Browsing by Author "Kasthurirathna, D"

Now showing 1 - 20 of 97

Open Access
2D Pose Estimation based Child Action Recognition
(Institute of Electrical and Electronics Engineers Inc., 2022-11) Mohottala, S; Abeygunawardana, S; Samarasinghe, P; Kasthurirathna, D; Abhayaratne, C
We present a graph convolutional network with 2D pose estimation for the first time on child action recognition task achieving on par results with LRCN on a benchmark dataset containing unconstrained environment based videos.
Embargo
Absorbing Markov Chain Approach to Modelling Disruptions in Supply Chain Networks
(IEEE, 2019-08) Perera, S; Bell, M; Kurauchi, F; Kasthurirathna, D
Recent developments in the area of network science has encouraged researchers to adopt a topological perspective in modelling Supply Chain Networks (SCNs). While topological models can provide macro level insights into the properties of SCN systems, the lack of specificity due to high level of abstraction in these models limit their real-world applicability, especially in relation to assessing the impact on SCNs arising due to individual firm or supply channel level disruptions. In particular, beyond the topological structure, a more comprehensive method should also incorporate the heterogeneity of various components (i.e. firms and inter-firm links) which together form the SCN. To fill the above gap, this work proposes using the idea of absorbing Markov chains to model disruption impacts on SCNs. Since this method does not require path enumeration to identify the number of supply chains which form the SCN, it is deemed more efficient compared to the other traditional methods.
Open Access
Agro-Genius: Crop Prediction Using Machine Learning
(https://ijisrt.com/agrogenius-crop-prediction-using-machine-learning, 2019-10) Gamage, M. P. A. W; Kasthurirathna, D; Paresith, M. M; Thayakaran, S; Suganya, S; Puvipavan, P
This paper present a way to aid farmers focusing on profitable vegetable cultivation in Sri Lanka. As agriculture creates an economic future for developing countries, the demand of modern technologies in this sector is higher. Key technologies used for this problem are Deep Learning, Machine Learning and Visualization. As the product, an android mobile application is developed. In this application the users should input their location to start the prediction process. Data preprocessing is started when the location is received to the system. The collected dataset divided into 3 parts. 80 percent for training, 10 percent for testing and 10 percent for validation. After that the model is created using LSTM RNN for vegetable prediction and ARIMA for price prediction. Finally, for given location profitable crop and predicted future price of vegetables are shown in the application. Other than the prediction, optimizing for multiple crop sowing according to the user requirements and visualizing cultivation and production data on map and graphs are also given in the application. This paper elaborates the procedure of model development, model training and model testing.
Open Access
Agro-Genius: Crop Prediction Using Machine Learning
(2019-10) Gamage, A; Kasthurirathna, D
This paper present a way to aid farmers focusing on profitable vegetable cultivation in Sri Lanka. As agriculture creates an economic future for developing countries, the demand of modern technologies in this sector is higher. Key technologies used for this problem are Deep Learning, Machine Learning and Visualization. As the product, an android mobile application is developed. In this application the users should input their location to start the prediction process. Data preprocessing is started when the location is received to the system. The collected dataset divided into 3 parts. 80 percent for training, 10 percent for testing and 10 percent for validation. After that the model is created using LSTM RNN for vegetable prediction and ARIMA for price prediction. Finally, for given location profitable crop and predicted future price of vegetables are shown in the application. Other than the prediction, optimizing for multiple crop sowing according to the user requirements and visualizing cultivation and production data on map and graphs are also given in the application. This paper elaborates the procedure of model development, model training and model testing.
Embargo
AI-Driven Smart Bin for Waste Management
(IEEE, 2020-12-10) Abeygunawardhana, A. G. D. T; Shalinda, R. M. M. M; Bandara, W. H. M. D; Anesta, W. D. S; Kasthurirathna, D; Abeysiri, L
With increasing urbanization, waste has become a major problem in the present world. Therefore, proper waste management is a must for a healthy and clean environment. Though government authorities in most countries provide various solutions for waste management, solid waste tends to make a significant impact on the environment as they do not decompose easily. This research focuses on AI (Artificial Intelligence)-driven smart waste bin that can classify the most widely available solid waste materials namely Metal, Glass, and Plastic. The smart waste bin performs the separation of waste using image processing and machine learning algorithms. The system also performs the continuous monitoring of the collected waste level by using ultrasonic sensors. A dedicated mobile application will generate the optimal routes for the available waste collectors to collect the filled bins. Moreover, with this smart bin, the challenge of recognizing each waste item is overcome by using visual data as the source. Therefore, the usage of expensive sensor devices and filtration techniques to determine the category is disregarded. The smart bin can recognize the category of solid waste, collect it to the specified container, and notify the garbage level in each container. So, it is a portable waste management system.
