DE102016015740B4 - Method for creating an earth observation image of a region using a synthetic aperture radar - Google Patents

Abstract

Method for creating an earth observation image of a region using a synthetic aperture radar (SAR) and a phase-controlled group antenna composed of individual antennas with an azimuthal swivel angle range that can be achieved by a control angle, each individual antenna having an element factor describing the individual antenna gain as a function of the solid angle and the group antenna having an array factor which describes the group antenna gain resulting from the arrangement and number of individual antennas as a function of the solid angle, the group antenna gain having several main maxima which are essentially the same size and which, however, with regard to the overall gain, which results from the element factor and the group antenna gain results, are attenuated towards increasing control angles θ, with the method - the earth observation image is created using multi-look processing, i.e. based on the processing of an overall image of a region which is composed of several created by illuminating over different viewing angle ranges Partial recordings of the region, i.e. looks (k), whereby the partial recordings are each filtered out of the overall recording, whereby - portions of the synthetic aperture that have a signal-to-noise ratio (SNR) at certain viewing angles that is smaller than a predeterminable one Limit value, can be made usable by a signal-to-noise-dependent summation of the individual looks (k), whereby the summation is derived either- location-independent directly from the azimuth antenna diagram by a normalized correction of the antenna diagram (LNPC) in the individual looks ( k), whereby the antenna diagram is first corrected and then normalization is carried out in each look so that at a certain azimuth angle or, if the operation is carried out in the spectrum, at a certain azimuth frequency specific to each look (k), the product of inverse Antenna diagram and normalization results in a value of one in each look, or - depending on the location, a different signal-to-noise-dependent weighting Ωm[k] of the looks (k) is carried out at each pixel position (P), whereby the weighting function (ALNPC, k(fa)) is created according to the desired optimization criteria.

Description

The invention relates to a method for creating an earth observation image of a region using a radar with synthetic aperture (SAR) and a phase-controlled group antenna composed of individual antennas with an azimuthal swivel angle range that can be achieved by a control angle, each individual antenna having an element factor describing the individual antenna gain as a function of the solid angle and the group antenna has an array factor which describes the group antenna gain resulting from the arrangement and number of individual antennas as a function of the solid angle, the group antenna gain having several main maxima which are essentially the same size and which, however, with regard to the overall gain resulting from the Element factor and the group antenna gain are attenuated towards increasing control angles. In particular, the invention relates to a method for improving resolution by expanding the usable swivel angle range of a phased array antenna for synthetic aperture radar (SAR) (hereinafter also referred to as wrapped staring spotlight) and for improving SAR image quality in parts of the synthetic aperture with poor SNR (hereinafter also with LNPC or omega weighting).

Staring Spotlight [1] is the SAR imaging mode (Synthetic Aperture Radar SAR) with the longest duration of illumination of a target and therefore also with the best resolution. The antenna is constantly aimed at the center of the scene to be imaged as the radar flies past (see 1 ). The duration of illumination and thus the azimuthal resolution is determined by the maximum azimuthal swivel angle range of the antenna.

The antenna is very often implemented as a phase-controlled group antenna, which allows rapid changes in the direction of view of the antenna without requiring the position of the carrier or influencing it. The direction of view is aligned electronically using excitation coefficients for the respective individual antennas. For each control angle, a set of excitation coefficients is typically stored on board the carrier. This means that the number or quantization of the adjustable control angles is limited. As a rule, the distance between the individual antennas that can be addressed by excitation coefficients is greater than half the wavelength, which leads to the appearance of grating lobes.

A state-of-the-art SAR system with staring spotlight imaging mode and phased array antenna is TerraSAR-X (TSX) [1]. In this exemplary system, the on-board tabulated excitation coefficients cover a range of azimuthal control angles between -2.2° and +2.2°. This angular range of 4.4° corresponds to the angular distance Δθ _GL between the main maxima of the azimuth antenna diagram. The main maxima are the lattice lobes and the main lobe. It is possible to expand the swivel angle range by distributing the control angles differently, but this leads to an increase in quantization and thus to loss of quality.

1 shows the oblique flight of a carrier past the spotlight scene to be recorded from an observation point that is opposite the flight path of the carrier so that the spotlight scene on the ground is in between. The carrier flies laterally over the scene to be recorded at the speed V _s . The azimuthal viewing direction of the antenna is guided so that it is constantly directed towards the center of the scene, i.e. it “stares” there. The in 1 The swivel angle range shown corresponds to the Δθ _GL of the TSX system. The antenna lobes shown in solid lines point to the control angles set by the excitation coefficients.

A technique that is used to support the wrapped starring spotlight mode described below is the diverging PRF setting [1],[2]. In SAR systems, the signal, which in principle is not band-limited, is sampled in azimuth at the pulse repetition frequency (PRF). Signals from directions that do not correspond to azimuth frequencies within the PRF band are also reflected into the PRF band by folding back. By specifically selecting the PRF used, the positioning of an undesirable controlled or folded-back main maximum can now be adjusted relative to a desired controlled or folded-back main maximum. Diverging PRF setting now means setting the position of the desired and undesired main maxima as separated from each other as possible. Bandpass filtering can then be used to remove as much of the unwanted main maximum as possible. 2 shows in a thick, solid line a desired main maximum centered in the middle of the PRF band, on the left with two unwanted main maximums in a thin dashed line in the Converging PRF setting, i.e. with these also centered in the middle of the PRF band. In the right illustration the undesirable main maxima are in the diverging PRF setting.

Conventional SAR processing in the azimuthal dimension carries out, among other things, a correction of the azimuth antenna diagram in the azimuth spectrum and then a weighting of each look in azimuth to control or suppress the side lobes of the impulse response. The side lobe suppression can also be located in the azimuth spectrum. For the processing of data from the exemplary TSX system, a “General Cosine Window” [1] with an α parameter of 0.6 is assumed below. This reduces the geometric resolution by a factor of 1.3 compared to processing without side lobe suppression. The maximum geometric azimuth resolution so far for the exemplary TSX system in the Staring Spotlight imaging mode, which results from the azimuth swivel angle range of 4.4°, is 0.165 m without spectral weighting for side lobe suppression and 0.225 m with side lobe suppression [1]. The largest sidelobe ratio related to the maximum of the impulse response (Peak Sidelobe Ratio - PSLR) is -13.2 dB without sidelobe suppression and -31.6 dB with suppression. There are two basic ways of visually representing the available azimuthal bandwidth: in maximum geometric resolution or by dividing the bandwidth into several views (“looks”) and adding them incoherently, which results in a representation with reduced speckle and degraded geometric resolution. Speckle is a fluctuation in brightness inherent in all SAR images that makes target detection difficult. Looks are often generated overlapping in SAR images and a common measure to quantify speckle reduction is the Equivalent Number of Looks (ENL).

Another important quality parameter of SAR images is the proportion of noise in the signal, which can be described by the signal-to-noise ratio (SNR). The dominant part of the noise is usually the thermal noise at the radar receiver. A high proportion of noise in the SAR signal appears in the SAR image like a gray haze that covers the image. A high proportion of noise also reduces the detectability of targets.

