Yapay açıklıklı radar görüntülerinde hareketli hedef tespiti

Abstract

Günümüzde gerçekleştirilen birçok uygulamada yapay açıklıklı radar (YAR) sistemlerinden yararlanılmaktadır. Bu sistemlerin sağladığı yüksek çözünürlüklü görüntüler kullanılarak belirli bir yeryüzü bölgesini uzaktan görüntüleme, bölge içinde hareketli hedef izleme ve görüntü analizi gibi uygulamalar başarılı bir şekilde gerçekleştirilebilmektedir.YAR sistemleriyle elde edilen görüntüler içindeki hareketli hedeflerin tespit edilmesi son zamanlarda üzerinde sıkça çalışılan bir konudur. Bu tez çalışmasında, elde edilen YAR görüntüleri içindeki hareketli hedeflerin tespit edilmesiyle ilgili iki yeni yöntem önerilmiştir.Önerilen ilk yöntemde (Yöntem I) yapay açıklıklı radar görüntüleri içindeki hareketli hedeflerin tespiti ve hareket parametrelerinin çıkartılması için yeni bir algoritma geliştirilmiştir. Önerilen algoritma ile hem menzil hem de azimut yönünde hareket eden hedefler tespit edilebilmektedir. Buna ilave olarak tespit edilen hareketli hedefin hız ve hareket doğrultusu gibi hareket parametreleri yüksek doğrulukla hesaplanabilmektedir.Bu yöntemde hedefin azimut yönündeki hareketi kesme ortalaması algoritmasıyla; menzil yönündeki hareketi ise alt açıklık algoritmasıyla tespit edilmektedir. Bu iki algoritmanın birbirinden bağımsız olarak çalışabilmesi hedef tespitinin menzil ve azimut için aynı anda paralel olarak yapılabilmesini sağlar. Ayrıca veriler bağımsız olarak işlenebildiği için elde edilen sonuçların doğruluğu ve kararlılığı yüksektir.Önerilen ikinci yöntemde (Yöntem II) YAR görüntüsü içindeki hareketli hedeflerin tespiti için alt yama algoritması kullanılmaktadır. Alt yama algoritması literatürdeki mevcut uygulamalarda görüntü çözünürlüğünü arttırmak amacıyla kullanılmaktadır. Ancak bu tez çalışmasında alt yama yönteminin görüntü çözünürlüğünü arttırmanın yanında görüntü içindeki hareketli hedefleri tespit edebilmek için kullanılabileceği gösterilmiştir.Bu yöntem ile hem menzil hem de azimut yönlerinde hareket eden hedefler başarılı bir şekilde tespit edilebilmektedir. Önerilen yöntemin bu çalışmada gösterilen uygulamasında ana görüntüden dört tane alt yama görüntüsü oluşturulmaktadır. Bu alt yama görüntüleri kullanılarak sistemde bulunan hareketli hedeflerin tespit edilme başarımı arttırılmaktadır. Yöntem I'de olduğu gibi görüntü içinde azimut ve menzil doğrultularında hareket eden hedefler birbirlerinden bağımsız olarak incelenebilirler. Kullanılan veriler bağımsız olarak işlenebildiği için elde edilen sonuçların doğruluk ve kararlılığı yüksektir.Önerilen her iki yöntem bilgisayar benzetimlerinde çalıştırılarak test edilmiştir. Bu yöntemlerde, hedef alanı içine yerleştirilen hareketli hedefler başarıyla tespit edilebilmiştir. Bu yöntemler kullanılarak herhangi bir yönde hareket eden hedeflerin yüksek doğrulukla tespit edilebildiği gösterilmiştir. Ayrıca her iki yöntemle hareketli hedefin görüntü içinde hareket ettiği doğrultu bulunabilmekteyken Yöntem I ile buna ilave olarak hareketli hedefin hızı da başarılı bir şekilde hesaplanabilmektedir.

