Patent Application
20210033725
The present disclosure relates to methods and systems for radar image processing. In particular, the disclosure relates to methods and systems for radar image processing in radar remote sensing of a surface of a body of water such as an ocean or sea. More particularly, the disclosure relates to methods and systems for radar image processing in ship-borne radar remote sensing of a surface of a body of water such as an ocean or sea.
A first aspect provides a method of remote sensing of a body of water comprising:
Applying the one or more calibration offsets may generate geographically registered radar imagery that is clear and stable enough to perform wave inversion techniques.
The method may be carried out on a computer. The method may be a computerized method.
The method may comprise applying two or three of the calibration offsets.
The radar intensity images and/or the time intensity images may include a portion of land adjacent the body of water. The portion of land adjacent the body of water may comprise for example a shoreline, a beach, a marsh. The portion of land adjacent the body of water may have thereon one or more buildings or other structures.
The radar apparatus may be operable to obtain x-band radar intensity images.
The navigation unit may be operable to record accurate positional and heading data, typically at high temporal resolution. The navigation unit may comprise a Global Navigation Satellite System (GNSS) and/or an Inertial Navigation System (INS).
Each of the plurality of radar intensity images may be converted into a geographically-referenced image using position and heading data recorded by the navigation unit. The or each geographically-referenced image may be a Cartesian co-ordinate image. Converting the radar intensity image(s) to the geographically-referenced image(s) may be performed using a bilinear interpolation.
A synthetic aperture image may be constructed from a number of sequential geographically-referenced radar intensity images, e.g. so that:
where p is the intensity of each pixel in scan Sc. I is therefore a time-integrated image of a set of n geographically-referenced radar intensity images.
The visual sharpness F of image I may be determined by the stability of static targets over the number of geographically-referenced radar intensity images and can be defined as the variance of the two-dimensional image gradient so that:
The sharpness F may therefore be maximised when the variance of gradients across static targets is maximised. This occurs in a visually sharp image as transient targets (e.g., sea clutter) will average out over n geographically-referenced radar intensity images to give a low pixel intensity whereas static targets will produce a high pixel intensity that in a sharp image has well defined edges. Therefore in a sharp time-integrated image the number of well-defined edges is low compared to the number of poorly defined edges (from sea clutter), producing a high F value.
The method may maximise F by applying one of more, e.g. all three, of the calibration offsets and generating image I. Image I may be generated through a search path, e.g. an efficient search path.
The method may assume that the three calibration offsets are not covariant and have an order of relative importance to image stability, for instance: angle α, range β and time γ.
Applying the calibration offsets may begin in a first pass with all three offsets initially set to equal zero, conducting a search for an optimum value of a first calibration offset, e.g. α, and then, optionally, using this offset value (with the other two calibration offsets initially set to zero) in a second pass to search for an optimum value of a second offset, e.g. β. The steps may then be repeated in a third pass to find a third offset, e.g. γ, using the previously-found values for the first and second offsets and γ initially set to zero. In this way, the maximum image sharpness Fmax should increase over each of the three passes in the program, with the maximum possible value of Fmax found when all three offsets have been optimally determined.
The vessel may comprise for example a ship, a boat, a raft, a hovercraft.
The body of water may comprise for example an ocean, a sea, a lake, a river, an estuary, a delta, a reservoir, a canal.
Surveying the surface of the body of water may comprise manoeuvring the vessel and, optionally, one or more further vessels, across at least a portion of the body of water.
A second aspect provides a system for remote sensing of a body of water comprising:
The computer may be configured to perform a method of the first aspect.
Applying the one or more calibration offsets may generate geographically registered radar imagery that is clear and stable enough to perform wave inversion techniques.
The computer may be configured to apply two or three of the calibration offsets.
The radar intensity images and/or the time intensity images may include a portion of land adjacent the body of water. The portion of land adjacent the body of water may comprise for example a shoreline, a beach, a marsh. The portion of land adjacent the body of water may have thereon one or more buildings or other structures.