Both speckle and thermal noise fluctuate the measured radar backscatter cross section. For area targets, ENL and SNR can be combined into one quality parameter and the detectability in the SAR image can be described as “radiometric resolution (RR)”. $$\text{RR}=1+\frac{1+1/\text{SNR}}{\sqrt{\text{ENL}}}.$$

wobei N_p die Anzahl von Bildpixeln mit Grauwerten von 0 bis 255 ist und g_p der Grauwert eines Pixels. Je kleiner C_rms ist, umso ähnlicher ist das Bild einem einzigen, idealen, das ganze Bild ausfüllende Flächenziel ohne Speckle und ohne Rauschen. Je größer C_rms für eine Szene mit gemischten Zielen, also Flächenzielen, urbane Gebieten und Punktzielen ist, umso weniger Rauschanteil ist vorhanden. Der maximale Kontrast ergibt sich, wenn die Hälfte der Pixel maximale und die andere Hälfte minimale Graustufe aufzeigen. Der maximale Kontrast entspricht der Hälfte der höchsten Graustufe, also 127,5 in Gleichung (1). Die Darstellung der zurückgestreuten Radarleistung erfolgt wegen des günstigeren Dynamikbereiches oft in der Wurzel der Intensität, also in Amplitude. Im Folgenden wird der Grauwert Null der Amplitude Null zugeordnet und der Grauwert 255 der Wurzel des 95. Perzentils der Intensität.Contrast is a helpful measurement for comparing differently displayed SAR images. It provides conclusions about a change in the noise component and describes the detectability of scenes with heterogeneous backscatter cross-sections, such as those that occur in urban areas. There is no uniform definition of contrast for SAR. The following definition of the “Root Mean Square Contrast” C _rms is used, which is similarly defined for optical images: $$\begin{array}{ccc}{\text{C}}_{\text{ms}}=\sqrt{\frac{1}{{\text{N}}_{\text{p}}}\cdot {\displaystyle \sum _{\text{p}=1}^{{\text{N}}_{\text{p}}}\left({\text{G}}_{\text{p}}-\overline{\text{G}}\right)}};& \overline{\text{G}}=\frac{1}{{\text{N}}_{\text{p}}}\cdot {\displaystyle \sum _{\text{p}=1}^{{\text{N}}_{\text{p}}}{\text{G}}_{\text{p}};}& {\text{G}}_{\text{p}}\in \left[\mathrm{0,}\dots \mathrm{,255}\right]\end{array},$$

where N _p is the number of image pixels with gray values from 0 to 255 and g _p is the gray value of a pixel. The smaller C _rms is, the more similar the image is to a single, ideal area target that fills the entire image without speckle and without noise. The larger C _rms is for a scene with mixed targets, i.e. area targets, urban areas and point targets, the less noise there is. The maximum contrast occurs when half of the pixels have a maximum gray level and the other half have a minimum gray level. The maximum contrast corresponds to half of the highest gray level, i.e. 127.5 in equation (1). Because of the more favorable dynamic range, the backscattered radar power is often represented in the root of the intensity, i.e. in amplitude. In the following, the gray value zero is assigned to the amplitude zero and the gray value 255 to the root of the 95th percentile of the intensity.

The representation of the amplitude in 3 from the value 0 to the root of the 95th percentile of all intensity values corresponds linearly to the gray values from 0 to 255. In Table 1, the value of the root is given as 11.76. The measured contrast of the whole in 3 shown area is 56.5 and for area D of the urban area is 75.2. A surface target should ideally show the lowest possible contrast, i.e. low speckle and thermal noise. For the three area targets A, B and C the contrast is in table 1, as well as the measured ENL values and the mean SNR. From the last two, according to equation (1), the radiometric resolution is approximately 2.3 for the three area targets.

1. a) In Staring Spotlight SAR ist entweder der azimutale Steuerwinkelbereich oder die Quantisierung der Steuerwinkel durch den verfügbaren On-board-Speicher begrenzt. Durch den bei Beibehaltung der Quantisierung begrenzten Steuerwinkelbereich ist die maximal mögliche geometrische Auflösung begrenzt. Bei Beibehaltung einer geometrischen Auflösung ist die radiometrische Auflösung begrenzt.
2. b) Wird bei SAR-Akquisitionen der Winkelbereich so weit ausgedehnt, dass auch Winkel in die synthetische Apertur eingehen, unter denen der Antennengewinn schlecht ist, dann erhöht sich das thermische Rauschen bei konventioneller SAR-Prozessierung stark und das SNR im SAR-Bild ist schlecht. Daher werden diese Winkelbereiche nicht verarbeitet und deren Informationspotential bleibt ungenutzt.

Table 1: Image quality parameters for the section of a staring spotlight image from the TerraSAR-X system.

1. a) In Staring Spotlight SAR, either the azimuthal control angle range or the quantization of the control angles is limited by the available on-board memory. The maximum possible geometric resolution is limited due to the limited control angle range while maintaining the quantization. While maintaining a geometric resolution, the radiometric resolution is limited.
2. b) If the angular range in SAR acquisitions is expanded to such an extent that the synthetic aperture also includes angles at which the antenna gain is poor, then the thermal noise in conventional SAR processing increases greatly and the SNR in the SAR image is poor . Therefore, these angular ranges are not processed and their information potential remains unused.

The object of the invention is to improve the resolution and image quality in SAR systems.

• - das Erdbeobachtungsbild anhand einer Multi-Look-Verarbeitung, d.h. anhand der Verarbeitung einer Gesamtaufnahme einer Region erstellt wird, die zusammengesetzt ist aus mehreren Teilaufnahmen (sogenannte Looks) der Region durch Beleuchten über jeweils unterschiedliche Blickwinkelbereiche hinweg, wobei die Teilaufnahmen jeweils aus der Gesamtaufnahme herausgefiltert werden, wobei
• - Anteile der synthetischen Apertur, die bei bestimmten Blickwinkeln ein Signal-zu-Rausch-Verhältnis aufweisen, das kleiner ist als ein vorgebbarer Grenzwert und damit „schlecht“ ist, weshalb diese Anteile der synthetischen Apertur bei bisherigen Verfahren nicht genutzt wurden, nutzbar gemacht werden durch eine Signal-zu-Rausch-abhängige Aufsummierung der einzelnen Looks, wobei die Aufsummierung entweder
  • - ortsunabhängig direkt aus dem Azimut-Antennendiagramm abgeleitet wird durch eine normalisierte Korrektur des Antennendiagramms in den einzelnen Looks (LNPC), wobei zunächst das Antennendiagramm korrigiert und danach in jedem Look eine Normierung so durchgeführt wird, dass bei einem bestimmten Azimutwinkel oder, falls die Operation im Spektrum durchgeführt wird, bei einer bestimmten, für jeden Look spezifischen Azimutfrequenz das Produkt aus inversem Antennendiagramm und Normierung in jedem Look einen Wert von eins ergibt,
  oder
• - ortabhängig durch eine an jeder Pixelposition unterschiedliche Signal-zu-Rausch-abhängige Gewichtung der Looks (Ω-Weighting) durchgeführt wird, wobei die Gewichtungsfunktion nach gewünschten Optimierungskriterien erstellt wird.

