京都大学 / 0048 / 新制・課程博士 / 博士(工学) / 甲第19411号 / 工博第4127号 / 新制||工||1636(附属図書館) / 32436 / 京都大学大学院工学研究科都市社会工学専攻 / (主査)教授 小池 克明, 教授 田村 正行, 教授 三ｹ田 均 / 学位規則第4条第1項該当 / Doctor of Philosophy (Engineering) / Kyoto University / DFAM

<p>In recent years, hyperspectral imaging technologies have been broadly applied to evaluate complex plant physiological features such as leaf moisture content, nutrient level and disease stress. A  critical  component  of  this  technique  is  white  referencing  used  to  remove  the  effect  of  non-uniform  lighting  intensity  in  different  wavelengths  on  raw  hyperspectral  images. Based  on  the  literature,  the leaf  geometry (e.g.,  tilt  angles)  and its interaction  with  the  illumination  severely impact  the  plant  reflectance  spectra  and vegetation  indices  such  as  the  normalized  difference  vegetation index (NDVI).  This thesis is  aimed to address the issues caused by the tilt angles across the leaf surface. To achieve this, two methods based on the fusion of the hyperspectral images and 3D  point  clouds  were  proposed.  The  first  method  was  to  build  a  3D  white  reference  library  in  which a point with almost the same tilt angle, height and position with the pixel on the plant leaf can be found, and then the white reference spectrum at that point can be used to calibrate the raw spectrum of the leaf pixel. The second method was to observe and summarize how the plant spectra and NDVI values changed with the leaf angles. Using the changing trends, the original NDVI and spectra  of  leaf  pixels  at  different  angles  can  be  calibrate  to  a  same  standard  as  if  the  leaf  was  imaged  at  a  flat  and  horizontal  surface.  The  approach  was  called  3D  calibration.  The  results  showed  that  the NDVI  values significantly  changed  with  leaf  angles  and  the  changing  trends differed  between  the  corn  and  soybean  species.  To evaluate the  performance  of  3D  calibration, 180 soybean plants with different genotypes, nitrogen (N), phosphorus (P) and water treatments were  grown  in  the  greenhouse. Each  plant  was  imaged  in three systems:  the high-throughput greenhouse hyperspectral imaging system, the indoor desktop imaging system with a visible-near infrared  (VINIR)  hyperspectral camera and  an  Intel  RealSense  depth  camera  and  the handheld device hyperspectral imaging system. In the greenhouse system, the whole canopy was captured. In the indoor desktop system, the partial canopy was captured because of the space limitation and the  top-matured  leaf  (the  middle  leaf  of  the  uppermost  matured  trifoliate)  was  focused.  The proposed  3D  calibration  was  applied  on  the  top-matured  leaf  to  remove  angle  impacts.  In  the  handheld device system, the flat top-matured leaf was captured. After done with imaging work, the plants were harvested to collect the ground truth data such as relative water content (RWC), N content and P content. Combined with the ground truth data, the NDVI values from three systems were  used  to  discriminate  different  genotypes  and  biochemical treatments,  whereas,  the  spectra from three systems were used to build partial least squares regression (PLSR) models for N, P and RWC. The results showed that the averaged tilt angles of top-matured leaves were impacted by different treatments. For instance, the low-nitrogen (LN) plants showed significantly higher leaf angles than high-nitrogen (HN) plants; the leaf angles on water-stressed (WS) plants were higher than those on well-watered (WW) plants. The leaf angles carried some signals that influenced not only the NDVI discrimination but also the PLSR modelling results. The signals were lost after 3D calibration.  For  the  top-matured  leaves,  the  discrimination  and  modelling  results  after  3D  calibration in the indoor desktop system were close to those from the flat leaves in the handheld device  system.  The  proposed  3D  calibration  approach  has  a  potential  to  eliminate  leaf  angle  impacts.</p>

This work aims to develop a deep learning model for land cover quantification through hyperspectral unmixing using an unsupervised autoencoder. Land cover identification and classification is instrumental in urban planning, environmental monitoring and land management. With the technological advancements in remote sensing, hyperspectral imagery which captures high resolution images of the earth's surface across hundreds of wavelength bands, is becoming increasingly popular. The high spectral information in these images can be analyzed to identify the various target materials present in the image scene based on their unique reflectance patterns. An autoencoder is a deep learning model that can perform spectral unmixing by decomposing the complex image spectra into its constituent materials and estimating their abundance compositions. The advantage of using this technique for land cover quantification is that it is completely unsupervised and eliminates the need for labelled data which generally requires years of field survey and formulation of detailed maps. We evaluate the performance of the autoencoder on various synthetic and real hyperspectral images consisting of different land covers using similarity metrics and abundance maps. The scalability of the technique with respect to landscapes is assessed by evaluating its performance on hyperspectral images spanning across 100m x 100m, 200m x 200m, 1000m x 1000m, 4000m x 4000m and 5000m x 5000m regions. Finally, we analyze the performance of this technique by comparing it to several supervised learning methods like Support Vector Machine (SVM), Random Forest (RF) and multilayer perceptron using F1-score, Precision and Recall metrics and other unsupervised techniques like K-Means, N-Findr, and VCA using cosine similarity, mean square error and estimated abundances. The land cover classification obtained using this technique is compared to the existing United States National Land Cover Database (NLCD) classification standard. / Master of Science / This work aims to develop an automated deep learning model for identifying and estimating the composition of the different land covers in a region using hyperspectral remote sensing imagery. With the technological advancements in remote sensing, hyperspectral imagery which captures high resolution images of the earth's surface across hundreds of wavelength bands, is becoming increasingly popular. As every surface has a unique reflectance pattern, the high spectral information contained in these images can be analyzed to identify the various target materials present in the image scene. An autoencoder is a deep learning model that can perform spectral unmixing by decomposing the complex image spectra into its constituent materials and estimate their percent compositions. The advantage of this method in land cover quantification is that it is an unsupervised technique which does not require labelled data which generally requires years of field survey and formulation of detailed maps. The performance of this technique is evaluated on various synthetic and real hyperspectral datasets consisting of different land covers. We assess the scalability of the model by evaluating its performance on images of different sizes spanning over a few hundred square meters to thousands of square meters. Finally, we compare the performance of the autoencoder based approach with other supervised and unsupervised deep learning techniques and with the current land cover classification standard.

V této práci je prezentována inovativní metoda zvaná \textit{Zobrazovací Reflektometrie}, která je založena na principu spektroskopické reflektometrie a je určena pro vyhodnocování optických vlastností tenkých vrstev .\ Spektrum odrazivosti je získáno z map intenzit zaznamenaných CCD kamerou. Každý záznam odpovídá předem nastavené vlnové délce a spektrum odrazivosti může být určeno ve zvoleném bodu nebo ve vybrané oblasti.\ Teoretický model odrazivosti se fituje na naměřená data pomocí Levenberg~-~Marquardtova algoritmu, jehož výsledky jsou optické vlastnosti vrstvy, jejich přesnost, a určení spolehlivosti dosažených výsledků pomocí analýzy citlivosti změn počátečních nastavení optimalizačního algoritmu.