This research aims to develop a state-of-the-art deep learning model to translate night-time images into day-time images, which has significant implications for improving the visibility of drivers at night. With the increasing number of vehicles on the road and the higher risk of accidents during nighttime, this research is crucial in enhancing the safety of driving, especially in low-light conditions. The proposed model utilizes the latest advances in deep learning and computer vision, including Generative Adversarial Networks (GANs) and attention mechanisms, to achieve realistic and high-quality image translations. The effectiveness of the proposed model will be evaluated on a large-scale dataset and compared with existing state-of-the-art methods. The outcome of this research will have practical implications for the development of advanced driver assistance systems (ADAS) and contribute to the advancement of computer vision and deep learning fields. The methodology of this research involves collecting a large dataset of night-time and day-time images, pre-processing and augmenting the dataset, designing, and training a novel GAN architecture with attention mechanisms for image enhancement, and evaluating the model’s performance using quantitative and qualitative metrics. The success of this research will have significant implications for the automotive industry, traffic safety, and other night-time computer vision research.

PUBLICATIONS:  

H.K.I.S. Lakmal, and M.B. Dissanayake, “Improving the Visibility at Night for Advanced Driver Assistance Systems Using Night-to-Day Image Translation,”, IEEE 6th SLAAI International Conference on Artificial Intelligence (SLAAI-ICAI-2022)

H.K.I.S. Lakmal, and M.B. Dissanayake,“Improving Visibility at Night with Cross Domain Image Translation for Advance Driver Assistance Systems,” 15th International Research Conference (IRC2022) KDU, pp. 3, 2022 

Deep Learning (DL)-based semantic segmentation has gained significant attention in various applications due to its ability to produce highly accurate segmented outputs. This research focuses on investigating deep learning techniques for segmenting water regions in aerial images. The primary objective of this study is to support public health agents in mosquito-vector control and eradication programs. By combining DL and convolutional neural networks (CNN) with image analysis, the identification of stagnant water in aerial images becomes feasible. This approach is particularly valuable for locating water pools in hard-to-reach areas inaccessible to humans. In line with the practical application scenario, the aerial image has arbitrarily shaped water-retaining areas. Hence the detection algorithm should possess the ability to identify objects and areas with different arbitrary shapes. The research aims to automate the segmentation of stagnant water areas in aerial images captured by a drone camera, employing the state-of-the-art CNN-based semantic segmentation method known as SegNet, incorporating the innovative attention unit because it is worth noting that superior segmentation results can be achieved by fine-tuning the model with additional attention units. Furthermore, various loss functions and evaluation metrics will be compared in the proposed model to identify the most effective one for water segmentation. On successful completion of the research, the proposed method is expected to automatically segment arbitrary-shaped stagnant water regions in aerial images as it can be applied to large-scale and small-scale water area segmentation tasks.

PUBLICATIONS:  

P. Mylvaganam and M. B. Dissanayake, "Deep Learning for Arbitrary-Shaped Water Pooling Region Detection on Aerial Images," Proceedings of 2022 Moratuwa Engineering Research Conference (MERCon 2022), Moratuwa, Sri Lanka, pp. 1-5, 2022

P. Mylvaganam and M. B. Dissanayake, "Detection of Mosquito Breeding areas using Semantic Segmentation," Proceedings of International Women in Eng. Symp. 2022, pp. 34-35, 2022.

The Lithium-ion battery is a popular energy storing device in power systems and electric vehicles due to its high energy density. These are consisting of several Lithium-ion cells connected in series and parallel. When one or more cells are degraded in such batteries, the charging and discharging of other healthy cells may be affected which leads to reduce the overall battery performance. Therefore, it is required to identify the degraded cells in a battery before they cause damages to the other cells. Since the degradation of a cell is shown as an increase of its internal resistance, to identify a degraded cell, it is required to estimate its internal resistance. There is no specific direct method to estimate the resistance without monitoring the voltage and current of an individual cell while in operation. The aim of this research is to develop a mathematical model to identify individual cell resistance changes by means of voltage, current and temperature variations of the battery, and thereby to detect the degraded cells. The proposed method modelled the thermal activity of a cell and expanded it into a multi-cell battery. The thermal equations are derived in the state-space and developed with an estimator for cell temperatures. Since the resistance is the key parameter of heat generation, sensing the temperature of selected cells and the current through modules, enables to detect the changes of resistances. An algorithm has been developed to quantitatively compare the estimated cell temperature with a threshold level to detect the degraded cells. The proposed model is implemented in battery simulation models. The simulation models of sixteen cells, thirty-two cells and fifty-six cells show that the detection algorithm detects the degraded cells accurately. The fifty-six-cell battery resembles a commercially available battery of an electric vehicle. Degraded cell detection system for batteries can be deployed in electric vehicles and storage devices in power systems. In addition to the existing hardware, this system requires temperature sensors to be installed on selected cells.

PUBLICATIONS:  

I. A. Premaratne, B. G. L. T. Samaranayake, and P. J. Binduhewa, “An Analytical study on Methods to Detect Degraded Cells in a Multi-Cell Lithium-ion Battery Pack,” presented at the 15th (IEEE) International Conference on Industrial and Information Systems (ICIIS) 2020, IIT Ropar, India, Nov. 2020

Performance improvement of printed log periodic dipole array (PLPDA) antennas using a model-based multi-objective optimization technique is proposed. Multiple antenna parameters will be accounted for during the optimization such that trade-offs are reduced. On successful completion of the research, the resulting antennas are expected to out-perform the conventional PLDPA in terms of size and antenna performance.

