RICIT http://ir.lib.ruh.ac.lk/handle/iruor/16744 2026-09-26T03:30:08Z 2026-09-26T03:30:08Z Wickramarathne, K.U. Gamage, C.Y. http://ir.lib.ruh.ac.lk/handle/iruor/21884 2026-09-25T09:52:21Z 2024-11-01T00:00:00Z

Identifying Sewing Machine Types According to the Stitches Using Machine Learning. Wickramarathne, K.U.; Gamage, C.Y. Identifying sewing machine types based on stitches plays a crucial role in ensuring the quality of garment production. For buying companies, to order garment products, it‟s a requirement to check the quality of the product. Deciding the specific quality of the product manually is a difficult task. Therefore, in the textile industry, there is a necessity for an automated system that can easily identify and verify whether the sewing machine type is correct or not, so the orders can be accepted or rejected easily. The model proposed in this paper helps to detect the type of sewing machine based on the stitches on clothing. This research is significant because it improves the quality control of garment production and ensures that buying companies can verify that orders meet the required standards. With the proposed model not only large-scale garments but small-scale garments can also identify the sewing machine types and solve their day-to-day problems effectively. The main objectives of the research are to create a data set of stitch images, segment the stitch patterns from images, extract features from the segmented images, and train the model to identify the sewing machine types. In this proposed methodology, the Keras Sequential Model has been employed to explore the automated identification of various sewing machine types through their unique stitch patterns. The research methodology involved enhancing image quality by reducing noise, selecting appropriate tools for system modeling, training the model with machine learning algorithms, and comparing the results with manual checks. The model achieved an accuracy of 82.80%. Identifying the defects in fabric during the manufacturing process and automatically categorizing or matching textile colors to improve dyeing and manufacturing processes are significant impacts that can be achieved in the future using the Keras Sequential Model.

2024-11-01T00:00:00Z chiranjitha, Gagana Wijerathna, Piyumi http://ir.lib.ruh.ac.lk/handle/iruor/21881 2026-09-25T09:45:16Z 2024-11-01T00:00:00Z

Real-Time Feedback System for Cricket Shot Defect Detection for Left- Handed Players Using Deep Learning Model. chiranjitha, Gagana; Wijerathna, Piyumi Cricket is a popular global sport that has many disciplines within it like batting, balling, and fielding. The batsman plays a prominent role and over the history game has produced many great batters. This research is focused on the innovative real-time cricket shot classification and feedback system for left-handed batsmen. The suggested methodology aims to address the gap in professional training availability and cost-effective tools for technical improvement for non-professional players. It mainly focuses on three basic cricket strokes: front foot defense, back foot defense, and cover drive, distinguishing between correct and incorrect executions. The model embeds into a feedback system that can accurately classify the correct or incorrect posture of these three shots while giving immediate feedback if the player is playing a wrong shot by calculating the angles between key points using algorithms. This method assists beginners by improving their skills on these 3 shots. The research involved creating a completely new data set of 58 lefthanded players across different age groups. This completely new data set consists of 1,960 videos, 490 per shot category, with correct and incorrect shots captured from multiple angles from the offside of the batsman. A 2D convolutional neural network (CNN) was developed for shot classification with the Nadam optimizer. The model demonstrates exceptional performance by achieving 97% accuracy on the test data set. Specifically, the system exhibited low error rates with a 9.7% overall false positive rate and 1.04% overall false negative rate. Further to that backfoot defense classification resulted in a 0.00% false negative rate, a 2.45% false negative for cover drive shot, and a 0.99% defensive false negative rate was demonstrated. Future work on this would involve extending the system to increase the number of cricket shots analyzed and enhancing the dataset to make the classification broader and more accurate.

2024-11-01T00:00:00Z Bandara, A.M.H.S.G. Wickrama Arachchi, R.S. http://ir.lib.ruh.ac.lk/handle/iruor/21879 2026-09-25T09:41:02Z 2024-11-01T00:00:00Z

Identification of Banana Type According to Banana Tree Using Machine Learning. Bandara, A.M.H.S.G.; Wickrama Arachchi, R.S. Bananas play a significant role in agriculture, especially in Sri Lanka, where approximately 29 varieties are cultivated. Proper identification of the types of bananas is essential for maintaining the best crop management practices, including fertilization, irrigation, and control of disease aspects. Identification of banana type by non-experts, done manually, may result in inefficiencies and errors. This research is intended to automate the identification of major banana species, such as Red Banana, Sour Banana, Cavendish, and Ash Plantain, with the help of machine learning coupled with image analysis. For this research, images were collected from multiple banana plantations across different regions, focusing on key identifying features such as petiole nature, color of the petiole, branching style, color of the midrib, and leaf blade base. Including the above features, two image data sets as branching style and leaf upper side per tree were collected. This data has been split into 80% for training and 20% for testing to ensure accurate model evaluation. The labeled data were used in training different machine learning models like VGG16, InceptionV3, VGG19, and a CNN custom model. The CNN custom model performed better among the selected techniques and gave a classification accuracy of 90%. The findings suggest that the custom CNN model can classify the banana types into four categories, which is a great approach to automating the identification process. Nonetheless, there are limitations, particularly the need for a bigger dataset size to cater to different conditions. These limitations could be mitigated in future studies by adding more banana types and a variety of data in the training of the models. From the findings of this research, it is obvious that farmers will be able to enhance crop management by employing this technique to boost agricultural productivity.

2024-11-01T00:00:00Z Vihangani, E. W. V Dilhan, A.W.A.T. http://ir.lib.ruh.ac.lk/handle/iruor/21877 2026-09-25T09:35:53Z 2024-11-01T00:00:00Z

Media Ethics Violation Detection in Sinhala News Headlines Using Natural Language Processing Techniques. Vihangani, E. W. V; Dilhan, A.W.A.T. The Sri Lankan media industry has undergone significant transformations in recent years. These changes are often driven by technological advancements and shifting audience preferences. As a result, upholding media ethics in news reporting has become more challenging. Over the years, the majority of machine learning-based research has been primarily devoted only to detecting hate speech, fake news, and offensive statements in developing semantic analysis systems. To tackle this pressing gap the study will aim to develop an automatic solution to uncover unethical reporting practices by identifying offensive patterns in Sinhala news headlines using natural language processing techniques. It will also focus on developing a solution to prevent its detrimental effects from thwarting them. Solution development involved several steps, and the data collection process was done by gathering over 2500 Sinhala news headlines from main digital media portals. Those data were annotated as "violated" and "non-violated" based on the code of ethics introduced by Verité Research Institute. Data preprocessing steps included tokenization, stop word removal, punctuation removal, spelling correction, numeric values removal, non-Sinhala values removal, and encoding. Two classification algorithms, Support Vector Machine, and Logistic Regression were used to train different models considering five feature extraction techniques. They were TF-IDF, N-Gram, Count Vectorizer, Word2Vec, and FastText. Based on the performance evaluated on the testing data, the combination of Support Vector Machine (SVM) with TF-IDF was chosen as the primary model due to its robust performance and adaptability across the vast data set and complex ethical scenarios in news headlines. This selected model outperformed other approaches achieving an accuracy of 91% on the tested dataset. Media ethics encompass a broad and complex field. This research focuses only on detecting violations in selected media ethics practices, specifically marginalization, prejudiced reporting, reporting on suicides, and cases of women abuse. Future research can expand on this work by addressing other ethical concerns in headlines, adding an automatic suggestion feature to correct violated headlines and extending the analysis to broader news content.

2024-11-01T00:00:00Z