The radar apparatus may be operable to obtain x-band radar intensity images.
The navigation unit may be operable to record accurate positional and heading data, typically at high temporal resolution. The navigation unit may comprise a Global Navigation Satellite System (GNSS) and/or an Inertial Navigation System (INS).
The computer may be configured to convert each of the plurality of radar intensity images into a geographically-referenced image using position and heading data recorded by the navigation unit. The or each geographically-referenced image may be a Cartesian co-ordinate image. Converting the radar intensity image(s) to the geographically-referenced image(s) may be performed using a bilinear interpolation.
The computer may be configured to construct a synthetic aperture image from a number of sequential geographically-referenced radar intensity images, e.g. so that:
where p is the intensity of each pixel in scan Sc. I is therefore a time-integrated image of a set of n geographically-referenced radar intensity images.
The visual sharpness F of image I may be be determined by the stability of static targets over the number of geographically-referenced radar intensity images and can be defined as the variance of the two-dimensional image gradient so that:
The sharpness F may therefore be maximised when the variance of gradients across static targets is maximised. This occurs in a visually sharp image as transient targets (e.g., sea clutter) will average out over n geographically-referenced radar intensity images to give a low pixel intensity whereas static targets will produce a high pixel intensity that in a sharp image has well defined edges. Therefore in a sharp time-integrated image the number of well-defined edges is low compared to the number of poorly defined edges (from sea clutter), producing a high F value.
The computer may be configured to maximise F by applying one of more, e.g. all three, of the calibration offsets and generating image I. Image I may be generated through a search path, e.g. an efficient search path.
The computer may be configured to assume that the three calibration offsets are not covariant and have an order of relative importance to image stability, for instance: angle α, range β and time γ.
Applying the calibration offsets may begin in a first pass with all three offsets initially set to equal zero, conducting a search for an optimum value of a first calibration offset, e.g. α, and then, optionally, using this offset value (with the other two calibration offsets initially set to zero) in a second pass to search for an optimum value of a second offset, e.g. β. The steps may then be repeated in a third pass to find a third offset, e.g. γ, using the previously-found values for the first and second offsets and γ initially set to zero. In this way, the maximum image sharpness Fmax should increase over each of the three passes in the program, with the maximum possible value of Fmax found when all three offsets have been optimally determined.
The vessel may comprise for example a ship, a boat, a raft, a hovercraft.
The body of water may comprise for example an ocean, a sea, a lake, a river, an estuary, a delta, a reservoir, a canal.
A third aspect provides a computer readable medium comprising software that, when loaded on a suitable computer, configures the computer to process a plurality of radar intensity images recorded on a data storage device so as to register a geographic location of each of the plurality of radar intensity images, wherein registering the geographic location of each of the plurality of radar intensity images comprises:
The software may configure the computer to perform a method according to the first aspect.
The skilled person will appreciate that except where mutually exclusive, a feature described in relation to any one of the above aspects may be applied mutatis mutandis to any other aspect. Furthermore except where mutually exclusive any feature described herein may be applied to any aspect and/or combined with any other feature described herein.
Embodiments will now be described by way of example only, with reference to the Figures, in which:
In embodiments, this disclosure may provide a simple offset “calibration” method for the accurate geographic registration of ship-borne X-band radar intensity imagery
X-band radar remote sensing of the ocean surface using ground-based installations and radar intensity imagery is a mature technology for the determination of ocean wave properties. In order to transition the analytical methods to data collected from a moving vessel the accurate geo-registration of radar images must be performed. However, finite offsets in azimuth, range and time are generated by the physical installation of equipment aboard a vessel that may not be measurable. This work details a simple, yet robust method to ‘calibrate’ data recorded by an arbitrary equipment installation to allow accurate geo-registration of the radar imagery. Time-integrated radar images are generated using a set of time-stamped radar intensity images and high-frequency, high-accuracy vessel heading and position information (from global satellite navigation systems and inertial navigation systems). The time-integrated image sharpness is found to be maximal for correctly determined angular, range and time offsets. An operational system for practising the method is also discussed.