Embargo
Architectural description based Overlay Networks
(2011-09-01) Kasthurirathna, D; Keppetiyagama, C
Overlay Networks are heavily used in Distributed computing applications. They often have heterogeneous architectures, such as Client Server, Peer to Peer or Hybrid. In this work, we try to abstract the Architecture of an Overlay Network into a document called an Architectural Description (AD). The Architectural Description document may contain the Roles and the Relationships of a particular Overlay Architecture. The Architectural Description documents may be exchanged among the nodes and parsed by the nodes themselves, enabling the nodes to adopt different roles and relationships. By introducing a new AD, a new Overlay Network can be formed dynamically. AD based Overlay Networks may open many new possibilities in Overlay Networking. This approach would allow heterogeneous Overlays to work collaboratively, while maintaining their respective Security settings using 'Security Roles'. It would also allow multiple overlays to be dynamically 'super-imposed' on top of each other. Apart from that, the AD based approach would allow the same set of nodes to switch between heterogeneous overlays at different time intervals. Architectural Descriptions can also be used as an efficient means of Security key management. A prototype framework was developed to explore these features, using sample distributed file sharing applications. Moreover, the possible enhancements and future directions of AD based approach in developing Overlay Networks are also discussed.
Embargo
Better you: Automated tool that evaluates mental health and provides guidance for university students
(Institute of Electrical and Electronics Engineers Inc., 2022-11-04) Eeswar, S. S; Samaratunga, J. S; Nivethika, G; Anjana, W.W.M.; Jayasingha, T.B.; Pandithakoralage, S; Kasthurirathna, D
This research paper proposes a system that evaluates mental health through text-based, voice-based and facial emotion recognition. After predicting the user's overall emotional state, activity suggestions and close contact interactions will be suggested to improve their mental health.
Embargo
Career Aura–Smart Resume and Employment Recommender
(IEEE, 2021-12-09) Dissanayake, K; Mendis, S; Subasinghe, R; Geethanjana, D; Lunugalage, D; Kasthurirathna, D
Recruitment and Job seeking are two major factors that are directly proportional to each other. Due to the competitive nature of the present world, the process of acquiring the best resource effectively and efficiently has become a challenging aspect for the companies. As a result, modern job portals have become increasingly popular to address the challenges identified in the early recruitment and job search process. The purpose of this research is to introduce an optimal solution to address the ineffective areas identified in the job and recruitment domain which can further enhance the recruitment and job seeking decisions by utilizing deep learning and sentiment analytic approach along with descriptive analysis. The proposed system recommends the relevant job opportunities by omitting the irrelevant job advertisements for job hunters who are interested in the IT job domain while they input their resume to the system and additionally, they can improve their career decisions by adhering to the prediction schemes. Moreover, the system facilitates recruiters to headhunt top talents efficiently once they input job requirements to the system and candidate suggestions are not only made depending on their resume information but also analyzing their LinkedIn endorsements.
Embargo
Cognitive Rehabilitation based Personalized Solution for Dementia Patients using Reinforcement Learning
(IEEE, 2021-04-15) Rathnayaka, M. H. K. R; Watawala, W. K. C. R; Manamendra, M. G; Silva, S. R. R. M; Kasthurirathna, D; Jayalath, T
Dementia is one of the most challenging health problems faced globally with the increase in the ageing population. The estimated current prevalence of dementia is 47.5 million worldwide. This number will nearly double in every 20 years globally. Dementia is basically, a syndrome which cannot be cured by medicine, but non-pharmacological therapy can be used to treat Dementia patients, this is known as Cognitive Rehabilitation Therapy. According to the recommendations of the doctors, the use of a brain training application could be better than traditional approaches. There are number of Brain training mobile applications in the world that could be useful in improving human concentration, attention and all sorts of brain activities but there isn’t any customized software solution that has games or activities. Patients can be in different stages of Dementia. Therefore, for a better cognitive rehabilitation they need personalized therapies with the games and activities. Accordingly, developing this application is an actual global requirement for dementia patients. The world is evolving with new technologies and this application includes the mind games based on such technologies as Reinforcement Learning which predict the next level for patients based on user behavior. And there are some activities on speech recognition using Deep Neural Network as well. Patients, caregivers and doctors can view the progress reports of the patients. All the games have designed along with the supervision and recommendation from a Consultant Psychiatrist in Sri Lanka. The main objective is to help the Dementia patients in cognitive rehabilitation to improve the quality of life with best suited personalized games and activities.