Synthetic aperture radar (SAR) systems are used widely in modern world applications. Any terrain image can be collected easily with the help of SAR imagery technology. These detailed images are useful for researching terrain properties and these images can be used for remote sensing, target tracking and image analysis applications.The main application area of SAR systems is aimed for high detailed imaging of specific earth terrains. Applications of SAR systems include environmental research such as calculating the rate of change of urban / forest areas, scientific researches, civilian applications such as observing population distribution, mapping, and military purposes like surveillance, enemy detection and tracking.One widely used application of SAR systems is detecting moving targets within the SAR images, and if it is possible, tracking the detected moving targets. Detection and motion parameter estimation of moving targets within the observed region is possible by using the SAR images. Information of detected moving objects can be used in very different applications, such as monitoring traffic flow, observation of military field, tracking of a specific moving target and motion parameter estimation of the moving targets.There are numerous papers exist in the literature defining different algorithms for detection of moving targets in SAR images. Most of the algorithms are detecting moving targets by using displaced phase centre antenna technique, along track interferometry, single-channel radar processing, and different kinds of focusing algorithms.The main success criteria for an algorithm used for moving target detection applications should be considered by the following three subjects: detecting the existence of the moving target, finding its motion direction and calculating its velocity correctly within a considerable error range.For detecting moving targets many algorithms are evaluated based on the typical blurring and displacement effects caused by the moving target within the image. Moving targets are appeared defocused or at wrong positions depending on the direction of the target motion within the SAR image. If a target moves in azimuth direction, motion causes blurring effect in azimuth direction, and if it moves in range direction, in addition to blurring effect, motion also causes a displacement in azimuth direction. If the velocity of the target is greater than a threshold value, moving target is even disappears.In this thesis, two new methods (called Method I and Method II) are proposed for detecting moving targets in a SAR image. These two new methods are based on the image defects described in the previous paragraph.The first method (Method I) proposes a new algorithm for detecting moving targets in SAR images. This algorithm combines two existing algorithms for detecting moving targets, which are shear averaging algorithm and sub-aperture processing algorithm. These two different algorithms are well defined, and can be independently used for detecting moving targets. The shear averaging algorithm can successfully detect moving targets travelling in the azimuth direction. On the other hand, sub-aperture processing algorithm can be used for detecting targets moving in the range direction. By combining these two algorithms moving targets travelling in any direction can be successfully detected.In the second proposed method (Method II), sub-patch algorithm is used to detect moving targets in SAR images. Sub-patch algorithm is a well known algorithm mainly used for image focusing applications for increasing the resolution of the low resolution images. In the existing SAR applications, sub-patch algorithm is used to enhance the resolution of a SAR image. But in this thesis, it is shown that sub-patch algorithm can be used to detect moving targets within the SAR images besides the enhancement of the SAR image resolution.Method II is capable of detecting targets moving in both range and azimuth directions. In the algorithm, four different subpatch images are formed from one whole scene image. By forming four different subpatch images of the same scene, the information for moving target detection process is increased; therefore moving targets can easily be detected.Both of the two proposed methods can be used in parallel processing applications. The moving target detection process can be run independently for azimuth and range directions. By using a modern parallel processor, the same input data can be used for analysing azimuth direction in one processor core and analysing range direction at the same time in the second processor core. After these two analyses completed, the results can be combined, and the target motion parameters can be successfully obtained. Moreover, detection performance and motion parameter estimation accuracy is high because of the non-sequential processing of range and azimuth direction motion detection.The performances of the proposed algorithms are tested in computer simulation. The simulations of the proposed algorithms are completed by considering single channel SAR system and using spotlight mode raw data. A good clutter cancellation process is applied before starting the algorithm steps. After all moving targets have been detected; number of detected targets, their movement directions and their motion parameters are reported.In the simulation scenarios, moving and stationary targets are put to the simulation scene data. The simulation scenes contain both stationary and moving targets. The motion parameters of the moving target are completely controllable and in the simulation scenarios, moving target is put into the simulation scene with different velocities. Therefore, the performances of the proposed algorithms are completely tested for different magnitude of target velocities and different moving direction of the target. The simulation results are provided in the simulation results section of the thesis.By using Method I, target moving not only in azimuth direction, but also in range direction (in different words, target moving in any direction) was detected successfully. Target moving in azimuth direction can be detected and its motion direction in azimuth can be found successfully and its azimuth direction velocity can be calculated within 10% error range. Target moving in range direction can be detected, its motion direction in range can be found successfully. But its velocity can?t be calculated within a sufficient accuracy, therefore additional work needed to improve the range direction velocity estimation of the moving target.In Method II, movement of the target in both azimuth and range directions can be detected successfully. In the application of this proposed method, four sub-patch images are generated from the main SAR image. The number of the sub-patch can be chosen according to the requirements of the application. In simulations used in this thesis analyzing these four images, performance of the detection process of the moving target is increased and satisfactory results could be obtained. As in the Method I, moving target detection calculations can be processed independently in azimuth and range directions and high accurate results are obtained.By using Method II, target moving not only in azimuth direction, but also in range direction was detected successfully. Target moving in azimuth direction can be detected, its motion direction in azimuth can be found successfully. Also, target moving in range direction can be detected and its motion direction in range can be found successfully.As a result, Method I and Method II can be used as quick and accurate algorithms to detect the presence of the moving targets within an observation scene. The algorithm results can be sent to more accurate algorithms for more detailed extraction of the parameters of detected moving targets.