accelerated Third-Order Convergence,
    B.Sc. (Special) Degree in Mathematics (2010).
    Supervisor(s): Sunethra Weerakoon and TGI Fernando
    Abstract: This report is a review of the citations of the research papers relevant to “A variant of Newton’s method with accelerated third-order convergence” published by S. Weerakoon and T.G. I. Fernando in the Elsevier journal of “Applied Mathematics Letters” in 2000. Within less than a decade it records over 200 citations. The majority of researchers refer this article to gain information regarding newly developed 3rd order iterative method for solving nonlinear equations while few researchers introduced modifications to the original method. Most subsequent researchers have made use of the technique in proving the third order convergence to prove their order of convergence. Almost all who came up with a new method of same or higher order used the set of sample functions used to test the Improved Newton’s Method. Some need INM as an efficient method for solving nonlinear equations arising in their particular research. Due to some modifications, sometimes the original method transforms to a higher order method. Thus the original method shows a way to develop other third order methods or higher order methods for solving nonlinear equations. Therefore the original method plays an important role in the branch of numerical methods. The paper entitled “New Variants of Newton’s Method for Nonlinear Unconstrained Optimization Problems” comments this and this paper saying that it indicates a way of solving nonlinear unconstrained optimization problems. In this report, we have highlighted some improvements done to the original method (i.e. INM). It may be useful for consequent researchers who interest in the development of new methods for solving nonlinear equations. In addition to the above objectives of this research, this is a collection of recently used iterative methods for solving nonlinear equations. It will helpful for the future researchers, who interest in this field.
  19. DVS Hettiarachchi, Mobile Navigation System,
    B.Sc. (Special) Degree in Computer Science (2009).
    Supervisor(s): TGI Fernando
  20. MDL Salgado, Degree Reduction of Bezier Curves,
    B.Sc. (Special) Degree in Mathematics (2009).
    Supervisor(s): Sunethra Weerakoon and TGI Fernando

