Browsing by Author "Dereje Tadesse"

Now showing 1 - 4 of 4
  • No Thumbnail Available
    Item
    Ensemble Learning with Attention and Audio for Robust Video Classification
    (Addis Ababa University, 2025-06) Dereje Tadesse; Beakal Gizachew (PhD)
    The classification of video scenes is a fundamental task for many applications, such as content recommendation, indexing, and monitoring broadcasts. Current methods often depend on annotation-dependent object detection models, restricting their generalizability when working with different types of broadcast content, particularly cases where visual clues like logos or brands may not have clear definition or presence. This thesis is intended to address the problems associated with current methods through describing a two-stage classification framework that integrates both recognized and unheard information to improve accuracy and robustness of classification. The first stage utilizes a detection model based on pretrained models of object detection and enhanced spatial attention to detect physical visual markers (such as program logo or branded intro sequences) in video program content. However, individual visual indicators are sometimes not robust enough to add confidence, especially in content such as situational comedies where logos do not exist. The second stage describes a twostaged, early fusion ensemble presentation of convolutional neural network-based visual features and recurrent neural network-based audio features. The two modes each use some complementary properties, thus could be used for more robust classification. Experiments were completed with a dataset of approximately 19 hours of content from 13 TV programs across three channels, all focused on intro, credit, and outro segments. The visual-only model achieved 96.83% accuracy, while the audio-only model achieved 90.91%. The proposed early fusion ensemble method achieved 94.13% accuracy and revealed more robustness in difficult situations when quality of visual data was low or ambiguous. Ablation studies contrasting model performance with different ensemble methods confirmed the greater utility of early fusion and its capturing of cross-modal interactions. The system is also designed to be computationally efficient allowing for operationalization in broadcast media settings. This work, while also demonstrating methodical video classification ability, fills a significant gap for scalable and generalizable video classification through the integration of multimodal learning, especially with large amounts of uncontrollable annotations which has previously been a hurdle to more typical models.
  • No Thumbnail Available
    Item
    Ensemble Learning with Attention and Audio for Robust Video Classification
    (Addis Ababa University, 2025-06) Dereje Tadesse; Beakal Gizachew (PhD)
    The classification of video scenes is a fundamental task for many applications, such as content recommendation, indexing, and monitoring broadcasts. Current methods often depend on annotation-dependent object detection models, restricting their generalizability when working with different types of broadcast content, particularly cases where visual clues like logos or brands may not have clear definition or presence. This thesis is intended to address the problems associated with current methods through describing a two-stage classification framework that integrates both recognized and unheard information to improve accuracy and robustness of classification. The first stage utilizes a detection model based on pretrained models of object detection and enhanced spatial attention to detect physical visual markers (such as program logo or branded intro sequences) in video program content. However, individual visual indicators are sometimes not robust enough to add confidence, especially in content such as situational comedies where logos do not exist. The second stage describes a twostaged, early fusion ensemble presentation of convolutional neural network-based visual features and recurrent neural network-based audio features. The two modes each use some complementary properties, thus could be used for more robust classification. Experiments were completed with a dataset of approximately 19 hours of content from 13 TV programs across three channels, all focused on intro, credit, and outro segments. The visual-only model achieved 96.83% accuracy, while the audio-only model achieved 90.91%. The proposed early fusion ensemble method achieved 94.13% accuracy and revealed more robustness in difficult situations when quality of visual data was low or ambiguous. Ablation studies contrasting model performance with different ensemble methods confirmed the greater utility of early fusion and its capturing of cross-modal interactions. The system is also designed to be computationally efficient allowing for operationalization in broadcast media settings. This work, while also demonstrating methodical video classification ability, fills a significant gap for scalable and generalizable video classification through the integration of multimodal learning, especially with large amounts of uncontrollable annotations which has previously been a hurdle to more typical models.
