Sentetik Açıklıklı Radar Görüntülerinde İnterferometri ve Radargrametri Tekniklerinin Füzyonu
Özet
With the development of satellite technology, advancements in active and passive remote sensing techniques are accelerated. The use of Synthetic Aperture Radar (SAR) sensor, which can perform electromagnetic imaging next to the optical sensors on satellites, makes it possible to obtain SAR images in unfavourable weather conditions (cloudy or even rainy) and in different times of the day (day or night). Satellite images are used for many purposes such as border security, assessments after a natural disaster, fault line tracking, railway and natural gas security, military and civilian mapping. Unlike optical systems, the SAR system produces its own beam source with its own sensor and operates as an active remote sensing system. This enables SAR system to gather complex images that contain phase information. To refine these images further, many techniques including SAR interferometry and radargrammetry have been developed and used. Methods that use phase information as well as amplitude information of SAR images are called SAR Interferometry (InSAR). Coherent Change Detection (CCD), Land Classification, Deformation Analysis and Digital Elevation Model (DEM) generation are amongst the numerous applications of SAR Interferometry technique. With the help of these methods, the SAR image can be assessed and enhanced. Radargrammetry is the first method to generate height information from radar images. The radargrammetry method can, briefly, be described as the extraction of geometric information from the radar images via application of photogrammetry technique on radar images. For this reason, it is also known as Stereo SAR. Unlike SAR interferometry, in the radargrammetry technique uses only the amplitude information and does not need the SAR image phase information. The differences in the distance between the targets in the imaged area and the sensors are used in this technique. Within the scope of the thesis; we conducted analyses on assessment of SAR images by creating a Digital Elevation Model with SAR Interferometry technique. It involves taking the SAR image pair and examining all steps of the algorithm until it creates a 3D height model. While examining the steps, the theoretical background of the the algorithms were also studied in detail and the simulation models were developed. Since the algorithm for generating DEM with the InSAR technique also covers the entire Coherent Change Detection algorithm with the InSAR technique, this application is also briefly mentioned in the thesis. The correlation map, which is the output of CCD, has been used as a feedback mechanism and guide map in two different steps in the generation of DEM with InSAR. In addition to the correlation map, many alternative versions such as pseudo correlation, phase derivative variance and amplitude change detection are also modelled. For similar scenarios, all of them were tried separately and comparative analyses were made. The emphasis in thesis is on the step-by-step phase unwrapping step, which is the heart of DEM generation with InSAR technique. Many phase unwrapping algorithms in the literature are analysed and many of them are simulated by employing different software languages. To reduce the required computation time, acceleration processes were applied for some of the algorithms in consideration, while others were developed by taking advantage of multiple programming languages to expedite the rendering. Subsequently, time and error analyses of all the phase unwrapping algorithms were performed and alternative solutions have been developed to correct the wrong phase unwrapping process, which is the main cause of errors when generating DEM. In addition, Radargrammetry technique, another technique of generating DEM of the region from SAR images, was also analysed. Similar to the SAR interferometry, the algorithmic steps in Radargrammetry technique were theoretically investigated one by one. Alternative algorithms in the literature were assesed and the best of them is selected. Then the selected technique is modelled in a simulation environment to run all the steps. Although both techniques require SAR images collected in different geometries, if they are fed with the SAR images of same region from the desired geometry for corresponding technique, they will be able to produce DEMs of the that particular region. Due to the nature of these techniques, they prone to make more mistakes in certain special cases compared to the rest when generating DEM. In this case, it causes erroneous zones to occur in the generated DEM. However, DEMs which are the outputs of the techniques, can be fused into a new DEM with higher accuracy. Fusion process can be performed statically by combining DEMs, by following a certain procedure. In this thesis, many static fusion alternatives have been studied and their results have been presented. However, it is impossible to correct the errors made in the same region in the output of both techniques via using static fusion. Instead, a different fusion algorithm was designed by running several techniques in tandem. A dynamic fusion algorithm is proposed in which the output of one technique is fed as an input to the other. To facilitate both the theoretical and simulation analysis of the techniques, we used real SAR images of suitable characteristics from different satellites. In addition, synthetic data were produced to feed SAR interferometry and radargrammetry techniques to better compare techniques and time analysis. All simulation models developed within this work accepts inputs via user-friendly graphical interfaces.
