Optik uydu görüntülerinin birleştirilmesinde frekans bölgesi filtrelerinin karşılaştırılması

Abstract

Her geçen gün daha da gelişmekte olan uydu teknolojilerine paralel olarak uzaktan algılama alanındaki uygulamalarda da önemli gelişmeler gözlenmektedir. Ancak fiziksel şartlardan kaynaklanan kısıtlamalar nedeni ile optik algılayıcıların mekansal ve spektral çözünürlükleri arasında farklar oluşmaktadır. Algılanmış tek bir görüntüde en iyi sonuçları elde etmek mümkün olamamaktadır. Pankromatik görüntüler yüksek mekansal çözünürlüklere sahipken, çok bantlı multispektral görüntüler yüksek spektral çözünürlüğe sahiptir. Bu problemi çözmek ve yüksek mekansal ve spektral çözünürlüklü renkli görüntüler elde etmek için görüntü birleştirme yöntemleri önem kazanmaktadır. Birleştirilmiş görüntüler çok daha güvenilir ve yorumlanabilir sonuçlar üretmekte, gözlem alanları için detaylı veriye sahip görüntüler elde edilebilmektedir. Bu çalışmada görüntü birleştirme yöntemleri; akademik gruplarca genel kabul görmüş olan modülasyon bazlı, bileşen değişimi bazlı, çoklu çözünürlük analizi bazlı başlıkları altında incelenmiştir. Modülasyon bazlı yöntemlerden Brovey yöntemi incelenmiş olup, bileşen değişim bazlı yöntemlerden PCA, IHS yöntemleri ele alınmıştır. Çoklu çözünürlük analizi bazlı yöntemlerden ise frekans bölgesindeki süzgeçleme yöntemleri ağırlıklı olarak incelenmiş ve aralarındaki ilişkiye göre sonuçlar yorumlanmıştır. Elde edilen sonuçlar literatürde kabul görmüş olan kalite değerlendirme metrikleri ışığında incelenmiş ve metriklerin ideal değerlerine göre sonuçlar yorumlanmıştır. Sonuçların yorumlanmasıyla görüntü birleştirme yöntemleri ile ilgili olarak araştırmalarda ve çalışmalarda kullanılabilecek alternatif bir model geliştirilmesi amaçlanmaktadır.

In parallel with the ever-evolving satellite technologies, significant improvements are observed in applications in the field of remote sensing. However, there are differences between the spatial and spectral resolutions of optical sensors due to restrictions caused by physical conditions. It is not possible to obtain the best results in a single perceived image. Panchromatic images have high spatial resolutions, while multiband multispectral images have high spectral resolution. To solve this problem and to obtain high spatial and spectral resolution color images, image-combining methods are gaining importance. Consolidated images produce much more reliable and interpretable results and images with detailed data can be obtained for observation areas. In this study, image fusion methods had been investigated under the headings of modulation-based, component-based, multi-resolution-based, which are generally accepted by academic groups. While Brovey method is discussed under modulation –based; the PCA and IHS methods are discussed under component-based methods. In multi-resolution analysis, frequency domain filtering methods have been studied extensively and then finaly the results are intepreted according to the relation between them. The obtained results are examined in the light of the accepted quality evaluation metrics in the literature and the results are interpreted according to the ideal values. The interpretation of the results is intended to develop an alternative model which can be used in researches and studies on the integration of SPOT 5 satellite images. As the optical earth's orbits rotate in their polar orbits, the orbit around the earth, the angles of turning right and left, the camera and optics on them, and so on. Depending on the constraints of the equipment, a strip of about 500-1000 km travels and does not turn. xx The captured images are sent to the most optical satellite platforms without being processed and converted into images on the ground systems. The satellite image data transmitted from the radio frequency terminal is first subjected to radiometric correction, followed by geometric correction. Coordinate correction and satellite image data fitted to the appropriate coordinate plane are transformed into workable formata on many commercial satellite image processing software and archived. The development of analytical and digital image fusion techniques has gained momentum as the use of remote sensing data for many disciplines in the light of developing technology has increased. Within this scope, a variety of researches and studies have been carried out on the development of models that produce results. Multiband image with high spatial and spectral resolution can be obtained by combining single band panchromatic images with multispectral multispectral images. A very different model was introduced from 1980 to be used during this assembly process, and detailed studies were carried out in this context. The spatial and spectral values of the resulting images are interpreted as having the most inclusive values and different metric values are used for this. In this study, comparative analyzes of Fourier domain based models were made and the results were interpreted according to different metric values. The results obtained in this context are interpreted and it is aimed to develop an alternative model to be used in researches and studies related to the merging of SPOT 5 satellite images. The steps of image fusion can be briefly summarized as follows:  Co-registration of the MS image with the PAN image  Sampling of MS and PAN view  Magnifying the size of the MS image to the same size as the PAN image  Image fusion: MS image addition of spatial information extracted from PAN image  Finally, the evaluation of the results obtained with reference to the quality metrics by referring to the original MS image and interpretation of its achievements. Because the MS and PAN image detection for the SPOT satellites is performed in the same time zone, the first step is not implemented since there is no time delay between the images. Unlike previous studies, the image combining methods described in this study were evaluated according to the six-faceted image evaluation method, focusing on how the output will be affected if some decision steps included in the intermediate steps of the image combining technique are not based on a single option but on different options. RMSE, SAM, RASE, ERGAS, UIQI and QNR metric are used for examining the results. SAM, RMSE, RASE and ERGAS methods are expressed as best when the results are 0; UIQI and QNR metrics are is best when the results are 1. The obtained results are compared according to these rules and detailed in tables. When the obtained results are examined, it is observed that the performance coefficient does not develop in direct proportion with the resolution level. It has evolved from Brovey's method to wavelet transform and multiresolution methods. In parallel with the developments in satellite technologies, it is expected that existing methods will be inadequate. New methods for developing additional features through hbyrid models are still underway. The most important problem encountered within the scope of the study is the large number of possibilities arising from the fact that there are too many options to be decided in the intermediate steps of the methods. MATLAB application tool is used in calculations. More detailed analysis on this field in this data stream can be made much faster with modern methods such as parallel programming and large data analysis on sophisticated computer clusters and servers. This thesis is based on the theoretical results and the new methods that will be modeled by going out from the theoretical results will be a pioneer of applications that can play an active role in many areas of our life…