http://dlib.iit.ac.lk/xmlui/handle/123456789/2624 Sat, 18 Apr 2026 00:34:17 GMT 2026-04-18T00:34:17Z http://dlib.iit.ac.lk/xmlui/handle/123456789/2759 A Gamified Solution to Managing Anger in Children with Conduct Disorder Amjah, Zikra "The problem of Conduct Disorder (CD) in children and adolescents arises through disruptive behaviours, including aggression, theft, and vandalism, often leading to long-term negative outcomes such as legal issues and poor physical health. Diagnosis relies on clinical evaluation and adherence to diagnostic criteria. Additionally, Aggression is a prominent feature of CD, as children with CD are more prone to experiencing heightened irritability and difficulty managing anger compared to their peers. Addressing the lack of digital interventions for managing anger tantrums in children with CD, this study proposes utilizing elements of which children are drawn to, being games and music. Drawing from Cognitive Behavioural Therapy (CBT) and Music Therapy principles, a gamified approach is planned, that being the application of digitalized-CBT (dCBT), with calming effects to mitigate aggressive episodes. Through the integration of rhythm game elements with dCBT and Music Therapy principles, the prototype effectively detects beats, manages player controls, and ensures beat correspondence. Testing demonstrates seamless interaction, with players able to tap in alignment with the beat to trigger sound emission and score points. These results indicate potential effectiveness in managing anger episodes in children with CD, but further testing and refinement are required to validate its effectiveness in therapeutic applications. " Mon, 01 Jan 2024 00:00:00 GMT http://dlib.iit.ac.lk/xmlui/handle/123456789/2759 2024-01-01T00:00:00Z http://dlib.iit.ac.lk/xmlui/handle/123456789/2758 Detecting Subtle Deepfakes in Celebrities Using Generalize Deep Learning Kankanamge, Yovindu "Deepfakes refer to altered digital content through deep generative models to achieve photorealistic standards. Deepfake instances related to celebrities are growing due to the accessibility of large quantities of data. Recently, several deepfake detection approaches have been developed to focus on identifying fake faces from genuine faces when they are tested on similar training forgery patterns. Somehow, these methods' performance is limited when tested on unseen patterns. In this research, the author proposes a deepfake detection method to increase the generalization of the model by detecting both pixel-level and noise-level manipulations to make the final prediction. The proposed network has two methods for extracting features. First, there's a pixel-level classifier that detects pixel-level changes using convolutional neural networks. Second, there's a noise-level classifier that identifies noise-level changes based on the Peak Signal-to-Noise Ratio (PSNR) to verify the authenticity of celebrity images. After that, a logical method was developed to compare the results from both classifiers. If either classifier detects the image as 'fake,' the final output will be labelled as fake, covering both pixel and noise manipulations. The proposed method extracts forgery traces caused by pixels and noise manipulations suggesting to have a good generalization ability. Extensive experimental results on three widely used datasets show that the proposed method achieves better generalization performance against unseen forgery patterns." Mon, 01 Jan 2024 00:00:00 GMT http://dlib.iit.ac.lk/xmlui/handle/123456789/2758 2024-01-01T00:00:00Z http://dlib.iit.ac.lk/xmlui/handle/123456789/2757 SignSync - Machine Learning Driven Sign Language Translation Assistant Luvishewa, Yasiru "The presented research addresses a significant challenge in the realm of communication: the language barrier faced by deaf and hard-of-hearing individuals. Existing sign language-to-text or speech and vice versa translation tools have made strides toward mitigating this issue; however, a gap remains in facilitating seamless communication among users of different sign languages. This gap, underscored by the variations and nuances specific to regional sign languages, such as American Sign Language (ASL) and British Sign Language (BSL), presents a formidable challenge in achieving effective and inclusive communication. To tackle this issue, the research proposes the development of a real-time machine learning powered sign language translation system. This system leverages advanced machine learning techniques, including Deep Neural Networks (DNNs) and Support Vector Machines (SVMs), integrated with the MediaPipe framework for enhanced gesture, hand movement, and facial expression recognition. These technologies are aimed at translating sign language in real-time, thereby bridging the communication gap between individuals using different sign languages. First, some quantitative measurements to confirm the effectiveness of the solution. Confusion metrics and AUC-ROC curves show the accuracy, recall, and overall performance of the model for classification tasks. According to preliminary data, accuracy rates are promising, which suggests that the application may have a positive effect on empowering the deaf community and removing obstacles to communication." Mon, 01 Jan 2024 00:00:00 GMT http://dlib.iit.ac.lk/xmlui/handle/123456789/2757 2024-01-01T00:00:00Z http://dlib.iit.ac.lk/xmlui/handle/123456789/2756 CleanSentry - Deep Learning-Based Beach Waste Material Classification and Report Generation System on Colombo District Seaside in Sri Lanka Samarakoon, Yashini "Waste pollution in Sri Lanka's coastal regions poses a significant challenge, affecting marine ecosystems, residents, tourism, and the environment. Addressing this issue necessitates innovative solutions to enhance sustainability efforts and optimize waste management processes. This research utilizes machine learning techniques to address the pressing need for effective waste classification. By leveraging convolutional neural networks (CNNs), the proposed solution aims to automate waste identification and categorization, thereby improving waste sorting practices. The methodology involves developing and training a CNN model using preprocessed drone-captured waste images. The MobileNetV2 model architecture, stripped of its top layer, is augmented with custom layers for global average pooling and a dense (output) layer, tailored to the specific classification task. The model achieves an accuracy of 80.74%, indicating its effectiveness in correctly classifying waste instances. However, the research faced limitations due to limited data availability and varying image data quality, impacting model robustness. The scope was further limited to generating classification reports rather than implementing real-time alerts, affecting immediate responsiveness. Future enhancements include implementing multiple object detection, integrating real-time reporting functionality, and incorporating geolocation mapping for enhanced waste management." Mon, 01 Jan 2024 00:00:00 GMT http://dlib.iit.ac.lk/xmlui/handle/123456789/2756 2024-01-01T00:00:00Z