To solve this problem, the invention proposes a method for creating an earth observation image of a region using a radar with a synthetic aperture (SAR) and a phase-controlled group antenna composed of individual antennas with an azimuthal swivel angle range that can be achieved by a control angle, each individual antenna having the individual antenna gain in Dependence on the element factor describing the solid angle and the group antenna has an array factor which describes the group antenna gain resulting from the arrangement and number of the individual antennas as a function of the solid angle, the group antenna gain having several main maxima which are essentially the same size and which, however, with regard to the Total gain, which results from the element factor and the group antenna gain, is attenuated towards increasing control angles, with the method

• - the earth observation image is created using multi-look processing, ie based on the processing of an overall image of a region, which is composed of several partial images (so-called looks) of the region by illuminating them over different viewing angles, with the partial images being filtered out of the overall image be, whereby
• - Portions of the synthetic aperture that, at certain viewing angles, have a signal-to-noise ratio that is smaller than a predeterminable limit value and are therefore “bad,” which is why these portions of the synthetic aperture were not used in previous methods can be made usable by a signal-to-noise-dependent summation of the individual looks, whereby the summation is either
  • - is derived independently of location directly from the azimuth antenna diagram by a normalized correction of the antenna diagram in the individual looks (LNPC), whereby the antenna diagram is first corrected and then normalization is carried out in each look in such a way that at a certain azimuth angle or, if the operation is carried out in the spectrum, at a certain azimuth frequency specific to each look, the product of the inverse antenna diagram and normalization results in a value of one in each look,
  or
• - Is carried out depending on the location by a signal-to-noise-dependent weighting of the looks (Ω-weighting) that is different at each pixel position, with the weighting function being created according to the desired optimization criteria.

• - eine möglichst gute radiometrische Auflösung und/oder
• - eine möglichst gutes Signal-zu-Rausch-Verhältnis im Multi-Look-Bild und/oder
• - eine möglichst hohe Anzahl Equivalent Number of Looks handelt/handeln kann.

Further possible variants of the invention for the location-dependent weighting of the looks can be seen in the fact that the optimization criteria for creating the weighting function are

• - the best possible radiometric resolution and/or
• - the best possible signal-to-noise ratio in the multi-look image and/or
• - the highest possible number of Equivalent Number of Looks trades/can trade.

A further aspect of the invention can be seen in the fact that recording in spotlight mode and thus by extending the illumination of the region to be recorded as a result of the expansion of the azimuthal swivel angle range, or in TOPSAR mode by extending the azimuthal swivel angle range, or in bi-directional SAR mode or in forward/backward looking SAR mode.

• 1 eine grafische Darstellung des bekannten Staring Spotlight Modus mit den maximalen azimutalen Steuerwinkeln des TerraSAR-X Systems (TSX),
• 2 eine grafische Darstellung des Converging PRF-Setting (links) und Diverging PRF-Setting (rechts),
• 3 ein beispielhafter Ausschnitt aus einem Staring-Spotlight-Bild des TerraSAR-X Systems (TSX), wobei drei Flächenziele (A, B und C) und eine urbane Fläche (D) mit Bebauung markiert sind,
• 4 eine schematische Darstellung des Staring Spotlight Modus mit erweitertem azimutalen Steuerwinkelbereich durch Ausnutzen der Hauptmaxima am Beispiel des TerraSAR-X Systems (TSX),
• 5 eine Auswahl der gewünschten Signalanteile bzw. die Unterdrückung der unerwünschten Signalanteile, durch zeitabhängige Bandpass-Filterung und Diverging PRF-Setting, wobei die angegebenen Winkel beispielhaft für das TerraSAR-X System (TSX) sind,
• 6 das prinzipielle Blockschaltbild der erfindungsgemäßen LNPC Multi-look-Verarbeitung anhand des range komprimierten und Azimut phasenkorrigierten vollständigen Azimut-Spektrums,
• 7 das prinzipielle Blockschaltbild der erfindungsgemäßen Ω-Weighting-Multilook-Verarbeitung anhand des range komprimierten und Azimut phasenkorrigierten vollständigen Azimut-Spektrums,
• 8 eine grafische Darstellung des Steuerwinkels der linken TerraSAR-X Wrapped-Staring-Spotlight-Aufnahme gemäß 9, wobei in dünner Linie der kommandierte, sich wiederholende, also „gewrappte“ Steuerwinkel, und in dicker Linie der „unwrappte“ Steuerwinkel, also der durch Filterung und Diverging-PRF-Setting nutzbar gemachte und letztendlich prozessierte Steuerwinkel dargestellt ist,
• 9 beispielhafte Wrapped-Staring-Spotlight-Bilder des TerraSAR-X Systems, wobei links eine Szene mit urbanen Gebieten und Flächenzielen gezeigt und ein Ausschnitt für eine in der Figurenbeschreibung beschriebene Auswertung markiert ist und wobei rechts eine Szene mit einem markierten Punktziel gezeigt ist, das in der Figurenbeschreibung genauer analysiert wird,
• 10 links die Azimutauflösung, gemessen für das in 9 ausgewählte Punktziel, wobei in gestrichelt die Messwerte für Element Pattern Weighting (EPW), als durchgezogene Linie die theoretischen/ simulierten Werte und gepunktet dazu im Vergleich die theoretischen/ simulierten Auflösungswerte für eine vollständige Korrektur des Antennendiagramms gezeigt sind und wobei rechts die gemessenen und simulierten Werte für das höchste Nebenzipfelverhältnis der Impulsantwort gezeigt sind,
• 11 Azimut-Antennendiagramme und Spektren, und zwar links für LNPC und rechts für Ω-Weighting, wobei die unterschiedliche Skalierung der dargestellten Amplituden zu beachten ist, und in den Diagrammen (a) das Profil des Azimutspektrums in durchgezogener und das gerechnete Antennendiagramm in gestrichelter Linie gezeigt ist, in den Diagrammen (b) die Azimut-Antennendiagramm-Korrektur spezifisch für jeden der neun Looks gezeigt ist, die abwechselnd in durchgezogener und in gestrichelter Linie gezeigt sind, in den Diagrammen (c) das Profil des Azimutspektrums nach der Antennendiagramm-Korrektur gezeigt ist und in den Diagrammen (d) das Profil des Azimutspektrums nach Korrektur des Antennendiagramms und Gewichtung zur Nebenzipfelunterdrückung gezeigt sind,
• 12 links ein optisches Google Earth Bild des Ausschnitts gemäß 9 links und rechts das Amplitudenbild der SAR-Aufnahme bei klassischer Prozessierung von 9L_FPC, wobei die neun Looks hinsichtlich des Azimut-Antennendiagramms vollständig korrigiert sind und wobei links und rechts verschiedene Flächenziele und Zielgebiete mit gemischtem Kontext durch weiße Rahmen markiert sind,
• 13. eine grafische Darstellung der optimierten Ω-Gewichte für M=32 Klassen und ein SNR_ML,min von 0 dB, wobei die X-Achse die Look-Nummer und die Y-Achse das SNR des zentralen Looks der Klassen zeigen,
• 14 ENL, SNR und radiometrische Auflösung für die optimierten Klassen des Ausführungsbeispiels, in dicker gestrichelter Linie für ein SNR_ML,min von 0 dB und in dünner durchgezogener Line für 3 dB und
• 15 markierte Ausschnitte der in 9 links gezeigten SAR-Akquisition, wobei 3L_FPC für drei Looks mit voller Antennendiagramm-Korrektur - dies entspricht in guter Näherung dem operationell implementierten Staring-Spotlight-Modus des TerraSAR-X Systems - steht, 9L_LNPC für neun Looks mit LNPC-Multilooking steht und 9L_Ω,0dB, 9L_Ω,3dB für Ω-Weighting-Multilooking in den Varianten 0 dB und 3 dB minimal gewünschtem SNR im Multilook-Bild stehen.

