Geri Dön

Monitoring rice fields using synthetic aperture radar

Başlık çevirisi mevcut değil.

PDF İndir

  1. Tez No: 648082
  2. Yazar: ONUR YÜZGÜLLÜ
  3. Danışmanlar: PROF. DR. IRENA HAJNSEK
  4. Tez Türü: Doktora
  5. Konular: Biyoteknoloji, Biyomühendislik, Çevre Mühendisliği, Biotechnology, Bioengineering, Environmental Engineering
  6. Anahtar Kelimeler: Belirtilmemiş.
  7. Yıl: 2017
  8. Dil: İngilizce
  9. Üniversite: Eıdgenossısche Technısche Hochschule Zurıch (eth)
  10. Enstitü: Yurtdışı Enstitü
  11. Ana Bilim Dalı: Belirtilmemiş.
  12. Bilim Dalı: Belirtilmemiş.
  13. Sayfa Sayısı: Belirtilmemiş.

Özet

Özet yok.

Özet (Çeviri)

This thesis focuses on the estimation of rice plant morphology and growth stage using an EM model inversion. For this purpose, different SAR image analysis techniques such as multi-polarimetric SAR and polarimetric SAR interferometry, have been employed. The algorithms are developed and validated using detailed ground measurements that were conducted between 2013 and 2015 from the Ipsala test site located in Turkey. This research presents the first implementation of stochastic inversion to an EM model, which considers agronomical, environmental, and modeling uncertainties. The study also contributes to the literature by introducing the use of surrogate models as an alternative for models that require high computation effort. This chapter synthesizes the dissertation with the main conclusions of the previous sections and concludes the thesis by providing investigation suggestions for the future of remote sensing based rice field monitoring. The temporal evolution of backscattering signatures in X-band SAR data is a non-monotonic function of time and growth stage. Therefore, some backscattering signatures can be measured in different stages of the growth. The proposed rice monitoring procedure proposes a method developed from copolar Pol-InSAR data. In 2015, TANDEM-X mission had a science phase to acquire large baseline bi-static interferometric acquisitions with height of ambiguity less than 5 meters. This mission allowed for the estimation of plant heights that are higher than 40 cm. RVoG model is used to predict height ranges of rice plants, which are later used as a constraint in the stochastic inversion. In the inversion approach developed in the study, the IRRI growth stages are used to organize the ground measured data to define the boundaries of the multi-dimensional parameter space of morphological descriptors. The chosen EM model with Monte Carlo simulations [82] is employed for simulating the backscattering intensities from a multi-dimensional parameter space. The performance of the EM model is assessed for each available polarimetric channel in X- and C-band for backscattering intensities. The accuracy analysis between model estimates and satellite measurements reached R2 values higher than 0.78 and RMSE values less than 2.85 dB. The integration of the Monte Carlo simulations into the multi-dimensional EM model leads to high computation effort. Towards solving this issue, the scattering model is substituted with its surrogate model, which was obtained through PCE algorithm [119]. The PCE metamodel was trained to mimic the behavior of the multi-dimensional EM model within the parameter space. The PCE-driven EM model reached R2 higher than 0.86 and RMSE less than 5.3 dB. The importance of the input parameters on the PCE-driven EM model simulations is investigated by conducting a global sensitivity analysis on a previously defined parameter space. The results showed that the stalk height and structural density parameters had a considerable influence over the estimated backscattering intensities. The proposed EM model inversion algorithm focuses on handling the effects of morphological variance within rice fields. In SAR based monitoring approaches, the information contained within a resolution cell represents an average of the scattering returns from hundreds of plants with different 71 Chapter 6 Summary and Conclusion morphologies. Different growth rates between those plants introduce a structural heterogeneity within fields. This heterogeneity can be observed in SAR images as parametric variance. The plant morphology based EM models become advantageous against other techniques on investigating this morphological variability. This thesis presents a new perspective to EM model inversion that aims to estimate the plant morphology as stalk height, stalk diameter, leaf length, and leaf width. The proposed approach uses a stochastic inversion, which searches the parameter space for the plant morphologies that have similar scattering returns to the measured backscattering signature. The success of the stochastic inversion algorithm depends on three conditions. The first condition is that high computation effort requirement is provided by employing a PCE metamodel. The other two conditions are provided as optimization constraints which are based on backscattering intensity and agronomical relations to limit the multi-dimensional parameter space. The measured intensity based constraint eliminates the plant morphologies that return different intensities. The agronomy based constraint removes the unlikely plant morphologies from the multi-dimensional parameter space using a convex-hull, which was defined using the data obtained from the ground measurements. The samples that are members of the complex-hull are considered to be agronomically possible morphologies for the measured backscattering intensity value. A detailed growth scale called BBCH classifies the crops according to their quantitative measures: number of leaves, tillers, and formation of flowers, panicles, and grains. Considering that there might be variations in plant morphology at a specific stage, the BBCH scale does not provide a direct link between growth stages and the morphologies. However, the stochastic inversion of the EM model provides estimations on the morphological measures of rice plants. The missing link is provided using high-degree polynomial relations generated by PCE metamodels. The results obtained from the stochastic inversion of the PCE-driven EM model with a priori IRRI growth phase showed accurate estimates of the crop height with absolute errors less than 15 cm. The integration of the EM model and the RVoG model inversion algorithms, on the other hand, reduced the errors in plant height estimations to values less than 10 cm. The proposed rice morphology estimation algorithms achieved R2 values higher than 0.6 between measured and estimated stalk diameter, leaf length, and leaf width. Lastly, in the BBCH stage estimations with the multi-dimensional non-linear relation provided by the PCE metamodel managed to have errors less than 10% compared to the measured values. The objectives of this research were accomplished by presenting a morphology based rice monitoring algorithm that is valid for multi-frequency and multi-polarimetric SAR image analysis techniques. As planned, the implementation of PCE surrogate model has reduced the computation effort significantly. Regarding the site independence, the algorithm was tested on different locations and presented promising results. The improvements in rice field monitoring for plant morphology and growth stage are expected to encourage agriculturists, local authorities, and insurance companies to use SAR data with different image analysis techniques. The major strengths, opportunities, and limitations of this dissertation are discussed below.