Open Access
Comparative Analysis of Deep Learning Models for Multi-Step Prediction of Financial Time Series
(researchgate.net, 2020-10-21) Aryal, S; Nadarajah, D; Rupasinghe, P.L; Jayawardena, C; Kasthurirathna, D
Financial time series prediction has been a key topic of interest among researchers considering the complexity of the domain and also due to its significant impact on a wide range of applications. In contrast to one-step ahead prediction, multi-step forecasting is more desirable in the industry but the task is more challenging. In recent days, advancement in deep learning has shown impressive accomplishments across various tasks including sequence learning and time series forecasting. Although most previous studies are focused on applications of deep learning models for single-step ahead prediction, multi-step financial time series forecasting has not been explored exhaustively. This paper aims at extensively evaluating the performance of various state-of-the-art deep learning models for multiple multi-steps ahead prediction horizons on real-world stock and forex markets dataset. Specifically, we focus on Long-Short Term Memory (LSTM) network and its variations, Encoder-Decoder based sequence to sequence models, Temporal Convolution Network (TCN), hybrid Exponential SmoothingRecurrent Neural Networks (ES-RNN) and Neural Basis Expansion Analysis for interpretable Time Series forecasting (N-BEATS). Experimental results show that the latest deep learning models such as NBEATS, ES-LSTM and TCN produced better results for all stock market related datasets by obtaining around 50% less Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE) scores for each prediction horizon as compared to other models. However, the conventional LSTM-based models still prove to be dominant in the forex domain by comparatively achieving around 2% less error values.
Embargo
Comparative analysis of the application of Deep Learning techniques for Forex Rate prediction
(2019-12-05) Nadarajah, D; Aryal, S; Kasthurirathna, D; Rupasinghe, L; Jayawardena, C
Forecasting the financial time series is an extensive field of study. Even though the econometric models, traditional machine learning models, artificial neural networks and deep learning models have been used to predict the financial time series, deep learning models have been recently employed to do predictions of financial time series. In this paper, three different deep learning models called Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN) and Temporal Convolution Network (TCN) have been used to predict the United States Dollar (USD) to Sri Lankan Rupees (LKR) exchange rate and compared the accuracy of the models. The results indicate the superiority of CNN model over other models. We conclude that CNN based models perform best in financial time series prediction.
Embargo
Comparative analysis of the application of Deep Learning techniques for Forex Rate prediction
(IEEE, 2019-12-05) Aryal, S; Nadarajah, D; Kasthurirathna, D; Rupasinghe, L; Jayawardena, C
Forecasting the financial time series is an extensive field of study. Even though the econometric models, traditional machine learning models, artificial neural networks and deep learning models have been used to predict the financial time series, deep learning models have been recently employed to do predictions of financial time series. In this paper, three different deep learning models called Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN) and Temporal Convolution Network (TCN) have been used to predict the United States Dollar (USD) to Sri Lankan Rupees (LKR) exchange rate and compared the accuracy of the models. The results indicate the superiority of CNN model over other models. We conclude that CNN based models perform best in financial time series prediction.
Embargo
Comparative analysis of the application of Deep Learning techniques for Forex Rate prediction
(IEEE, 2019-12-05) Aryal, S; Nadarajah, D; Kasthurirathna, D; Rupasinghe, L; Jayawardena, C
Forecasting the financial time series is an extensive field of study. Even though the econometric models, traditional machine learning models, artificial neural networks and deep learning models have been used to predict the financial time series, deep learning models have been recently employed to do predictions of financial time series. In this paper, three different deep learning models called Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN) and Temporal Convolution Network (TCN) have been used to predict the United States Dollar (USD) to Sri Lankan Rupees (LKR) exchange rate and compared the accuracy of the models. The results indicate the superiority of CNN model over other models. We conclude that CNN based models perform best in financial time series prediction.