The various aspects of the invention are explained in more detail below with reference to the drawing. In detail we show:

• 1 a graphical representation of the well-known Staring Spotlight mode with the maximum azimuthal control angles of the TerraSAR-X system (TSX),
• 2 a graphical representation of the converging PRF setting (left) and diverging PRF setting (right),
• 3 an exemplary section of a staring spotlight image from the TerraSAR-X system (TSX), with three area targets (A, B and C) and an urban area (D) marked with buildings,
• 4 a schematic representation of the staring spotlight mode with extended azimuthal control angle range by exploiting the main maxima using the example of the TerraSAR-X system (TSX),
• 5 a selection of the desired signal components or the suppression of the unwanted signal components, through time-dependent bandpass filtering and diverging PRF setting, whereby the specified angles are examples for the TerraSAR-X system (TSX),
• 6 the basic block diagram of the LNPC multi-look processing according to the invention based on the range compressed and azimuth phase-corrected complete azimuth spectrum,
• 7 the basic block diagram of the Ω-weighting multilook processing according to the invention based on the range compressed and azimuth phase-corrected complete azimuth spectrum,
• 8th a graphical representation of the control angle of the left TerraSAR-X wrapped staring spotlight image 9 , whereby the commanded, repeating, i.e. “wrapped” control angle is shown in a thin line, and the “unwrapped” control angle is shown in a thick line, i.e. the control angle made usable and ultimately processed through filtering and diverging PRF setting,
• 9 exemplary wrapped staring spotlight images of the TerraSAR-X system, with a scene with urban areas and area targets shown on the left and an excerpt for one in the figure description the evaluation described is marked and a scene with a marked point target is shown on the right, which is analyzed in more detail in the figure description,
• 10 on the left the azimuth resolution, measured for the in 9 selected point target, where the measured values for Element Pattern Weighting (EPW) are shown in dashed lines, the theoretical/simulated values are shown as a solid line, and the theoretical/simulated resolution values for a complete correction of the antenna diagram are shown in dotted lines in comparison, and the measured and simulated values are shown on the right are shown for the highest sidelobe ratio of the impulse response,
• 11 Azimuth antenna diagrams and spectra, namely on the left for LNPC and on the right for Ω-weighting, whereby the different scaling of the amplitudes shown must be taken into account, and in diagrams (a) the profile of the azimuth spectrum is shown in a solid line and the calculated antenna diagram in a dashed line in diagrams (b) the azimuth antenna pattern correction is shown specifically for each of the nine looks shown alternately in solid and dashed lines, in diagrams (c) the profile of the azimuth spectrum after antenna pattern correction is shown and in diagrams (d) the profile of the azimuth spectrum is shown after correction of the antenna diagram and weighting for sidelobe suppression,
• 12 On the left is an optical Google Earth image of the section 9 left and right the amplitude image of the SAR recording with classic processing of 9L _FPC , where the nine looks are fully corrected with regard to the azimuth antenna diagram and where different area targets and target areas with mixed context are marked by white frames on the left and right,
• 13 . a graphical representation of the optimized Ω weights for M=32 classes and an SNR _ML,min of 0 dB, where the x-axis shows the look number and the y-axis shows the SNR of the central look of the classes,
• 14 ENL, SNR and radiometric resolution for the optimized classes of the exemplary embodiment, in thick dashed line for an SNR _ML,min of 0 dB and in thin solid line for 3 dB and
• 15 marked sections of the in 9 SAR acquisition shown on the left, where 3L _FPC stands for three looks with full antenna pattern correction - this corresponds to a good approximation of the operationally implemented staring spotlight mode of the TerraSAR-X system -, 9L _LNPC stands for nine looks with LNPC multilooking and 9L _Ω,0dB , 9L _Ω,3dB for Ω-weighting multilooking in the variants 0 dB and 3 dB minimum desired SNR in the multilook image.

Embodiment not according to the invention

In an existing SAR system with a phase-controlled array antenna, the recording mode described below, which is not part of the invention (hereinafter referred to as wrapped staring spotlight mode), significantly expands the swivel angle range compared to the conventional staring spotlight recording mode, without any additional on-board -Storage space requirement and without a deterioration in the quantization of the control angles. The prerequisite is that the angular distance Δθ _GL between the main maxima is covered in sufficient quantization of the control angles. The expansion is then carried out by repeatedly adjusting the control angle within Δθ _GL , but not the signal components of the control direction of the synthetic aperture actually set by the excitation coefficients, but those signal components that come from the desired direction of another main maximum of the antenna diagram are added.

4 again shows the recording geometry of the exemplary TSX system, but for a wrapped staring spotlight recording. The middle three antenna positions are already in 1 shown for Staring Spotlight mode. The two outer antenna positions in 4 now show the expansion of the swivel angle range. In order to record signals from the directions ± 4.4°, the excitation coefficients for 0° control angle are set again at the two positions. The desired control angles are then made usable through diverging PRF setting and corresponding filtering in the processing. For point targets, the extended control angle range directly improves the geometric resolution.

The selection of the desired signal components or the suppression of the undesired ones is carried out using time-dependent bandpass filtering. This process is in 5 illustrated. During the acquisition, the desired control angle is set at any time directly or by exploiting an ambiguous (“wrapped”) main maximum. This is in 5 shown by the thick line drawn in dash-dots, which varies continuously from +4.4° control angle to -4.4° over the azimuth time t _a , ie the recording time Control angle drops. The vertical extent of the band drawn around this line in a thin solid line around this control angle is the so-called scene angle, which indicates the extent of the processed scene in the azimuth direction. It is set either by an azimuth-time-dependent bandpass filter during processing, or equivalently after processing by selecting the scene extent in the finished processed image. Specifically, the azimuth time-dependent filter can be implemented, for example, through subaperture processing with different center frequencies in the subapertures [3]. The variation of the desired control angle over the azimuth acquisition time determines the maximum geometric resolution and is referred to as the unwrapped resolution angle. The main maxima shown in dashed lines in the upper right and lower left parts of the illustration are shifted from the desired ones by the angle Δθ _GL . The gain ratios shown correspond qualitatively to those of the exemplary TerraSAR-X system. The bottom row of diagrams represents the relative positioning of the desired and strongest unwanted main maxima in the PRF band for diverging PRF setting. For the control angles of ± 2.2°, the desired and unwanted main maxima are the same size, but due to the diverging PRF -Setting separated from each other. The unwanted main maxima can be suppressed by the scene angle corresponding to a bandpass. For the very large control of up to ± 4.4° in the exemplary TerraSAR-X system, the gain values from the desired control angle directions are significantly lower than those from undesired directions. Even when using diverging PRF setting, the remaining gain from these directions is greater than that from the desired directions.

In an existing SAR system with a phase-controlled group antenna and a distance between the individual antennas that can be addressed by excitation coefficients greater than half the wavelength and on-board stored excitation coefficients for antenna control in azimuth, the wrapped staring spotlight recording mode extends the swivel angle range compared to the conventional staring spotlight recording mode significantly. This is done through additional utilization of main maxima, which form next to the main maximum in the actually set main maximum direction, without additional on-board storage space requirements and without a deterioration in the quantization of the control angles. Diverging PRF setting during data acquisition supports the selection of the desired signal components or the suppression of the unwanted signal components, which is done through time-dependent bandpass filtering. The added value for point targets is the improvement in geometric resolution associated with the expansion of the swivel angle range. The added value for area targets and mixed image content is improved radiometric resolution and contrast in multi-look images.

The core of the wrapped staring spotlight method can also be used for other SAR modes, such as bi-directional SAR or forward/backward-looking SAR, i.e. azimuth viewing directions can be evaluated in existing SAR systems are not within the originally designed control angle range.

Embodiments according to the invention

The SAR images resulting from the formation of the synthetic aperture taking into account control angle ranges with poor SNR can be compared with conventional SAR processing in terms of the two quality parameters “contrast” and “radiometric resolution” by the weighted multi according to the invention, which is dependent on the SNR of the individual looks -Look processing can be significantly improved. Two main variants of the novel multi-look processing are given below as an exemplary embodiment, namely Look Normalized Antenna Pattern Correction (LNPC) with a weighting of the individual look images that is constant across the pixels and therefore location-independent, and Q- Weighting (English: Ω-Weighting) with a pixel-dependent and therefore location-variant weighting of the look images. LNPC is easier to implement. Ω-Weighting is more flexible and can be better tailored to the detection of specific targets.

LNPC variant

LNPC weights the individual azimuth look images differently when added to the multilook image. The different weighting results from a normalized correction of the azimuth antenna diagram. For example, the normalization can be related to the mean azimuth frequency of the look in each look, i.e. for the mean azimuth frequencies of the looks the correction would then be 1. Each pixel in the resulting multi-look image has exactly the same weighted shares of the individual looks. LNPC is therefore location invariant.

6 shows the basic block diagram of the LNPC multi-looking. The starting point is the complete azimuth spectrum, which is range-compressed and fully phase-corrected in azimuth, including azimuth modulation correction.

What is important for LNPC multi-looking is the spectral correction of the antenna diagram by the amplitude weighting function A _LNPC,k (f _a ) in equation (3), which is dependent on the azimuth frequency f _a and the number of the look. The amplitude of the full spectrum two-way azimuth antenna pattern is labeled AAP _2W (f _a ). The central Doppler frequency (Doppler centroid) of the look k is denoted by f _c,k . The rectangle function rect selects the look bandwidth B _L centered around f _c, k for each look k from the total spectrum. The antenna diagram is corrected for all frequencies f _a but normalized to the value 1 at the central Doppler frequency by multiplying by AAP _2W (f _c,k ). $${\text{A}}_{\text{LNPC},\text{k}}\left({\text{f}}_{\text{a}}\right)=\frac{{\text{AAP}}_{2\text{W}}\left({\text{f}}_{\text{c},\text{k}}\right)}{{\text{AAP}}_{2\text{W}}\left({\text{f}}_{\text{a}}\right)}\cdot \text{rect}\left[\frac{{\text{f}}_{\text{a}}-{\text{f}}_{\text{c},\text{k}}}{{\text{b}}_{\text{L}}}\right]$$

At the same point in the process of 6 , at which the multiplication with A _LNPC,k (f _a ) takes place, the desired weighting for sidelobe suppression is also carried out by a multiplication with M _L,k (f _a ). This weighting function spans the bandwidth B _L and is shifted for each look to be centered around f _c,k . According to the inverse Fast Fourier Transform (IFFT) in 6 For each image pixel p there are k look pixels with the intensities I _p,k . These are then summed according to the rule in equation (4) to give the multi-look pixel intensity I _p,ML . $${\text{I}}_{\text{p},\text{ML}}=\frac{{\text{C}}_{\text{cal}}}{{\displaystyle \sum _{\text{k}=1}^{{\text{N}}_{\text{L}}}\left({\text{w}}_{\text{I},\text{s}}\left[\text{k}\right]\cdot {\displaystyle \underset{{\text{f}}_{\text{c},\text{k}}-{\text{b}}_{\text{L}}/2}{\overset{{\text{f}}_{\text{c},\text{k}}+{\text{b}}_{\text{L}}/2}{\int }}{\left[{\text{AAP}}_{2\text{W}}\left({\text{f}}_{\text{A}}\right)\right]}^2\text{df}}\right)}}\cdot {\displaystyle \sum _{\text{k}=1}^{{\text{N}}_{\text{L}}}{\text{I}}_{\text{p},\text{k}}}$$

C _cal is an arbitrary calibration constant that combines all necessary but not required constants for the description of the procedure. Because of the different weighting of the look pixels in the summation to the multi-look pixel, the sum must be normalized by the denominator in equation (4). Here W _I,S [k] is a factor for look k defined in equation (5), which describes the ratio of the intensity of a look pixel before and after the spectral LNPC weighting and the weighting for side lobe suppression. The integral over the look bandwidth B _L is centered at f _c,k , and includes the weighting of the intensity by the azimuth antenna pattern acting on it over the look bandwidth k. $${\text{w}}_{\text{I},\text{S}}\left[\text{k}\right]=\frac{{\displaystyle \underset{{\text{f}}_{\text{c},\text{k}}-{\text{b}}_{\text{L}}/2}{\overset{{\text{f}}_{\text{c},\text{k}}+{\text{b}}_{\text{L}}/2}{\int }}{\left[{\text{W}}_{\text{L},\text{k}}\left({\text{f}}_{\text{a}}\right)\cdot {\text{AAP}}_{2\text{W}}\left({\text{f}}_{\text{c},\text{k}}\right)\right]}^2{\text{df}}_{\text{a}}}}{{\displaystyle \underset{{\text{f}}_{\text{c},\text{k}}-{\text{b}}_{\text{L}}/2}{\overset{{\text{f}}_{\text{c},\text{k}}+{\text{b}}_{\text{L}}/2}{\int }}{\left[{\text{AAP}}_{2\text{W}}\left({\text{f}}_{\text{a}}\right)\right]}^2{\text{df}}_{\text{a}}}}$$

Ω-weighting variant

The name Ω-Weighting is intended to make it clear that this weighting is the last one applied to the individual looks, and only in the look image area. As in 7 shown, the correction of the antenna diagram and the weighting for sidelobe suppression take place beforehand, still in the azimuth spectral range. The correction of the azimuth antenna diagram is carried out, as shown in equation (6), entirely by multiplying the amplitude function A _FPC,k (f _a ) on the amplitude spectrum. The index FPC stands for full pattern correction. $${\text{A}}_{\text{FPC},\text{k}}\left({\text{f}}_{\text{a}}\right)=\frac{1}{{\text{AAP}}_{2\text{W}}\left({\text{f}}_{\text{a}}\right)}\cdot \text{rect}\left[\frac{{\text{f}}_{\text{a}}-{\text{f}}_{\text{c},\text{k}}}{{\text{b}}_{\text{L}}}\right]$$

With Ω-weighting, the multi-look summation of the intensities of the individual look pixels I _p,k,FPC occurs differently for each pixel p over the looks k as described in equation (7). The determination of the individual weights Ω _p,k depends on the SNR of the individual look pixels and is described further below. $$\text{ML},\mathrm{\Omega }=\frac{{\text{C}}_{\text{cal}}}{{\displaystyle \sum _{\text{k}=1}^{{\text{N}}_{\text{L}}}\left({\mathrm{\Omega }}_{\text{p},\text{k}}\cdot {\text{w}}_{\text{I},\text{S},\text{FPC}}\left[\text{k}\right]\cdot {\displaystyle \underset{{\text{f}}_{\text{c},\text{k}}-{\text{b}}_{\text{L}}/2}{\overset{{\text{f}}_{\text{c},\text{k}}-{\text{b}}_{\text{L}}/2}{\int }}{\left[{\text{AAP}}_{2\text{W}}\left({\text{f}}_{\text{a}}\right)\right]}^2{\text{df}}_{\text{a}}}\right)}}\cdot {\displaystyle \sum _{\text{k}=1}^{{\text{N}}_{\text{L}}}\left({\mathrm{\Omega }}_{\text{p},\text{k}}\cdot {\text{I}}_{\text{p},\text{k},\text{FPC}}\right)}$$

W _I,S,FPC [k] is the factor for look k defined in equation (8), which describes the ratio of the intensity of a look pixel before and after the spectral full correction of the antenna pattern and the weighting for side lobe suppression. $${\text{w}}_{\text{I},\text{S},\text{FPC}}\left[\text{k}\right]=\frac{{\displaystyle \underset{{\text{f}}_{\text{c},\text{k}}-{\text{b}}_{\text{L}}/2}{\overset{{\text{f}}_{\text{c},\text{k}}+{\text{b}}_{\text{L}}/2}{\int }}{\left[{\text{W}}_{\text{L},\text{k}}\left({\text{f}}_{\text{a}}\right)\right]}^2{\text{df}}_{\text{a}}}}{{\displaystyle \underset{{\text{f}}_{\text{c},\text{k}}-{\text{b}}_{\text{L}}/2}{\overset{{\text{f}}_{\text{c},\text{k}}+{\text{b}}_{\text{L}}/2}{\int }}{\left[{\text{AAP}}_{2\text{W}}\left({\text{f}}_{\text{a}}\right)\right]}^2{\text{df}}_{\text{a}}}}$$

Determination of the weights Ω _p,k

• i) dass gegenüber einer konventionellen SAR-Verarbeitung eine verbesserte radiometrische Auflösung zeigt und
• ii) bei dem sich der Bildkontrast durch die Erhöhung des thermischen Rauschens nicht verschlechtert bzw. sogar verbessert.

The objective of optimizing the weights is to generate a multi-look image including looks with poor SNR,

• i) that shows improved radiometric resolution compared to conventional SAR processing and
• ii) in which the image contrast does not deteriorate or even improves due to the increase in thermal noise.

A simple solution would be to derive the weight Ω _p,k directly from the measured SNR of pixel p in look k, i.e. directly from S̃NR _p,FPC [k]. The tilde indicates a measured value. The disadvantage is that look pixels with high measured intensity are always heavily weighted and pixels of low intensity are always lightly weighted, which increases the intensity fluctuations induced by the speckle, thus increasing the speckle. This solution is therefore ruled out.

To limit the influence of speckle on the weights, Ω-weighting uses the information from all looks of an image pixel. As in the basic block diagram of 7 shown, the method according to the invention defines classes of pixel SNR curves over the looks using the antenna diagram AAP _2W (f _a ) and the measured SNR values of all pixels and looks S̃NR _p,FPC [k]. Then the weights Ω _m [k] for each class m are optimized so that the radiometric resolution is as small as possible for a given minimum required SNR in the multi-look image SNR _ML,min . If SNR _ML,min is not reached for a class, all weights of that class are set to zero. Finally, for each multi-look image pixel to be formed, the associated class m is determined using the SNR curve of the measured look pixels S̃NR _p,FPC [k] and the weighting Ω _m [k] of this class in the multi-look summation of equation (7), that is, Ω _p,k [k] is set to Ω _m [k].

For Ω-weighting, the thermal noise is reduced for areas with a low backscatter cross-section, i.e. exactly in the areas where the thermal noise is most disturbing. For higher backscatter cross sections, more noise is tolerated, as this improves the number of looks and ultimately the radiometric resolution.

Without loss of generality, one option for defining the classes, for optimizing the weights of the classes and for assigning the SNR curve of a pixel over the looks to a class is shown in the exemplary embodiment below.

In summary, the aspects essential to the invention can be described as follows.

The variant of the LNPC - Look Normalized Antenna Pattern Correction) improves the resulting multi-look SAR images with regard to the quality parameters "contrast" and "when forming the synthetic aperture, including control angle areas with poor SNR. radiometric resolution” compared to conventional, state-of-the-art SAR processing through a new type of weighted resolution that depends on the SNR of the individual looks Multi-look processing. With LNPC, the individual look images are weighted in a constant and therefore location-independent manner based on the pixel positions.

LNPC can also be applied to other SAR modes with spatially varying SNR in the azimuthal direction, such as ScanSAR, TOPSAR, Staring Spotlight and Sliding Spotlight.

When forming the synthetic aperture, including control angle ranges with poor SNR, Ω-weighting improves the resulting multi-look SAR images in terms of the quality parameters “contrast” and “radiometric resolution” compared to conventional ones State-of-the-art SAR processing through a novel weighted multi-look processing that depends on the SNR of the individual looks. With Ω weighting, the look images are weighted depending on the location of the pixels and thus location-variant.

Ω-weighting can also be applied to other SAR modes with spatially varying SNR in the azimuthal direction, such as ScanSAR, TOPSAR, Staring Spotlight and Sliding Spotlight.

As described above, two acquisitions were carried out with the TerraSAR-X system in the wrapped staring spotlight acquisition mode not according to the invention. The angles commanded over the acquisition time for the left in 9 The recording shown is in 8th shown in thin line. The repeated setting of the control angle within Δθ _GL = 4.4° can be clearly seen. The large control angle range of 8.8° achieved by using the grid lobes is plotted in a thick black line, which plots over the thin line in the middle range from approx. 3.5 s to 10 s.

9 shows the two wrapped staring spotlight images in an overview with several looks. Azimuth processing was performed with the full acquired azimuth control angle range of 8.8°. Parts of the two images are examined in more detail below to demonstrate the advantages of the wrapped staring spotlight capture method, as well as LNPC multi-look processing and Ω-weighting multi-look processing.

The extent of the Wrapped Staring Spotlight images in 9 Corresponds in the azimuth direction to a scene angle of the size of the azimuth 3dB opening angle of the antenna, ie the projection of the opening angle onto the earth's surface at the distance of the scene center. The marked section in the left shot in 9 is in 3 shown in a multi-look representation. 3 approximately corresponds to the representation of the state-of-the-art staring spotlight imaging mode of the exemplary TSX system. The highlighted section in the right photo of 9 contains a three-surface reflector (corner reflector), which appears as a point target in the image. Both in 9 Marked sections are analyzed below and thus the advantages of the invention are demonstrated.

Point targets

The impulse response to the point target in the right section of 9 has been analyzed for increasing processed bandwidth. The bandwidth of the TerraSAR-X system in staring spotlight mode is 38 kHz. All bandwidths beyond this have been created using the Wrapped Staring Spotlight mode. The measured geometric azimuth resolution improves with increasing processed bandwidth.

The processing was carried out in azimuth without correction of the antenna pattern. This means that the weighting of the element antenna diagram is retained in the raw data recording. This weighting is referred to below as EPW (Element Pattern Weighting). As the bandwidth increases, this weighting changes from a similarity to the rectangular weighting to a weighting with strong suppression of the side lobes. This is shown in 10 in the right diagram, which shows the strongest sidelobe ratio (Peak-to-Side Lobe Ratio PSLR) of the point target response over the processed bandwidth. The PSLR values of an ideal simulation are shown in dashed lines. The values measured in the TerraSAR-X Wrapped Staring Spotlight image for the point target are shown in a solid line. The PSLR decreases from the square-wave weighting value of -13.2 dB as the bandwidth increases as the effective weighting evolves toward greater sidelobe rejection. The measured values agree very well with the simulation.

The geometric resolution on the left in 10 is measured for the EPW weighting over the bandwidth of the operationally existing staring spotlight mode of the TerraSAR-X system. She agrees very well correspond to the theoretical/simulated values. From 55 kHz there is a small deviation; for the maximum processed bandwidth of 76 kHz it is 5%, which is probably due to the high requirements for the height information of the location of the point target. However, this is a requirement for the method used for focusing in the azimuth direction. However, the increasing improvement in the measured geometric resolution shows the improvement in the geometric resolution compared to the operationally existing staring spotlight mode of the TerraSAR-X system, which delivers an azimuth resolution of 0.19 m at 38 kHz maximum bandwidth and EPW. The control angle ranges additionally developed by the wrapped staring spotlight can be successfully integrated into the synthetic aperture for point targets and thereby extend it or allow improved geometric resolution in azimuth.

For comparison, the simulated geometric resolution for a rectangle weighting (rect), i.e. for a complete correction of the azimuth antenna diagram, is also given.

Area targets and mixed image content

In order to demonstrate an improvement in image quality for area targets and for image parts with mixed target areas, the entire range of wrapped staring spotlight recordings were carried out in 9 On the left a number of nine looks with a bandwidth of 15.2 kHz each are formed. The looks overlap by 50% and each look is weighted with the sidelobe suppression weighting described above. 11 shows profiles of the azimuth spectrum for LNPC according to the invention (left) and Ω-weighting (right). The upper diagrams (a) are identical and show the profile of the spectrum and the calculated antenna diagram in a dashed line. Both the spectrum and the antenna diagram show fluctuations due to the quantization of the antenna control and the traveling pulses (ie sometimes different transmit and receive antenna diagrams due to the long signal transit time > 1/PRF). For the outer, i.e. the larger azimuth frequencies, the higher profile of the spectrum compared to the antenna diagram (pattern) shows a greatly increased proportion of noise in the actually recorded data. The calculated antenna diagram does not contain any noise.

The diagrams (b) of the 11 show the applied corrections to the antenna diagram. With LNPC, the amplitude of the correction is equal to one at the central look frequency. With Ω-Weighting, the antenna pattern is completely corrected (Full Pattern Correction FPC). The profiles of the corrected azimuth spectra point sharply upward at the ends of the outer looks, which means that the noise is “pulled up”. This effect is much more pronounced with Ω-weighting. Please note the different scaling of the amplitudes and the limitation of the amplitude shown. The lower plots (d) show the spectra after additional weighting for sidelobe suppression.

Both LNPC and Ω-weighting multilooking produce the same shape of impulse responses to a point target in all looks. This means that the signal spectra of a point target are weighted equally. In 11 The total spectra are shown, i.e. signal and noise. The differences in 11 (d) arise due to the differently “raised” noise component. This leads to the same performance values of the point target response for point targets in each look: geometric resolution and side lobe ratios.

12 shows an optical Google Earth image of the section on the left 9 Left. Right in 12 The result of a classic azimuth processing of the nine looks with complete correction of the azimuth antenna pattern and sidelobe suppression can be seen, hereinafter referred to as 9L _FPC . Different area targets and target areas with mixed contexts are marked by white frames.

Out of 12 It immediately becomes clear that the SNR in the multilook image with classic 9L _FPC processing is completely insufficient for image evaluation due to the poor SNR of the external looks. 15 shows the results for LNPC-Multilooking (9L _LNPC ) and Ω-Weighting-Multilooking in the variants 0dB minimum desired SNR in the multi-look image (9L _Ω,0dB ) and 3dB (9L _Ω,3dB ). For comparison, the variant with three looks and full azimuth antenna pattern correction (3L _FPC ) is also shown, which corresponds to a good approximation of the operationally implemented staring spotlight mode of the TerraSAR-X system. The representation of the amplitude is in all amplitude images shown 15 as described above. The azimuth resolution in all looks and in the multi-look image is 0.55m with a sampling of 0.26 m. In the distance dimension (here oblique view display) the geometric resolution is 0.58 m with a sampling of 0.22 m .

• i) Das SNR wird in jedem Look für jeden Pixel gemessen. Um den Einfluss des Speckle zu verringern, wird dabei eine gleitende Mittelwertbildung unter Einbeziehung der acht benachbarten Pixel angewendet.
• ii) Für den zentralen Look wird ein Histogramm erstellt und M Klassen werden so definiert, dass jede Klasse m in etwa gleich viele SNR-Werte enthält.
• iii) Die SNR-Werte der anderen Looks werden ausgehend vom SNR-Wert des zentralen Looks mit Hilfe des Azimut-Antennendiagramms ergänzt.
• iv) Die Optimierung wird für jede Klasse durchgeführt, so wie oben beschrieben. Es wird also für ein vorgegebenes gewünschtes minimales SNR eines Pixels des Multi-Look-Bildes SNR_ML,min eine Optimierung der Ω-Gewichte Ω_m[k] der einzelnen Looks k so durchgeführt, dass eine öglichst gute radiometrische Auflösung erzielt wird.
• v) An jeder Pixelposition wird dem Verlauf der SNR-Werte (ohne gleitende Mittelwertbildung) über den einzelnen Looks eine Klasse zugeordnet. Dies erfolgt über den minimalen Gesamtabstand zwischen dem Verlauf der gemessenen SNR-Werte und der SNR-Werte der Klassen.
• vi) Die für die Klasse optimierten Ω-Gewichte werden in der Multi-Look-Summation für die jeweilige Pixelposition laut Gleichung (7) verwendet.

The individual weights of the Ω-weighting were set as follows in this exemplary embodiment according to the invention.

• i) The SNR is measured for each pixel in each look. In order to reduce the influence of speckle, a moving average is used taking into account the eight neighboring pixels.
• ii) For the central look, a histogram is created and M classes are defined so that each class contains m approximately the same number of SNR values.
• iii) The SNR values of the other looks are supplemented based on the SNR value of the central look using the azimuth antenna diagram.
• iv) Optimization is performed for each class as described above. For a given desired minimum SNR of a pixel of the multi-look image SNR _ML,min, an optimization of the Ω weights Ω _m [k] of the individual looks k is carried out in such a way that the best possible radiometric resolution is achieved.
• v) At each pixel position, a class is assigned to the progression of the SNR values (without moving averaging) over the individual looks. This is done via the minimum overall distance between the course of the measured SNR values and the SNR values of the classes.
• vi) The Ω weights optimized for the class are used in the multi-look summation for the respective pixel position according to equation (7).

13 shows the optimized weights for M=32 classes and a minimum desired SNR of 0 dB in each pixel of the multi-look image. The Y-axis does not directly show the number of the class, but rather the SNR of the central look of the class. 14 shows the result of optimizing the Ω weights for the classes, namely the Equivalent Number of Looks (ENL), the SNR and the radiometric resolution (RR).

Comparison of image results for the entire partial image, area targets and mixed image content

Table 2 provides an overview of what is required for the respective multilook image 15 Multilook methods used, indicates the amplitude limit values used to display the gray values and indicates measured values for different image areas.

Table 2: Measured values for the in 15 shown multi-look images that were generated using various multi-look methods. Some of the values for 3L _FPC have already been given in Table 1.