Benzer Tezler

  1. Tez No

    472626

    Paddy-rice leaf area index (LAI) estimation using radar and optical imagery

    Radar ve optik görüntüler kullanarak çeltik bitkisi yaprak alan indeks kestirim

    ELNAZ NAJATISHENDI

    Yüksek Lisans

    İngilizce

    İngilizce

    2017

    İletişim Bilimleriİstanbul Teknik Üniversitesi

    İletişim Bilimleri Ana Bilim Dalı

    DOÇ. DR. ESRA ERTEN

  2. Tez No

    1019691

    Sar ve optik veri entegrasyonuna dayalı çeltik alanlarının tespiti ve açıklanabilir yapay zeka (SHAP) ile analizi

    Identification of rice fields based on sar and optical data integration and analysis with explainable artificial intelligence (SHAP)

    MERİÇ BEKTAŞ

    Yüksek Lisans

    Türkçe

    Türkçe

    2026

    Jeodezi ve FotogrametriZonguldak Bülent Ecevit Üniversitesi

    Geomatik Mühendisliği Ana Bilim Dalı

    DOÇ. DR. ÇAĞLAR BAYIK

  3. Tez No

    2740

    Çocukluk çağı aritmilerinde sol ventrikül fonksiyonlarının iki boyutlu ekokardiyografi ile değerlendirilmesi

    Başlık çevirisi yok

    M.AKİF GÜVEN

    Tıpta Uzmanlık

    Türkçe

    Türkçe

    1988

    Çocuk Sağlığı ve HastalıklarıGazi Üniversitesi

    Çocuk Sağlığı ve Hastalıkları Ana Bilim Dalı

  4. Tez No

    2889

    Digital position encoder

    Başlık çevirisi yok

    VEDAT MEHMET KARSLI

    Yüksek Lisans

    İngilizce

    İngilizce

    1988

    Elektrik ve Elektronik MühendisliğiOrta Doğu Teknik Üniversitesi

    DOÇ. DR. CELAL KORASLI

  5. Tez No

    2899

    Design of a convergence recorder and improvement of a load monitoring system for measurements in underground mines

    Başlık çevirisi yok

    TALAT DEVRİM VURAL

    Yüksek Lisans

    İngilizce

    İngilizce

    1988

    Maden Mühendisliği ve MadencilikOrta Doğu Teknik Üniversitesi

    DOÇ. DR. ERDAL ÜNAL

Geri Dön