Embargo
Computer Vision and NLP based Multimodal Ensemble Attentiveness Detection API for E-Learning
(IEEE, 2021-04-21) Wijeratne, M. D; Lakmal, R. H. G. A; Geethadhari, W. K. S; Athalage, M. A; Gamage, A; Kasthurirathna, D
Attention is the fundamental element of effective learning, memory, and interaction. Learning however, with the evolvement of technologies in the modern digital age, has surpassed traditional learning systems to more convenient online or e-learning systems. Nevertheless, unlike in the traditional learning systems, attention detection of a student in an e-learning environment remains one of the barely explored areas in Human Computer Interaction. This study proposes a multimodal ensemble solution to detect the level of attentiveness of a student in an e-learning environment, with the use of computer vision, natural language processing, and deep learning to overcome the barriers in identifying user attention in e-learning. The proposed multimodal captures, processes, and predicts user attentiveness levels of individual students, which are subsequently aggregated through an ensemble model to derive an overall outcome of better accuracy than individual model outcomes. The final outcome of the ensemble model produces a range of percentages, within which the attentiveness level of the student lies during a single online lesson. This range is consequently delivered to the users through an Application Programming Interface.
Open Access
Computer vision based indoor navigation for shopping complexes
(acm.org, 2020-12-09) Perera, G. S. T; Madhubhashini, K. W. R; Lunugalage, D; Piyathilaka, D. V. S; Lakshani, W. H. U; Kasthurirathna, D
Smartphone-based indoor navigation systems are frantically required in indoor situations. This limitation of clients is significant. Global Positioning System (GPS) isn't plausible for indoor areas as it gives exceptionally helpless outcomes for indoor restriction. In this research paper, we present a Computer Vision-Based Indoor Navigation System for shopping complexes. Computer vision is used in this system to find the exact location/current location of the user. It contains a mobile android application for positioning, navigating, and displaying the current location for showing on 2D Map. The system will detect the user's position, generate a GIS map, display the shortest path using A* search algorithm, and provides step-by-step direction to the destination using audio instruction for localization with Augmented Reality (AR) map and navigation using mobile phone sensor technologies like accelerometer, gyroscope, and magnetometer. The audio instructions include active guidance for upcoming turns in the traveling path, distance of each section between turns. This system uses a suggestion-based Chabot that uses a trained model to improve the user's experience. Thus, this research expects to build a cost-effective, efficient, and timely response system that will help the users for a smart shopping experience.
Embargo
Computer Vision Based Privacy Protected Fall Detection and Behavior Monitoring System for the Care of the Elderly
(IEEE, 2021-09-07) Fernando, Y. P. N; Gunasekara, K. D. B; Sirikumara, K. P; Galappaththi, U. E; Thilakarathna, T; Kasthurirathna, D
The elderly population constitutes a large percentage of the society hence making elderly care a top priority. Falls have been identified as a leading issue among major problems faced by them. Concerning this, many monitoring devices have been developed, most of them focusing solely on one specific health care aspect or related to fall detection, and are based on sensors and wearable devices which are usually uncomfortable for daily use. Considering these aspects, the solution proposed in this research is a real time computer vision-based system that monitors behavior and detects anomalies through deep learning. The monitoring is mainly focused on detecting unusual behavior including falls, and monitoring routine activities to detect deviations. A device approach is used to deploy the deep learning models and consists of IP camera-based monitoring which uses a special privacy protected procedure that ensures the detection is done based on meta data and therefore no camera image or footage is stored. The research is mainly focused on four major components which are user identification, fall detection, routine variance detection and device configuration.
Embargo
Computer Vision Enabled Drowning Detection System
(IEEE, 2021-12-09) Handalage, U; Nikapotha, N; Subasinghe, C; Prasanga, T; Thilakarthna, T; Kasthurirathna, D
Safety is paramount in all swimming pools. The current systems expected to address the problem of ensuring safety at swimming pools have significant problems due to their technical aspects, such as underwater cameras and methodological aspects such as the need for human intervention in the rescue mission. The use of an automated visual-based monitoring system can help to reduce drownings and assure pool safety effectively. This study introduces a revolutionary technology that identifies drowning victims in a minimum amount of time and dispatches an automated drone to save them. Using convolutional neural network (CNN) models, it can detect a drowning person in three stages. Whenever such a situation like this is detected, the inflatable tube-mounted self-driven drone will go on a rescue mission, sounding an alarm to inform the nearby lifeguards. The system also keeps an eye out for potentially dangerous actions that could result in drowning. This system’s ability to save a drowning victim in under a minute has been demonstrated in prototype experiments' performance evaluations.