Postgraduate Research Projects Supervised

  1. IDID Ariyasingha, Random Weight-based Ant Colony Optimization Algorithm for Multi-objective Optimization Problems,
    M.Phil. Degree in Computer Science (2011-2017).
    Supervisor(s): TGI Fernando
    Abstract: 
    Most real world optimisation problems are concerned with multiple objectives and they are very difficult to optimise simultaneously. Researchers have proposed several ant colony optimisation algorithms for solving multiple objective problems over the last few decades. Therefore, this thesis concentrates on analysing, improving and developing ant colony optimisation algorithms for solving multiple objectives.Initially, the thesis reviews the recently proposed ant colony optimisation algorithms which have been introduced for optimising multiple objectives simultaneously. Then, it analyses the performances of these multi-objective ant colony optimisation (MO-ACO) algorithms when applied to some combinatorial optimisation problems. Therefore, at the beginning of the research, performances of the MOACO algorithms are analysed by applying them to several benchmark instances of the travelling salesman problem (TSP). It considers various objectives, such as two, three and four objectives, and changes the number of ants and number of iterations for understanding their effects on the performances of MOACO algorithms. The results of the detailed analysis have shown that some MOACO algorithms achieve better performances. Also, their performance slightly depends on the number of ants, the number of objectives and the number of iterations used in the colony.Single objective optimisation problems are considered in most of the algorithms when solving the job shop scheduling problem in the literature. However, some ant colony optimisation algorithms for solving the job shop scheduling problem with multiple objectives have been proposed in recent years, because real world applications are concerned with multiple objectives. Hence, this thesis analyses the performance of some recent multi-objective ant colony optimisation algorithms by applying them to sixteen benchmark problem instances of the job shop scheduling problem on up to 20 jobs * 5 machines. Also, it considers two, three, and four objectives and it optimises four criteria – makespan, mean flow time, mean tardiness, and mean machine idle time – simultaneously. Furthermore, different numbers of ants are used in a colony, to see their effects on the performance of the algorithms. The results obtained have shown that the performance of some multi-objective ant colony optimisation algorithms depends on the number of objectives and the number of ants.This thesis proposes a new ant colony optimisation (ACO) algorithm named the random weight-based ant colony optimisation algorithm (RWACO) to solve multiple objectives simultaneously. The RWACO algorithm is based on the ant colony system (ACS) algorithm and uses randomly generated weights for each objective associated with heuristic information which is called the random weight-based method. The performance of the newly proposed algorithm, RWACO, is evaluated by comparing it with more recent MOACO algorithms when applied to the three combinatorial optimisation problems: the travelling salesman problem (TSP), the job shop scheduling problem (JSSP) and the quadratic assignment problem (QAP). According to the results obtained with these studies, it has been shown that the RWACO algorithm achieves better performances than all the other MOACO algorithms considered in these studies.The random weight-based method, which has been introduced for the RWACO algorithm, is applied to the Pareto strength ant colony optimisation algorithm (PSACO) to examine its effects on the performance of the PSACO algorithm. The performance is evaluated by applying it to the travelling salesman problem. The experimental results have shown that the PSACO algorithm performs better when the random weight-based method is applied.
  2. AJP Samarawickrama, A Recurrent Neural Network Approach in Predicting Daily Stock Prices: An Application to the Sri Lankan Stock Market, M.Sc. Degree in Computer Science (2017).
    Supervisor(s): TGI Fernando
    Abstract:
    Many studies have been carried out to predict stock prices using different Artificial Neural Network (ANN) models during past time in different countries. Recurrent Neural Networks (RNNs) is a sub-field of neural networks that use feedback connections. Several types of RNN models have been used by researchers in predicting financial time series. This study was conducted in order to develop models to predict daily stock prices of selected listed companies of Colombo Stock Exchange (CSE) based on Recurrent Neural Network (RNN) Approach and to measure the accuracy of the models developed and identify the shortcomings of the models if present. Feed Forward, Simple Recurrent Neural Network, Gated Recurrent Unit and Long Short Term Memory architectures were used to build models. Each network has 6 input neurons and 1 output neuron. Closing, High and Low prices of past two days were selected as input variables for each company. Stock prices from 2002/01/01 to 2013/06/30 selected for the study. Most recent data was selected for the test data set, next recent set was selected for validation and the other set was selected for training. Keras package was used as the software to build and train neural networks. When considering both iterative errors and forecasting errors feed forward networks produce the highest and lowest errors. The forecasting accuracy of the best feed forward networks is approximately 99%. SRNN and LSTM networks generally produce lower errors compared with feed forward networks but in some occasions, the error is higher than feed forward networks. Compared to other two networks, GRU networks are having comparatively higher forecasting errors.
  3. DAA Deepal, Convolutional Neural Network Approach for the Detection of Lung Cancer in Chest X-Ray Images,
    M.Sc. Degree in Computer Science (2016).
    Supervisor(s): TGI Fernando
    Abstract:
    The chest X-rays are considered to be the most widely used technique within the health industry for the detection of lung cancer. Nevertheless, it is very difficult to identify lung nodules using raw chest X-ray images and analysis of such medical images has become a very complicated and tedious task. This study mainly concerned on Convolutional Neural Network Approach to identify whether the suspicious area is a nodule or non-nodule. The JSRT digital image of the chest X-ray database developed by the Japanese Society of Radiological Technology (JSRT) and was used to train and test these models. Further, the support vector machine and Multilayer perceptron were used for comparison with Convolutional Neural Network model. “Pylearn2” research library is used to build the Convolutional Neural Network model and Multilayer perceptron model. “Scikit-learn” research library is used to build the Support vector machine model. “MATLAB R2013a” is used to extract nodule and non-nodule locations from the original images and other Image processing parts.
  4. HPS Nishani, Improved Newton’s Method to Solve Systems of Nonlinear Equations, M.Sc. Degree in Industrial Mathematics (2011).
    Supervisor(s): S Weerakoon, TGI Fernando and M Liyanage
    Abstract: Improved Newton’s Method (INM) is a widely accepted third order iterative method introduced in the late 90’s to solve nonlinear equations. It has become so popular among numerical analysts that it records more than 500 citations in recognized international journals. However, even after more than a decade of the initial introduction of INM, nobody took the challenge of extending the INM for systems of nonlinear equations.The objective of this research was to prove the third order convergence of the Improved Newton’s Method in solving systems of nonlinear equations. We were able to extend the Improved Newton’s Method to functions of several variables and provide a rigorous proof for the third order convergence. This theory was supported by computational results using several systems of nonlinear equations. Computational algorithms were implemented using MATLAB.
  5. RP Abeysooriya, Genetic Optimization of Cut Order Planning in Apparel Manufacturing,
    M.Sc. Degree in Industrial Mathematics (2011).
    Supervisor(s): TGI Fernando
    Abstract: 
    Cutting is one of the main value-adding processes of the apparel manufacturing process. It serves as the major input provider for the sewing process by feeding cut panels required to sew garments. Fabric cutting process acts as a second major cost contributor to the manufacturing process due to the high expenditure on the marker making fabric spreading and cutting, which is about 5-10% of the total manufacturing cost. Owing to this reason, apparel manufacturers heavily focused on reducing the cost incurred in the cutting department. The work plan of the cutting department is termed as Cut order plan, which plans the entire cutting process; number of markers needed, sizes to be included in each marker, quantity of garments from each size need to be included in the marker and the number of fabric plies that will be cut from each marker. Researchers highlight that an effective cut order plan results in reducing the above-mentioned cost factors of the cutting process, thereby reducing the entire manufacturing cost to a greater extent. This study aims at optimizing the cut order planning process of garment manufacturing process. Genetic Algorithm (GA) principles are adopted in achieving this goal. A comprehensive literature study was carried out to understanding the cutting process and the theoretical background of cut order planning. Furthermore, GA principles and their application possibility of cut order planning are studied in detail. Next, the optimization algorithm was developed based on GA principles and the computer-based program was developed to execute the algorithm, using MATLAB. With the aim of validating the developed algorithm, cut order plan data was collected from several apparel manufacturing companies, and then the algorithm was validated using them. Same data was input to the software tools available for generating cut order plans, with the aim of comparing their results with the results obtained by the developed program. Finally, the output was finalized by doing the necessary changes based on the comparison.
  6. CP Amarajeewa, Optical Character Recognition (OCR) of Sinhala Characters by Feature Analysis Followed by Matrix Matching,
    M.Sc. Degree in Industrial Mathematics (2002).
    Supervisor(s): GK Watugala, TGI Fernando and HKGdeZ Amarasekara
  7. PAS Pathiraja, Modelling Sinhala Characters using Cubic Rational Bezier Curves,
    M.Sc. Degree in Industrial Mathematics (2002).
    Supervisor(s): GK Watugala, TGI Fernando and HKGdeZ Amarasekara

Advertisements
%d bloggers like this: