https://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4688 Thu, 26 Mar 2026 05:14:12 GMT 2026-03-26T05:14:12Z https://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4706 Title: Image Processing Based Approach to Determine the Angle of Incidence for Bullet Holes, in Aid of Shooting Incident Reconstruction Authors: Ariyarathna, W. R. C. Abstract: Identification of a shooter’s location in a shooting incident is a piece of critical information which investigators need to understand for scene reconstruction. Bullet holes and their characteristics play a significant role in tracking bullet paths to their originated locations. Few methods are currently employed to estimate the angle of incidence of fired bullets by means of bullet holes. But each method has its own limitations. The accuracy level of the estimated trajectories depends on many factors. Considering the commonly reported incidents, out of many bullet types and associated target surfaces, the AK family rifles and Zinc coated 1 mm sheet metal are selected for this study. The study analyzed the special deformation features and metal surface debris spread around the outer perimeter of the bullet impact marks, targeting the determination of incident angle. It also demonstrated two inversely proportional mathematical relationships considering the area of the impact mark and the length of the lead-in crease to the incident angle of the fired AK bullet. And it revealed the possibility of using the bullet impact area to predict the approximated incident angles from 10o to 90o and the usability of lead-in crease length to predict the approximated incident angles below 40o, referring to the AK bullets impact marks identified on Zinc coated 1 mm sheet metal in actual crime scenes. Thu, 22 Jun 2023 00:00:00 GMT https://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4706 2023-06-22T00:00:00Z https://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4705 Title: Information Extraction From Scanned Invoices using Machine Learning, OCR and Spatial Feature Mapping Techniques Authors: Darsha, W.B. Abstract: Receiving invoices as scanned images is one of the biggest problems business organizations are still facing. Consuming human effort for converting scanned invoices to text documents is not sustainable because of their low performance even inherently capable of. With the recent escalations of Computer Vision technology with Machine Learning we were seeing new dimensions for addressing this bursting problem. Optical Character Reading (OCR) is the latest way of extracting text from images in general context, but the output was not much helpful for identifying key parameters from invoices. Hence we employed an object detection algorithm called You Only Looks Once (YOLO) first to capture text blobs in granular level, then streamlined them to OCR and finally processed spatial information with pattern matching techniques. Using this improved approach we could successfully extract not only key parameters like merchant information, invoice no, datetime, total but also the invoice items in the table body, and indeed with a high performance. Thus methodology we developed can be adapted to any scanned invoice dataset with proper adjustments, and also for any other document type. Thu, 22 Jun 2023 00:00:00 GMT https://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4705 2023-06-22T00:00:00Z https://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4704 Title: Information Retrieval System for Circulars Authors: Wijeratne, D. P. Abstract: Revolution of physical infrastructure, increasing population, and advancement in technology contributed to rapid increase in published information. This rapid growth of published information will give potential users problems such as how to find specific information from a large document corpus.The researcher focused on circular corpus and identified the absence of a well functioning circular retrieval system will lead to significant problems such as need to put a lot of effort and time to search for a specific circular which interested in. This dissertation will address these issues by implementing a circular storage and retrieval system that will enable users to quickly and easily retrieve the circulars what they are looking for.Along with circular management system a personalized circular recommendation system will also be built which recommends circulars based on previously viewed circulars. Circular storage and retrieval system will be implemented on top of circular document collection maintained by a government ministry. To enable users to quickly and easily retrieve circulars, the system 1) Allows users to enter tags specific to circulars when uploading circulars. 2) Allows users to narrow down the searched results by advanced filter criteria such as words to include, words to exclude, matching phrase, year of circular and tags. 3) Query match happens using an inverted index and circulars will be ranked using BM25 algorithm before passing the results to front end. Also for the ease of the user there will be a personalized circular recommendation system to recommend circulars based on previously viewed circulars and that is implemented using TF-IDF matrix and cosine-similarity matrix. Thu, 22 Jun 2023 00:00:00 GMT https://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4704 2023-06-22T00:00:00Z https://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4703 Title: Exploring Model Level Transfer Learning For Improving Sinhala Speech Recognition Authors: Nanayakkara, A.L. Abstract: Automatic Speech Recognition (ASR) is the process which accurately translate spoken utterances into its corresponding textual format. However, in ASR, it will only translate the given speech data into text and not worry on the semantic aspect on it. Through accurate ASR we can easily build an interface for the both illiterate and literate users. Anyway, ASR gives better results for the most widely used data rich languages likes, English and German, but not in the data scarcity languages like Sinhala. During past few years many researchers conducted several studies on developing more accurate ASR for Sinhala, but they failed to succeeded on it due to low resource problem. This project represents a study to build ASR system for a low resource Sinhala language, which is known as morphologically rich complex language. To tackle the data scarcity issue, we have used new mechanism called transfer learning. It is capable to transfer knowledge from data rich model to data scarce model. We carry out several experiments on Sinhala speech recognition on DeepSpeech by considering various aspects such as applications on language optimizations, external scorer and data augmentations. Initially we start our experiments on transfer learning from pre-trained English to Sinhala without considering any data augmentation and achieved 22.92% in WER and 8.84% in CER. Later, when we applied data augmentation on the transfer learned model then it showed drastically reduction on WER and CER with compared to the initial models. It showed 17.19% for the WER and 5.9% in CER for the model which consists of 10% of reverb together with 30% of overlay augmented type with others on 40% in each by considering its default values and it is explained in this document. Experiments were conducted for the Sinhala speech dataset gathered from Language Technology Research Laboratory at UCSC. It consists of 40 hours of data coverage including both male and female speakers, which were recorded with the support from Praat and RedStart tools. All the experiments conducted by using it with the external support on 4-gram language model which build on KenLM toolkit. Finally, in the user evaluation it gives fairly good results for our model. Thu, 22 Jun 2023 00:00:00 GMT https://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4703 2023-06-22T00:00:00Z