Submitted by Lucienne Costa (lucienneferreira@ufcg.edu.br) on 2018-09-12T19:09:12Z No. of bitstreams: 1 JOÃO NAILSON DE CASTRO SILVA – DISSERTAÇÃO (PPGRN) 2017.pdf: 6066847 bytes, checksum: 06d6d215ba35522ddc64fb5c435d43f4 (MD5) / Made available in DSpace on 2018-09-12T19:09:12Z (GMT). No. of bitstreams: 1 JOÃO NAILSON DE CASTRO SILVA – DISSERTAÇÃO (PPGRN) 2017.pdf: 6066847 bytes, checksum: 06d6d215ba35522ddc64fb5c435d43f4 (MD5) Previous issue date: 2017-08-25 / Capes / A Caatinga é um bioma único que só ocorre na região do Semiárido do Brasil (SAB). Este bioma se diferencia principalmente pela capacidade de desenvolver mecanismos de adaptação aos baixos índices pluviométricos da região, além de apresentar alta resiliência nesse ambiente de elevada pressão antrópica. Tendo em vista que a cobertura vegetal exerce um papel muito importante no equilíbrio ambiental, as técnicas de sensoriamento remoto têm sido muito utilizadas para extrair informações biofísicas da vegetação. Este trabalho, teve como objetivo estabelecer uma metodologia que incorpore a capacidade de caracterização temporal do índice de vegetação à técnica de classificação de imagens para melhorar a identificação das classes de cobertura da Terra na bacia do Rio Sucuru, no Cariri paraibano. Neste sentido, busca-se uma classificação de referência dos padrões de cobertura da terra a partir de imagens de SR, para um período específico, no qual, seja possível uma validação em campo. Além disso, é realizada uma avaliação de uma série temporal de um índice de vegetação para melhoria da classificação realizada. Nessa pesquisa foi adotado parte da metodologia proposta por Chaves et al. (2008), para classificar os padrões de cobertura do solo e uma série temporal de EVI, processadas com 88 imagens selecionadas dos sensores ETM+ e OLI/TIRS, da série Landsat, para o período entre outubro de 2014 a setembro de 2016. Os resultados evidenciaram que analisar a cobertura vegetal utilizando apenas um único momento não retrata fidedignamente os padrões de cobertura do solo, visto que nesse ambiente semiárido as respostas que a vegetação apresenta diante da presença ou ausência de chuva são muito rápidas. Nesse sentido, os resultados mostram que uma análise espaço-temporal, utilizando um índice de vegetação, pode estabelecer uma melhor distinção das categorias atribuídas em uma classificação de padrões de cobertura do solo, possibilitando uma melhor percepção do comportamento da vegetação para um período de 24 meses observado. / The Caatinga is a unique biome that only occurs in the semi-arid region of Brazil. This biome is distinguished, mainly, by the capacity to develop mechanisms of adaptation to the low rainfall rates of the region. Plus, it also has a high resilience level upon high anthropic pressure. Considering that land cover plays a very important role in environmental balance, remote sensing techniques have been widely used to extract biophysical information from vegetation. The objective of this work is to establish a methodology that incorporates a time series vegetation index characterization to the image classification technique for improving the land cover classification in the Sucuru River basin in Cariri, Paraíba. In this sense, a Land Cover classification is acquired from SR images, for a specific data. For this data, there is a ground truth validation. In addition, an evaluation of a vegetation index time series is performed to improve the classification. In this research was adopted part of the methodology proposed by Chaves et al. (2008), to classify land cover patterns and a time series of EVI, processed with 88 images selected from the ETM + and OLI / TIRS sensors of the Landsat series for the period between October 2014 to September 2016. The results evidence that analyzing the land cover for a single time stamp could not reliably portray the land cover patterns since we often have fast changes before and after a rainfall event in this semi-arid environment. In this sense, the results show that a spatiotemporal analysis, using a vegetation index, can establish a better distinction of the categories assigned to a classification of land cover patterns, allowing a better perception of vegetation behavior for a period of 24 Months observed.

<p>Domain adaptation is one of the hottest directions in solving annotation insufficiency problem of deep learning. General domain adaptation is not consistent with the practical scenarios in the industry. In this thesis, we focus on two concerns as below.</p> <p>  </p> <p>  First is that labeled data are generally collected from multiple domains. In other words, multi-source adaptation is a more common situation. Simply extending these single-source approaches to the multi-source cases could cause sub-optimal inference, so specialized multi-source adaptation methods are essential. The main challenge in the multi-source scenario is a more complex divergence situation. Not only the divergence between target and each source plays a role, but the divergences among distinct sources matter as well. However, the significance of maintaining consistency among multiple sources didn't gain enough attention in previous work. In this thesis, we propose an Enhanced Consistency Multi-Source Adaptation (EC-MSA) framework to address it from three perspectives. First, we mitigate feature-level discrepancy by cross-domain conditional alignment, narrowing the divergence between each source and target domain class-wisely. Second, we enhance multi-source consistency via dual mix-up, diminishing the disagreements among different sources. Third, we deploy a target distilling mechanism to handle the uncertainty of target prediction, aiming to provide high-quality pseudo-labeled target samples to benefit the previous two aspects. Extensive experiments are conducted on several common benchmark datasets and demonstrate that our model outperforms the state-of-the-art methods.</p> <p>  </p> <p>  Second is that data privacy and security is necessary in practice. That is, we hope to keep the raw data stored locally while can still obtain a satisfied model. In such a case, the risk of data leakage greatly decreases. Therefore, it is natural for us to combine the federated learning paradigm with domain adaptation. Under the source-private setting, the main challenge for us is to expose information from the source domain to the target domain while make sure that the communication process is safe enough. In this thesis, we propose a method named Fourier Transform-Assisted Federated Domain Adaptation (FTA-FDA) to alleviate the difficulties in two ways. We apply Fast Fourier Transform to the raw data and transfer only the amplitude spectra during the communication. Then frequency space interpolations between these two domains are conducted, minimizing the discrepancies while ensuring the contact of them and keeping raw data safe. What's more, we make prototype alignments by using the model weights together with target features, trying to reduce the discrepancy in the class level. Experiments on Office-31 demonstrate the effectiveness and competitiveness of our approach, and further analyses prove that our algorithm can help protect privacy and security.</p>