Proceedings of the 5th Annual Sessions,  Sri Lanka Association for Artificial Intelligence

31st October 2008, Colombo



Using Neural Networks for Recognition of Handwritten Mathematical Documents

Nilupa Liyanage , Asoka S. Karunananda

Abstract-Advancements in modern technologies cannot still override the importance of preparation of handwritten documentations. In particular, handwritten documentations are inevitable in mathematical calculations, mathematical tutorials, preparation of marking schemes and financial reports. This paper presents our approach to the design and implementation of Artificial Neural Network solutions for recognition of handwritten mathematical documents and producing text files. The system consists of three modules for image processing, character recognition and text formation. The Image processing module of the system has been designed to perform thresholding, normalization, segmentation and feature extraction of the handwritten numeric characters. The Image processing module captures the features of handwritten characters and to produce quality inputs for the ANN module. The Artificial Neural Network module for character recognition has been designed with a three layer architecture to use back propagation training algorithm. Image processing has been done through MATLAB while NeuroSolution toolkit has been used for the development of ANN and formation of textual output.



Web-based English to Sinhala Selected Texts Translation system

B. Hettige, Asoka. S. Karunananda

Abstract-This paper presents English to Sinhala Machine Translation system that can translate selected English text into Sinhala through the web. This Translation system contains two modules, namely; web-based machine translation system and java based user interface. Core of the translation system runs on a web server and can be accessed by user interface. The core of the translation system contains seven modules, namely, English Morphological analyzer, English Parser, Translator, Sinhala Morphological generator, Sinhala parser, Transliteration module and three Lexicon Databases. Java based user interface provides a mechanism for on-demand translation of selected texts from an English document. This enables users to get translated a selected set of English sentences while reading a document.



Advising Farmers on Crop Selecting Using an Expert System

M. M. A. Premachandra and Uditha Ratnayake

Abstract-Agriculture and plantation is an important and interesting research area everywhere in the world and Sri Lanka is no exception. Nowadays available land area for a plantation is becoming scarce. This scarce resource is frequently wasted through our bad practices and improper management. Cultivation is a more economical but complex process. Selecting and maintaining suitable crops for the maximum profit involves a sequence of tasks. These tasks and the whole process need a lot of expert knowledge and experience. But unfortunately, people having this type of knowledge are very limited. Their assistance is not available when the person who is going to cultivate needs it. We propose a knowledge-based approach to land evaluation for the selection of suitable agricultural crops – Crop Advisor.“Crop Advisor” is a Knowledge-based Decision Support System (KBSS) for crop selection. The expert system is powered primarily by human knowledge collected from crop experts. It also considers economic feasibility of raising a crop by taking market price, cost of production, access to market and yield levels. The “Crop Advisor” expert system then suggests with consultation with farmer (through a graphicalu1ser interface) a suitable agricultural crop that can be grown in a land unit with reasoning.

PDF (234KB)


Using Agent technology to access the Semantic Web

Sanjaya Ratnayake, Ruvindee Rupasinghe, Anuruddha Ranatunga, Shalinda Adikari ,Sajayahan de Zoysa, Kamala Tennakoon, Asoka S. Karunananda

Abstract-Semantic web consists of heterogeneous sources of knowledge including texts, graphics, blogs, animations, audios, and videos. However, there have been limited researches conducted to present semantic web information in the form suitable to individuals. This paper reports on the design and implementation of ‘Divon’, a swarm of agents that emulate a user profile driven approach to present semantic web information in the forms suitable to individuals. Divon has been structured with four agents, namely, Message Agent, Query Handler Agent, Presentation Agent, and User Profiler Agent. The Message Agent plays a key role in guiding the search process and displaying the information in a suitable form for the user whereas User Profiler Agent creates individual user profiles according to individual preferences. Divon has been developed on JADE environment and can run on any computer in connection with an arbitrary search engine.