Ground-based X-band radar using intensity imagery is an established tool for the remote sensing of the sea surface through primary determination of ocean wave spectra ([1], [2], [3]) and utilizing this spectral information to determine bathymetry through wave-inversion techniques ([4], [5], [6]) and near-surface currents through the Doppler shift of directional wave spectra [6]. Wave-inversion has been demonstrated to work with radar data recorded from a moving vessel for the determination of bathymetry [7] and surface currents [8] as well as estimates of surface elevation [9] and wind fields [10]. The capabilities of shore based radars for depth and current mapping are well documented, but these systems are inherently limited to line of sight and limited range, usually of the order of 4 km. If the same methods can be applied to data from moving vessels, there is potential for a single system to cover a vastly greater area and avoid the blind spots inherent in static systems. For example, if a pilot vessel, tug or port survey vessel were to be equipped with such a system, it could be returning almost continuous updates from entire port operation areas—perhaps monitoring the evolution of sandbanks adjacent to navigation channels. Such systems could also be used for rapid assessment of hurricane or typhoon damage in order to guide more detailed high resolution surveys by the limited resources of conventional survey boats.
A problem associated with transitioning ground-based radar techniques to shipborne use is the accurate frame-by-frame geographic registration of the digitised radar images. To achieve this, accurate (and frequent) position and heading data for the vessel must be obtained. It is common practice to use multi-antenna Global Navigation Satellite Systems (GNSS) and Inertial Navigation Systems (INS) to record accurate positional and heading data at high temporal resolution. However, the physical installation and mounting of such systems can introduce an unknown (i.e., not routinely or simply measured) angular offsets between the GPNSS/INS unit, radar transceiver and the vessel's true ‘bow up’ heading.
As the vessel moves in geographic space any recorded sea echoes must originate from consistent points in space; i.e., the movement of the vessel (horizontal motion or changes in heading) must not affect the recorded position of a sea echo between frames. This is a similar problem as is faced in Synthetic Aperture Radar (SAR) or Synthetic Aperture Sonar (SAS) signal processing. In order to generate a clear, high-contrast SAR or SAS image every recorded signal must be accurately and consistently geo-registered.
Three principal (‘first order’) systematic offsets are here identified that need to be known in order to accurately geo-register marine radar data using an accurate GNSS/INS system for positional and navigation data.
Correcting for these first-order offsets creates geo-registered radar imagery clear and stable enough to perform wave inversion techniques. There are a number of physical effects that can generate additional offsets that are neglected in this disclosure. These include:
Reference [8] calculated the alignment error α by adopting the calibration methods of [11] developed for ship-borne Acoustic Doppler Current Profiler (ADCP) measurements. The method requires the determination of radar-derived near-surface current components (via wave inversion techniques) during a period of vessel manoeuvring. The method has been shown to be highly effective, however individual estimates of α are dependent on accurate determination of near surface current vectors and therefore tend to be noisy [8], requiring averaging across a number of estimates.
The requirement for the use of sea clutter wave inversion techniques also requires the presence of adequate sea clutter (and therefore wave height and wind speed). There are no published methods for determining the start range and system time errors β and γ, leaving trial-and-error as the present solution for the determination of these parameters. For ground-based radar installations this trial-and-error approach may be satisfactory as it technically needs only to be performed once per installation. However for unknown vessel equipment installations and the potential for frequent equipment changes this approach will be prohibitively time consuming. Additionally, the accuracy of trial-and-error offset determination is dependent on the time applied to the problem. An automated, accurate method is therefore required.
As wave inversion using ground-based marine radar intensity imagery is a mature technology there remain significant opportunities expanding the approach to vessel-borne data. However a simple and robust approach is needed to derive these offsets that are critical for accurate image geo-registration. The method presented in this work for the determination of α, β and γ avoids the need for physical parameters (such as wave spectra or surface current components) to be derived from radar intensity imagery, instead relying on accurate position and heading information (from GPS/INS) and a simple, yet robust, image processing technique. The robustness of the method lies in the SAR-type images that are generated as a result: clear, sharp, time-integrated images can only be generated if the geographic coordinates of each pixel in each recorded frame is accurate and stable as the vessel moves.
