Alam, Md. Golam RabiulTahsin, AnikaFairooz, MaishaRabbi, Gazi Rehan2024-09-082024-09-08©20242024-05ID 23141058ID 23141060ID 20101080http://hdl.handle.net/10361/23998Cataloged from PDF version of thesis.Includes bibliographical references (pages 68-71).This thesis is submitted in partial fulfillment of the requirements for the degree of Bachelor of Science in Computer Science, 2024.This research delves deeply into the intricate dynamics of wetlands in Bangladesh, with a particular focus on the haors, utilizing continuous monitoring to grasp the nuanced temporal changes that occur. It introduces an innovative unsupervised se mantic segmentation methodology tailored for analyzing the yearly fluctuations in wetlands. Leveraging the rich dataset provided by multi-temporal satellite imagery and cutting-edge unsupervised learning algorithms, this approach stands poised to revolutionize our understanding of wetland dynamics. At the heart of our method ology lies the strategic application of feature extraction and advanced clustering techniques, with a notable inclusion being the decoder model. These techniques enable the segmentation of wetland regions based on discernible patterns of expan sion and contraction. Moreover, our research extends beyond mere segmentation, incorporating time series methods to forecast wetland fluctuations. By integrating predictive analytics into our framework, we strive to provide not just a snapshot of wetland conditions but also insights into their future trajectories. To validate the efficacy of our approach, rigorous comparative analyses with actual data are conducted. This empirical validation serves to enrich our comprehension of river system dynamics and lends support to ongoing wildlife preservation initiatives. Our methodology represents a significant advancement in unsupervised learning meth ods, adept at adapting to dynamic conditions without the constraints of labeled training data. Furthermore, the incorporation of advanced clustering techniques enhances our ability to pinpoint regions undergoing substantial changes, thereby facilitating targeted conservation efforts. Crucially, the journey continues after seg mentation and prediction. Post-processing of segmentation results allows for metic ulous accuracy assessment, ensuring the reliability of our findings. Through a series of meticulously designed experiments, we showcase the robustness and effective ness of our methodology and model. By pushing the boundaries of unsupervised semantic segmentation and environmental research, we aspire to make meaningful contributions to the broader scientific community and pave the way for informed conservation strategies.71 pagesenBrac University theses are protected by copyright. They may be viewed from this source for any purpose, but reproduction or distribution in any format is prohibited without written permission.Wetland localizationSemantic segmentationImage clusteringGaussian Hidden MarkovWetland mitigation.Semantic computing.Thesis