京都大学 / 0048 / 新制・課程博士 / 博士(農学) / 甲第21429号 / 農博第2307号 / 学位論文||H30||N5157(農学部図書室) / 京都大学大学院農学研究科地域環境科学専攻 / (主査)教授 近藤 直, 准教授 小川 雄一, 教授 飯田 訓久 / 学位規則第4条第1項該当 / Doctor of Agricultural Science / Kyoto University / DGAM

No description available.

No description available.

No description available.

No description available.

No description available.

No description available.

No description available.

Buildings are essential parts to human life, which provide the place to dwell, educate, entertain, etc. However, they are usually vulnerable to earthquakes, and collapsed buildings are the main factor of fatalities and directly impact livelihoods. It is particularly important to quickly and accurately obtain damaged building conditions for further planning rescue. Remote sensing has the ability to quickly capture the information of damaged buildings in a large area, and remote sensing imagery has been used by government organizations, international agencies, and insurance industries for assessing post-event damage. The application of deep learning is encouraged by recent technological developments, enabling the processing of increasing amounts of data in a reasonable time as well as the use of more complex models. In this thesis, deep learning is explored for identifying collapsed buildings using very high-resolution remote sensing imagery after the 2010 Haiti earthquake. In the present study, a simple architecture of convolutional neural network (CNN) model was proposed to evaluate the potential of CNN for extracting features and detecting collapsed buildings using only post-event very high-resolution remote sensing imagery. Three balancing methods were considered to reduce the effect of the imbalance problem for the performance of the CNN, and the results showed that a suitable balancing method should be considered when facing imbalance dataset to retrieve the distribution of collapsed buildings. To improve the classification accuracy, pre- and post-event very high-resolution remote sensing imagery were considered, and a conventional classification method was combined with the CNN. Compared to conventional texture features, deep features learnt from CNNs had better performance for identifying collapsed buildings, and the accuracy was further improved by combing CNN features with random forest classifier. For the limited dataset, a pretrained CNN model was applied to detect collapsed buildings, and the effect of data augmentation was also investigated. The experimental results demonstrated that the pretrained CNN model outperformed the model trained from scratch for identifying collapsed buildings.

Predicting wall heat flux accurately in wall-bounded turbulent flows is critical for a variety of engineering applications, including thermal management systems and energy-efficient designs. Traditional methods, which rely on expensive numerical simulations, are hampered by increasing complexity and extremly high computation cost. Recent advances in deep neural networks (DNNs), however, offer an effective solution by predicting wall heat flux using non-intrusive measurements derived from off-wall quantities. This study introduces a novel approach, the convolution-compacted vision transformer (ViT), which integrates convolutional neural networks (CNNs) and ViT to predict instantaneous fields of wall heat flux accurately based on off-wall quantities including velocity components at three directions and temperature. Our method is applied to an existing database of wall-bounded turbulent flows obtained from direct numerical simulations (DNS). We first conduct an ablation study to examine the effects of incorporating convolution-based modules into ViT architectures and report on the impact of different modules. Subsequently, we utilize fully-convolutional neural networks (FCNs) with various architectures to identify the distinctions between FCN models and the convolution-compacted ViT. Our optimized ViT model surpasses the FCN models in terms of instantaneous field predictions, learning turbulence statistics, and accurately capturing energy spectra. Finally, we undertake a sensitivity analysis using a gradient map to enhance the understanding of the nonlinear relationship established by DNN models, thus augmenting the interpretability of these models. / <p>Presentation online</p>