The new automated calibration algorithm is based on the concept that echoes from static targets (e.g., radar marker buoys, sea walls, land clutter from buildings, saltmarshes, etc.) must remain in the same position on a geo-referenced radar image even after an arbitrarily large number of sequential scans while the scanning origin (the radar) is both in motion and its reference azimuth (the heading of the vessel) is changing. As the vessel manoeuvres the position of the echoes become steadily less correct as the errors in geo-referencing are cumulative with the reference point in motion. This apparent motion of static targets can be minimised with prior knowledge of their origin. However, it is also possible to quantify their ‘static-ness’ using simple image processing techniques.
First, each digitised polar-coordinate (azimuth-range) image Sp is converted into a geo-referenced Cartesian coordinate (e.g. OS Grid, UTM) image Sc using high-resolution position and heading information, e.g., from a multi-antenna GNSS/INS system. In this case the Cartesian conversion is performed using a simple but robust bilinear interpolation.
The Cartesian radar image Sc is defined in geographical coordinates (x,y), and is a function of the polar-coordinate image Sp (coordinates azimuth a and range b), defined as:
S
c
=F(Sp{a−α,b−β})
where the radar beam azimuth α is relative to the ship heading which is recorded alongside the vessel position in (x,y) coordinates by GNSS navigational equipment at time T=T−γ. The calibration offsets α, β and γ are applied to a, b and T respectively.
Next, a ‘synthetic aperture’ image I is constructed from an arbitrarily large number n of sequential Cartesian image scans Sc so that:
where p is the intensity of each pixel in scan Sc. I is therefore a time-integrated image of a set of n scans. The visual sharpness F of image I will be determined by the stability of static targets over the number of scans and can be defined as the variance of the two-dimensional image gradient so that:
The sharpness F is therefore maximised when the variance of gradients across static targets is maximised. This occurs in a visually sharp image as transient targets (e.g., sea clutter) will average out over n scans to give a low pixel intensity whereas static targets will produce a high pixel intensity that in a sharp image has well defined edges. Therefore in a sharp time-integrated image the number of well-defined edges is low compared to the number of poorly defined edges (from sea clutter), producing a high F value.
The new calibration algorithm acts to maximise F by applying calibration offsets and generating image I through an efficient search path. Any suitable search path may be employed, although generally a more efficient search path may be preferred, since a more efficient search path may determine the correct values of the calibration offsets (i.e. α, β and γ) more quickly and/or using less computing resource. The approach assumes that the three calibration offsets are not covariant and have an order of relative importance to image stability: angle α, range β and time γ. There may be a degree of co-dependency between the angular and range offsets and their effect on F, although the effect is minimal and is discussed later. The program begins with all offsets initially set to equal zero, searches for an optimum value of α and then uses this offset value (with the other two initially set to zero) when gridding images Sc for the second pass to search for the optimum value of β. This process is repeated for the final pass to find γ, with the previously found values for the other two offsets and γ initially set to zero. In this way, the maximum image sharpness Fmax should increase over each of the three passes in the program, with the maximum possible value of Fmax found when all three offsets have been optimally determined, i.e. when α, β and γ are all correct.
In principal the calibration method requires an arbitrary number of intensity images containing static targets to function correctly. However the accuracy of the process to search for Fmax is weakly dependent on the number of images provided to the algorithm as well as the relative proportion of static targets to sea in the total gridded area. The implications of the quantity and quality of input data on offset determination are discussed in a later section.