PUBLICATIONS:  

H. M. S. M. Herath, J. M. J. W. Jayasinghe and D. N. Uduwawala (2021). Performance Improvement of Printed Log Periodic Dipole Array Antennas, 2021 IEEE 16th International Conference on Industrial and Information Systems (ICIIS), Kandy, Sri Lanka, 2021, pp. 52-57

Printed yagi arrays have both the features of dipole yagi arrays as well as the inherent advantages of printed patch antennas. This study aims at improving the performance of the printed version using different element geometries. The parameters such as element spacing and substrate properties will also be optimized.

PUBLICATIONS:  

Uduwawala, D., Jayasinghe, J., & Narampanawe, N. (2019, December). Genetically Designed High Gain Sierpinski Carpet Fractal Antenna. In 2019 14th Conference on Industrial and Information Systems (ICIIS) (pp. 11-14).

Visible light communication (VLC) is a promising solution to the problems of spectrum crunch, bandwidth limitations, and security issues related with radio frequency (RF) communication. In VLC, 380nm - 780nm range of the electromagnetic spectrum which is known as visible light range is used for the communication. For that purpose, a light-emitting diode or laser is used as the transmitter while a photodiode is used as the receiver. Intensity modulation and direct detection techniques are used to achieve the communication. Modern research has shown that VLC systems can be deployed in underwater environments to achieve high performance compared to conventional RF and acoustic communications. To fill the research gaps of VLC, we design and analyse several indoor and underwater VLC systems which are capable of obtaining higher performance gains. Designed VLC systems are as follows: Parallel Relay Assisted Underwater VLC Systems, Relay-Aided Non-Orthogonal Multiple Access (NOMA) based Underwater VLC Systems, Autonomous Underwater Vehicle (AUV) Placement for Coverage Maximization of Underwater Sensor Network, Cooperative NOMA aided Underwater VLC Systems, Indoor VLC Systems.

PUBLICATIONS:  

K. W. S. Palitharathna, R. I. Godaliyadda, V. R. Herath and H. A. Suraweera, "Relay-Assisted Optical Wireless Communications in Turbid Water," in Proc. 13th ACM International Conference on Underwater Networks & Systems (WUWNet '18), Shenzhen, China, Dec. 2018, Art. no. 40, pp. 1-5,

K. W. S. Palitharathna, H. A. Suraweera , R. I. Godaliyadda, V. R. Herath and Z. Ding, "Impact of Receiver Orientation on Full-Duplex Relay Aided NOMA Underwater Optical Wireless Systems," in Proc. IEEE International Conference on Communications (ICC 2020), Dublin, Ireland, Jun. 2020, Art. No. 1, pp. 1-7

K. W. S. Palitharathna, H. A. Suraweera , R. I. Godaliyadda, V. R. Herath and J. S. Thompson, "Multi-AUV Placement for Coverage Maximization in Underwater Optical Wireless Sensor Networks," in Proc. IEEE Global OCEANS 2020, Marine Bay, Singapore, Oct. 2020, Art. No. 1. pp. 1-8.

K. W. S. Palitharathna, H. A. Suraweera, R. I. Godaliyadda, and V. R. Herath, "Rate maximization for ightwave power transfer-enabled cooperative half/full-duplex UOWC systems," in Proc. IEEE International Workshop on Signal Processing Advances in Wireless Communications (SPAWC 2021), Lucca, Italy, Sept. 2021

To handle random behavior of RES (Renewable Energy Sources), future Power Systems focused on being rich with RES need flexibility in operations more than it is needed now. Flexibility is defined as the ability of a system to deploy its resources to respond to changes in net load, where net load is defined as the remaining system load not served by RES generation. This research is focused to quantify the flexibility of a power system in terms of a risk metric, which is formulated by accounting the uncertainties in Net Load forecast and generation availability. Furthermore, a flexibility constrained unit commitment (UC) problem which utilizes the risk metric is formulated to enhance the flexibility of the next day operation schedule. A Probabilistic uncertainty range is defined for the net load based on a desired risk level. With the application of flexibility constrained UC, a flexible operation is achieved by making sure that the flexibility limits of the scheduled operation of the flexible units in the system is adequate enough to match the defined uncertainty range.

PUBLICATIONS:  

Herath, A., Manaz, M. M., Liyanage, K. M., Masuta, T., Lu, C. N., & Nishio, K. (2022). Frequency Excursion Likelihood Constrained Resource Scheduling for Large-Scale Renewable Energy Integration. IEEE Access, 10, 90563-90575.

Herath, A., Liyanage, K. M., Manaz, M. M., Masuta, T., & Lu, C. N. (2021). Flexibility Metric for Power System Planning Under Forecast Uncertainty. In 2021 IEEE 16th International Conference on Industrial and Information Systems (ICIIS) (pp. 116-121).

Nishio, K., Yoshida, K., Masuta, T., Herath, A. A., Liyanage, K. M., & Kulasekara, H. (2020). Input Data Extrapolation of Generator Unit Commitment and Optimal Load Dispatch for AGC 30 model. In 2020 International Conference on Smart Grids and Energy Systems (SGES) (pp. 498-503).