Engineeringhttps://ir.kdu.ac.lk/handle/345/83542026-04-08T13:46:50Z2026-04-08T13:46:50ZIntegration of Image Processing with Underwater Rover to Monitor Coral Growth.Adhipaththu, WAMPLAbeysinghe, AMMHBRanasinghe, BYSameera, UVHThilakarathne, BLShttps://ir.kdu.ac.lk/handle/345/88802025-08-29T12:06:54Z2024-09-29T00:00:00ZIntegration of Image Processing with Underwater Rover to Monitor Coral Growth.
Adhipaththu, WAMPL; Abeysinghe, AMMHB; Ranasinghe, BY; Sameera, UVH; Thilakarathne, BLS
This research outlines the integration of image
processing with an Aquabot, an underwater rover to
monitor the growth of corals. Designed system composed
of underwater vehicle, a floating station for communication
and coral monitoring unit. This monitoring unit can
classify the coral type and also to identify available
diseases on the coral. Communication occurs through a
combination of wired and cloud methods. The coral
diseasedetection unit was evaluated on a data set of 121
images containing 489 instances of coral disease and the
coral varieties detection model was evaluated on a dataset
of 124 images containing 492 instances of coral varieties.
These images are taken from corals reefs and coral
nurseries around Sri Lanka. In both cases the models
identified the diseases and varieties correctly with overall
confidence levels of 33.1% and 59.6% respectively. The two
models achieved box precision of 0.346 and 0.679, the box
recallsof 0.331 and 0.558, the mean average precision
(mAP50) of 0.295 and
0.596 and the mean average precision at IoUthresholds of
0.5 to 0.95 (mAP50-95) of 0.188 and 0.425 respectively.
The results of the evaluations show that both models are
effective for respective tasks. This research introduces a
combination of AquaBot and an image processing system,
which is capable of real time monitoring and identification
of corals and its diseases.
2024-09-29T00:00:00ZDesigning a Portable Solar Power Station for Outdoor ActivitiesEranda, HASRathnayake, RAITRodrigo, PMPKarunarathne, KGNSKarunadasa, JPhttps://ir.kdu.ac.lk/handle/345/88782025-08-29T11:59:55Z2024-09-29T00:00:00ZDesigning a Portable Solar Power Station for Outdoor Activities
Eranda, HAS; Rathnayake, RAIT; Rodrigo, PMP; Karunarathne, KGNS; Karunadasa, JP
Portable solar power stations with compact
solar panels and lithium-ion batteries are useful for
outdoor activities like camping and hiking for clean and
renewable energy. This paper presents the development of
a backpack style portable solar power station with weight
less than 6 kg, that delivers 500 W for 30 minutes after one
charge. It uses a 100 Ah, 3.7 V, lithium-ion battery with
foldable solar panels and delivers sinusoidal 230 V, 50 Hz,
AC output through standard socket outlets. Multiple
outputs are provided to enable several devices to take
power simultaneously, for example charging multiple
devices. Solar panel output is fed to a push-pull converter
that produces constant 400 V DC voltage for an H-bridge
inverter which is operated with sinusoidal PWM to produce
sinusoidal, constant voltage, and constant frequency AC
output. The overall design is compact, sustainable, and
adaptable to various weather conditions, to provide a
green option for outdoor enthusiasts for a better outdoor
experience
2024-09-29T00:00:00ZEnhancing Electrical Grid Reliability through Predictive Cycle Detection with Graph Neural NetworksWIJEKOON, WMKGVBHETTIARACHCHI, HPPPhttps://ir.kdu.ac.lk/handle/345/88772025-08-29T11:33:12Z2024-09-29T00:00:00ZEnhancing Electrical Grid Reliability through Predictive Cycle Detection with Graph Neural Networks
WIJEKOON, WMKGVB; HETTIARACHCHI, HPPP
This paper represents end-end study focused on
improving the grid reliability with the application of Graph
Neural Networks (GNNs). Graph Representation of the
electrical grid which yields the model of nodes of
substations and transformers interconnection of power
lines constructed by the data from the National Grid
Electricity System Operator (ESO) Data Portal. Based on
their connections, node feature updating and encoding by
predict grid reliability with multi-layered Graph Attention
Network (GAT) was employed. In predicting failure
regions, proposed model with rigorously trained and tested
state shows higher accuracy compared to existing methods.
Results of the model signifies the model capability to
efficiently manage large-scale data with actionable insight
generation for specific usecases such as predictive
maintanance, which ensure the resilience of modern power
systems and integrating renewable energy in the modern
power system.
2024-09-29T00:00:00ZEdge Computing using FPGA with the Deployment of Neural Networks for General Purpose ApplicationPerera, KeviniHettihewa, ChamodWickramasinghe, ManupaSandanayake, AshanRajapaksha, ChamaliPathirana, Pubuduhttps://ir.kdu.ac.lk/handle/345/88762025-08-29T11:22:23Z2024-09-29T00:00:00ZEdge Computing using FPGA with the Deployment of Neural Networks for General Purpose Application
Perera, Kevini; Hettihewa, Chamod; Wickramasinghe, Manupa; Sandanayake, Ashan; Rajapaksha, Chamali; Pathirana, Pubudu
Artificial intelligence and deep learning are
gaining traction in edge computing to extract insights from
Internet of Things (IoT) devices. Hardware accelerators
like Field Programmable Gate Arrays (FPGAs) accelerate
deep learning efficiently due to their energy efficiency,
parallelism, flexibility, and reconfigurability. However,
resource constraints of FPGAs pose deployment
challenges. This research explores hardware-accelerated
applications’ dynamic deployment on the Kria KV260
platform with a Xilinx Kria K26 system-on-module,
equipped with a Zynq multiprocessor system-on-chip. It
presents an innovative solution to dynamically reconfigure
deep neural networks by running multiple neural networks
and Deep Processing Units concurrently. This research
advances Edge Computing using FPGAs to facilitate
efficient deployment of Neural Networks in resource
constrained edge environments
2024-09-29T00:00:00Z