Embargo
Computer Vision for Autonomous Driving
(IEEE, 2021-12-09) Kanchana, B; Peiris, R; Perera, D; Jayasinghe, D; Kasthurirathna, D
Computer vision in self-driving vehicles can lead to research and development of futuristic vehicles that can mitigate the road accidents and assist in a safer driving environment. By using the self-driving technology, the riders can be roamed to their destinations without using human interaction. But in recent times self-driving vehicle technology is still at the early stage. Mostly in the rushed areas like cities it becomes challenging to deploy such autonomous systems because even a small amount of data can cause a critical accident situation. In Order to increase the autonomous driving conditions computer vision and deep learning-based approaches are tended to be used. Finding the obstacles on the road and analyzing the current traffic flow are mainly focused areas using computer vision-based approaches. As well as many researchers using deep learning-based approaches like convolutional neural networks to enhance the autonomous driving conditions. This research paper focused on the evaluation of computer vision used in self-driving vehicles.
Embargo
Computer-Vision Enabled Waste Management System for Green Environment
(IEEE, 2021-12-09) Hewagamage, P; Mihiranga, A; Perera, D; Fernando, R; Thilakarathna, T; Kasthurirathna, D
Waste management has become a critical requirement to maintain a green environment in Sri Lanka as well as other countries. Town councils have to regularly collect different types of wastes to clean cities/towns. Hence managing the waste of the cities is a challenging task. However, most of the urban councils currently use a manual approach to managing waste. However, it results in many difficulties for the people and cleaning staff who involve in the process by following strict guidelines. Issues due to waste contamination, no proper information management of waste collection, and no punctuality in removing waste from the garbage bins are some of the significant issues arising from the manual process. Due to the drawbacks of the manual approach, social issues, environmental issues, health issues can occur easily. This paper proposes a better solution to replace this manual system with an automated system to overcome these issues. Hence, the main objective of this research is to introduce an ICT-based innovative design that can be used to develop an effective waste management system in town councils. In the proposed model, we will introduce a Computer Vision-based smart waste bin system with real-time monitoring that incorporates various technologies such as computer vision, sensor-based IoT devices, and geographical information system (GIS) related technologies. Our proposed solution consists of a waste bin system, which is capable of automated waste segregation. Our design facilitates the admin users to expand the waste bin kit by adding more waste categories in a user-friendly manner, making our product adaptive in any environment. At the same time, waste bins can notify the real-time waste status. Our system generates the optimum collection routing path and displays it in a mobile app using those real-time status details. We also demonstrate a low-cost prototype.
Open Access
Consumer Surplus based Method for Quantifying and Improving the Material Flow Supply Chain Network Robustness
(2018-06-01) Perera, S; Bell, M. G. H; Kurauchi, F; Bliemer, M. C. J; Kasthurirathna, D
Recent advances in network science has encouraged researchers to adopt a topological view when characterising the robustness of supply chain networks (SCNs). However, topology based characterisations, without considering the heterogeneity among the supply chains which form the SCN, can only provide a partial understanding of robustness. Hitherto, focus of robustness studies have been on cyclic SCNs, with unweighted and undirected links representing general inter-firm interactions. Here, we consider the specific case of a material flow SCN with multi-sourcing, which is characterised by a tiered structure with directed and weighted links. The proposed method uses the multinomial logit model to estimate the utility levels of supply chains within the SCN, as perceived by a focal firm which is indicative of the SCN consumers. The robustness of the SCN is characterised by considering the degree to which supply chains overlap with each other as a cost in the logit formulation. Finally, using a randomisation scheme to generate ensembles of SCN configurations which preserve the number of connections at each firm, the configuration which maximises the consumer surplus for the focal firm is identified. The proposed method is implemented on a real world SCN to identify the optimal configuration in terms of robustness.