  • No Thumbnail Available
    Item
    Financing Urban Infrastructure and Services in Ethiopia: The Case of Solid Waste Management in Adama (Nazareth) town
    (Addis Ababa University, 2001-06) Dereje Tadesse; Meheret Ayenew
    This study reports the situation and financing of solid waste management sen'ices in Adama town of Oromia Region (Ethiopia). The study showed that the town is currently providing inadequate solid waste management services. The town administration collects and disposes less than 15 percent of the wastes generated by the town annually. It is also found out that the solid waste collection and disposal systems are backward and are not economical as well. The institutional arrangements' for the management of both liquid and solid wastes are also very weak. The town administration allocates a marginal budget to this service which is less than 10 percent of its total budget as opposed to towns of developing countries that commit between 20 and 40 percent of their budget for the same. Low budget and low performances are generally attributed to scarcity of resources. However, problems of insufficient mobilization of the available resources and misuses are found to be equally serious according to this study. This shows that studying the financing system of urban services is an important area to reckon since it is helpfit! to look for ways by which the service may be improved by using the available resources differently. It may also be of advantage to check whether the available meager resources are effectively and efficiently used for the priority and basic needs of the town. The study also suggests some lines of action that are essential to improve the current inadequate solid waste management service of Adama town.
  • No Thumbnail Available
    Item
    A Practitioner Inquihv Into Pre-Service EFL Reflective Practicum of Haramaya Universitv: A Condition of Inabilltv to Reflect and Determinants of Effective Reflection
    (Addis Ababa University, 2009-07) Dereje Tadesse ; Hailom Banteyerga
    The thrust of this study was experiential observation of pre-service EFL student teachers' inability to reflect inion their practices at Haramaya University practicum context. The overall aim of the study was to, firstly, critically analyze the problem in its context with the intention to understand the facts of the situation of the problem and, secondly, make an mqulry into ways for maximizing the student teachers' ability to reflect in their context. A qualitative paradigm and practitioner inquiry design were adopted. Grounded Theory Method of data analysis was employed to systematically thematize, categorize and discover patterns and processes in the data. Ten student teachers, as respondents and practitioners, as well as four of their teacher educators, as informants, were selected by means of purposeful sampling, to take part in the study. Participant observation methods that involve unobtrusive observation, complete classroom observation, unstructured interview reflective journaling and discussions and practicum document gathering were used to collect qualitative data. The findings of the Contextual Analysis showed that the core factor for the EFL student teachers' inability to reflect is mainly their lack of effective reflection tools for reflectionfor/ on/about-actions and lack of effective time for reflection-in-action. The data suggests that, consequently, the central strategy of reflective practice they adopted was overdependence on and replication of the existing school syllabi without critical reflection. Based on these Contextual Analysis findings an Inquiry was next conducted, whereby participant student teachers were engaged on reflective journaling for two semesters. The findings of the Inquiry showed significant levels of improvement in their ability to reflect forlin/on action. For instance, the data analysis showed that they steadily began to reduce non-reflective behaviors such as over-advocating own actions, protectionism of self and peers, and exchange of distorted information, each of which initially blocked reflectivity. Gradually, they began to take such reflective actions as reflective observation of pupils' behaviors, reflective planning of lessons and actions, reflective classroom acts such as appreciative judgments of pupils' behavior, revision of some taken for granted assluuptions, promotion of sharing of information in classrooms and effective time management. Yet, some non-reflectivity behaviors such as context dilemma continued to persist due to influence of macro factors. All the domains of EFL refl ective skills- English, teaching and inquiry-improved. From the findin gs, a conclusion has been reached that the student teachers' inabil ity to effectively reflect related to their lack of effective reflection tools and control over lesson times and material s. By providing these conditions, student teachers' potential to effectively reflect and reconstruct new skills and knowledge from their experience can be prompted. Studies on conditions for cross-institutional and interdisciplinary practitioners' reflection are suggested as a major further area to be researched.