3rd International Conference on Advancements in Computing [ICAC] 2021

Permanent URI for this collectionhttps://rda.sliit.lk/handle/123456789/947

Browse

Filters

20212

Advanced Search

Filter by

Settings

Subject: search.filters.subject.Deep Learning (DL)

Search Results

Now showing 1 - 2 of 2
  • Thumbnail Image
    PublicationEmbargo
    Early Warning for Pre and Post Flood Risk Management by Using IoT and Machine Learning
    (2021 3rd International Conference on Advancements in Computing (ICAC), SLIIT, 2021-12-09) Ilukkumbure, S.P.M.K.W.; Samarasiri, V.Y.; Mohamed, M.F.; Selvaratnam, V.; Rajapaksha, U.U.S.
    Flooding has been a very treacherous situation in Sri Lanka. Therefore, developing a structure to forecast risky weather conditions will be a great aid for citizens who are affected from flood d isasters. I n t his s tudy, t he a uthors explore the use of Machine Learning (ML), Deep Learning (DL), Internet of Things (IoT), and crowdsourcing to provide insights into the development of the pre and post flood r isk management system as a solution to manage and mitigate potential flood risks. Machine learning and deep learning algorithms are used to predict upcoming flooding s ituations and r ainfall occurrences by using predicted weather information and historical data set of flood a nd r ainfall. Crowdsourcing i s u sed a s a n ovel method for identifying flood t hreatening a reas. Weather i nformation is gathered from citizens and it will help to build a procedure to notify the public and authorities of imminent flood risks. The IoT device tracks the real-time meteorological conditions and monitors continuously. The overall outcome showcases that machine learning models, deep learning algorithms, IoT and crowdsourcing information are equally contributing to predict and forecast risky weather conditions. The integration of the above components with machine learning techniques, together with the availability of historical data set, can forecast flood occurrences and disastrous weather conditions with above 0.70 accuracy in specific areas of Sri Lanka.
  • Thumbnail Image
    PublicationEmbargo
    Deep Transfer Learning Approach for Facial and Verbal Disease Detection
    (2021 3rd International Conference on Advancements in Computing (ICAC), SLIIT, 2021-12-09) Manage, D.M.; Alahakoon, A.M.I.S.; Weerathunga, K.; Weeratunga, T.; Lunugalage, D.; De Silva, H.
    Millions of people have been subjected to different kind of acute diseases, some of them are eye diseases, facial skin diseases, tongue diseases and voice abnormalities. Most of eye diseases cause fully or partial blindness. Skin and tongue complications can be signs of cancers. Voice abnormalities can be cured at initial stages. Well-practiced medical practitioners have the ability of diagnose these diseases, but due to the pandemic situations and high consultation costs people do not tend to consult doctors. This research is predominantly focused on development of an application for automatic detection of eye, skin, tongue and verbal diseases using transfer learning (TL) based deep learning (DL) approach. Deep learning is a part of machine learning (ML) which has been used in most computer vision approaches. Transfer learning has been used to rebuild the existing convolutional neural network (CNN) models and used in disease detection. DenseNet121, MobileNetV2, RestNet152V2, models have been used to detect eye, skin and tongue diseases respectively and a new model has been used to detect voice abnormalities. CNN models are capable of automatically extracting features from the given images and voice data. All the trained models have been given accuracy rate of 80%-95%.