Some Interesting Related Neural Network Research from the University of Sri Lanka

Jay Jayanka, A Computer-Based System for Identifying Sri Lankan Sinhala Ayurvedic Herbal Plants,
BSc (Special) in Computer Science (2016).
Supervisor: TGI Fernando
Abstract: In this research, we developed a computer system based on deep learning for identifying medicinal plant leaves. Convolutional neural networks with RGB and grayscale images, as well as multilayer neural networks with RGB images, were used for identifying Ayurvedic plant leaves. In this work, we used a convolutional neural network with RGB images that achieved 97.71% accuracy for identifying Sinhala medicinal plant leaves.

HTM Perera, A Deep Learning Approach for Sinhala Handwritten Character Recognition,
BSc (Special) in Computer Science (2015).
Supervisor: TGI Fernando
Abstract: Handwritten character recognition can be considered one of the major subfields of computer vision and machine intelligence. With the recent development of deep-learning-based computer vision methods, most complex image recognition tasks have become easier and more accurate. Deep-learning methods perform better than other classical methods used for recognizing handwritten characters. Sinhala handwritten characters have considerable variation due to the unique shapes of Sinhala characters. So far, only classical image processing methods and basic machine learning methods have been used in some research on Sinhala handwritten character recognition. The main purpose of this research is to identify and implement an efficient deep-learning-based method for Sinhala handwritten character recognition. In this research, we were able to achieve top-1 and top-5 error rates of 2.74% and 0.07% respectively with a convolutional neural network for Sinhala handwritten character recognition.

RMEJ Rathnayaka, Discovering a Tool to Find the Astrological Effects of a Name,
BSc (Special) in Computer Science (2014).
Supervisor: TGI Fernando
Abstract: Sri Lankans, especially Sinhala Buddhists and Hindus, follow customs and rituals introduced by astrology. These rituals begin with the birth of a newborn and continue until the time of death. The first of these rituals is naming a baby. Currently, many parents want to name their children with astrologically auspicious names. Although they spend money on this, there is no reliable way to determine whether a given name from an astrologer conforms to astrological rules. In common practice, the names they are given do not meet astrological rules. Naming a baby is related not only to astrology but also to language. In this research, the Sinhala language and its concepts were taken into account. Many concepts in astrology have been absorbed into the Sinhala language. Sinhala people have practiced astrology for centuries, and Buddhist priests were the pioneers who preserved astrology and language in ancient Sri Lanka. This may be the reason for the close relationship between the Sinhala language and astrology. Astrology has concepts that classify words according to their pronunciation patterns. Using these astrological concepts, the effect can be given when a word is pronounced or placed as the first word of a poem. In this research, astrological and linguistic concepts were studied to develop a web-based application that implements concepts for predicting the effects of a given name. The introduced and studied algorithms and concepts may offer a new way of thinking about word pronunciation; in particular, the concept of වර්ණ (warn) in the Sinhala language may help find a more reliable grapheme-to-phoneme conversion.

KS Ilmini, Identifying Human Characteristics Using Neural Networks (Physiognomy Recognition)
BSc (Special) in Computer Science (2013).
Supervisor: TGI Fernando
Abstract: The research background of this paper is the development of a human personality recognition system using machine learning techniques. Identifying personality traits from face images includes three independent algorithms: an artificial neural network (ANN) with a backpropagation learning algorithm, support vector machines (SVM), and deep learning. The facial region in the image is identified by a color segmentation algorithm. The extracted image is then fed into the personality recognition process. Facial features are manually identified in ANN and SVM. The main research area of the project is developing a multi-class recognition system using artificial neural networks to identify personality traits from extracted face images.