Whole partial images

The 3L _FPC image contains little noise because only the inner 3 looks with good SNR were used, but there is a high speckle, which is shown by the ENL value of approx. 2 for the area targets. Although a fixed low weight is applied for the external looks for 9L _LNPC , the ENL is improved to a value of 4. However, the noise is increased due to the external looks, which is visible in the image through the gray haze that can be seen when comparing with 3L _FPC . Another indicator of the increased thermal noise is the reduction in contrast from 56.5 to 48.3. The improvement in terms of speckle can be seen in the improved ENL values. The image contrast is improved to 57.1 for Ω-weighting with a desired minimum multilook SNR of SNR _ML,min = 3dB.

Area targets

The improvement of the ENL is for the three area targets A, B, and C 15 remeasured. They have a moderate and high radar backscatter cross section. The improvement in ENL is greatest for 9L _LNPC , as the increase in noise also helps improve the ENL statistics. However, the improvement for 9L _Ω.0dB and 9L _Ω.3dB is of the same order of magnitude. For the high radar backscatter cross section of area target C, the improvement in the Ω-weighting images is higher than in 9L _LNPC , since more looks with higher weights were summed as a consequence of the better SNR. The higher weighting is in 13 visible. The measured SNR is best for 3L _FPC and lower for the other three images. The radiometric resolution, which results from SNR and ENL, is for the invention The appropriate methods “LNPC and Ω-Weighting” have been significantly improved. The measured image contrast in Table 2 decreases significantly in the area targets for both methods according to the invention due to the improved radiometric resolution.

Mixed image content

The improvement in speckle and noise is also evident in the scenery of area targets and routes in target area E. In 3L _FPC the detectability of the paths is limited by the high speckle and in 9L _LNPC by the high noise. The best detectability for this scenery is achieved with 9L _Ω.3dB , i.e. with Ω weighting and a high desired SNR in the multi-look image. The reduced speckle is evident and the measured contrast is similar to 3L _FPC .

Man-made objects such as buildings generate high backscatter, for example on edges or on surfaces that are perpendicular to the direction of the radar, and extremely low backscatter in shadowed areas. This leads to high image contrast. High noise with low backscatter reduces image contrast and thus the detection and classification options for urban areas. Table 2 gives the measured contrast for the target area D with buildings and the target area F in 15 at. The increase in noise from 3L _FPC to 9L _LNPC reduces contrast because the areas with low radar backscatter cross sections appear brighter due to the increase in noise. Ω-Weighting significantly improves image contrast by high weighting of external looks for strong targets and at the same time low weighting of external looks for the weak radar backscatter cross section of shadowed areas. Target area F contains a roof structure. The images with Ω-weighting show reduced speckle for the area target around the roof structure. The improvement in SNR from 9L _LNPC over 9 _LΩ.0dB to 9L _Ω.3dB is clearly observable. At 9L _Ω,3dB , several pixels are zeroed because the expected minimum multi-look SNR is less than the desired 3 dB.

Existing SAR systems can be improved in the swivel angle range and thus in the geometric resolution in azimuth using the wrapped staring spotlight mode, which is not according to the invention. Examples of systems currently in operation include the TerraSAR-X system, the Canadian Radarsat system, the Japanese ALOS 2/PalSAR 2 system or the Italian Cosmo-Skymed system. The Spanish PAZ system, which is about to be launched, is also a potential system that can be improved by Wrapped Staring Spotlight. All future SAR systems with phased array antennas will benefit from the wrapped staring spotlight mode. Aircraft-borne SAR systems with phase-controlled array antennas that can be controlled in the azimuth direction can also be subsequently improved in terms of geometric and/or radiometric resolution using the wrapped staring spotlight mode. The core of the wrapped staring spotlight method can also be used for other SAR modes, such as bi-directional SAR or forward/backward-looking SAR, i.e. azimuth viewing directions can be evaluated in existing SAR systems are not within the originally designed control angle range. The TOPSAR mode can also be expanded by extending the swivel angle range. An example of such a system is Sentinel-1.

The image quality can be significantly improved in SAR acquisitions with poor SNR in parts of the synthetic aperture using the LNPC or Ω-weighting according to the invention. Both methods can be built into SAR processors from the outset or subsequently and enable multi-look SAR images to be improved in terms of the quality parameters of contrast and radiometric resolution. Examples are again the SAR systems mentioned above in the wrapped staring spotlight mode.

bibliography

• [1] J. Mittermayer, S. Wollstadt, P. Prats-Iraola, R. Scheiber: “The TerraSAR-X Staring Spotlight Mode Concept”, IEEE Trans. on Geosc. Remote Sens., vol. 52, no. 6, pp. 3695-3706 , June 2014.
• [2] J. Mittermayer, S. Wollstadt, P. Prats, P. Lopez-Dekker, G. Krieger, A. Moreira: “Bidirectional SAR Imaging Mode”, IEEE Trans. on Geosc. Remote Sens., vol. 51, no. 1, pp. 601-614 , January, 2013.
• [3] J. Mittermayer, A. Moreira and O. Loffeld, “Spotlight SAR Data Processing Using the Frequency Scaling Algorithm,” IEEE Trans. on Geosc. Remote Sens., vol. 37, no. 5, pp. 2198-2214 , September, 1999.

Claims (3)

Method for creating an earth observation image of a region using a synthetic aperture radar (SAR) and a phase-controlled group antenna composed of individual antennas with an azimuthal swivel angle range that can be achieved by a control angle, each individual antenna having an element factor describing the individual antenna gain as a function of the solid angle and the group antenna having an array factor which describes the group antenna gain resulting from the arrangement and number of individual antennas as a function of the solid angle, the group antenna gain having several main maxima which are essentially the same size and which, however, with regard to the overall gain, which results from the element factor and the group antenna gain results, are attenuated towards increasing control angles θ, with the method - the earth observation image is created using multi-look processing, ie based on the processing of an overall image of a region which is composed of several created by illuminating over different viewing angle ranges Partial recordings of the region, ie looks (k), whereby the partial recordings are each filtered out of the overall recording, whereby - portions of the synthetic aperture that have a signal-to-noise ratio (SNR) at certain viewing angles that is smaller than a predeterminable one Limit value, can be made usable by a signal-to-noise-dependent summation of the individual looks (k), whereby the summation is either - derived directly from the azimuth antenna diagram, regardless of location, by a normalized correction of the antenna diagram (LNPC) in the individual looks ( k), whereby the antenna diagram is first corrected and then normalization is carried out in each look so that at a certain azimuth angle or, if the operation is carried out in the spectrum, at a certain azimuth frequency specific to each look (k), the product of inverse Antenna diagram and normalization results in a value of one in each look, or - depending on the location, a different signal-to-noise-dependent weighting Ω _m [k] of the looks (k) is carried out at each pixel position (P), whereby the weighting function (A _LNPC,k (f _a )) is created according to the desired optimization criteria. Procedure according to Claim 1 , characterized in that with location-dependent weighting of the looks (k), the optimization criterion for creating the weighting function (A _LNPC,k (f _a )) is - the best possible radiometric resolution and / or - the best possible signal-to-noise ratio in the multi-look image and/or - the highest possible number of Equivalent Number of Looks. Method according to one of the preceding claims, characterized in that the recording is carried out in spotlight mode and thus by extending the illumination of the region to be recorded as a result of expanding the azimuthal swivel angle range, or in TOPSAR mode by extending the azimuthal swivel angle range, or in BiDirectional SAR mode or in forward/backward looking SAR mode.

Images