To test the new calibration algorithm a sample radar intensity image dataset was selected, recorded by the Natural Environment Research Council (NERC) research vessel RSS Discovery upon leaving port in Southampton and travelling South-East through the Solent. The RSS Discovery is equipped with an incoherent 25 kW Furuno X-band radar operating on short pulse (50 ns) for scientific purposes (separate from the navigational radar), with the transceiver and 1.8 m horizontally-polarised (‘HH’) antenna (providing a 3 dB beamwidth of 1° in the horizontal and approximately 20° in the vertical) installed approximately 30 m above the waterline on the ship's instrumentation tower. The raw radar intensity data were captured and digitised at 30 MHz and 12 bits by a WaMoS II (OceanWaves GmbH) radar computer and compressed to polar coordinate images with a pixel resolution of 0.3° in azimuth and 7.5 m in range. The digitiser was set to record images out to 2 km range with an antenna rotation rate of 2.4 s (25 rpm) in records 64 images in length. 14 image records were selected providing a total of 896 images in the sample set. The bilinear, nearest-neighbour Cartesian image interpolation was performed to a 20 m pixel size to perform the calibration routines and a 5 m pixel size for imagery to visually inspect the results. A second dataset of 512 images recorded as the Discovery sailed through the Sound of Mull, Scotland, was also obtained to demonstrate the image stabilisation effect possible with correctly determined offset variables.
In each of
Regardless of the choice of post-processing of vessel-borne marine radar imagery, the quality of any derived data products will be proportional to the accuracy of geo-registration. As the physical equipment mounting positions, effect of radar cable length and computer internal delays are not easily or readily measured the offsets that are required for accurate registration need to be derived using minimal information. The method proposed in this study requires only a sequence of suitable radar images and the associated vessel heading and position data for the time of recording.
This work has described an accurate and robust method for determining the offset parameters required to correctly geo-register vessel-based marine radar imagery. However in order to create an operational system a number of issued must be addressed.
It is apparent that there is a mild co-dependency of α and β on the calculation of Fmax. This dependency may be due to the form of input imagery to the algorithm combined with the programmatic method of determination of F (i.e., with β and γ set to zero when initially searching for α). The co-dependency manifests an image I with a calculated highest F value but which remains slightly blurred; especially around small, static targets. A simple solution for this could be to use a first guess for β and γ based on prior knowledge of the system under calibration which will lead to a more accurate determination of α.
The number and content of radar intensity images required for the algorithm to perform accurately is dependent on vessel motion and manoeuvres during image acquisition. The most useful images are those that contain between 20% and 50% land (preferably with a sharply defined land/sea boundary such as coastal defences or harbour walls) and are recorded when the vessel is in motion and turning. In this ‘best-case’ scenario an image set of between 512 and 1024 scans are recommended to accurately perform the described method.
Accurately geo-registered radar intensity images recorded by ship-borne radars have a number of uses aside from wave inversion of sea clutter. The SAR-type, time-integrated images that are created by the presented method utilise radar data that is not routinely recorded by vessel operators. Assuming the operator has access to GPS/INS position and heading data the only additional requirement to transform the ephemeral information presented by the radar's plan position indicator (PPI) to high resolution (˜5 m) maps of the coast and other static targets is the provision of accurate angular, spatial and time offsets. These high-resolution, geo-registered images could be utilised for coastal surveying, for example using the waterline method [12] for intertidal areas or wave inversion for wet areas in the presence of waves. Survey mapping is also a possibility: especially salt-marsh monitoring (e.g.,
By applying one or more of the calibration offsets to the polar-coordinate radar image(s), the sharpness of the Cartesian radar image(s) may be improved sufficiently to enable more reliable and/or accurate geographic registration.
Any number of the calibration offsets, i.e. one, two or three of α, β and γ, may be applied. The necessary calibration offsets may be applied in any order. The specific number and/or order of calibration offsets that may need to be applied may vary, e.g. depending upon the specific equipment set-up on a given vessel.
In general, if α>1°, then time integrated radar images from a moving vessel produced without applying the calibration offset a will be too blurry to be used in the remote sensing of a body of water.
It will be appreciated that the values of α, β and γ may vary from one vessel to another. The values of α, β and γ may not be constant for a given vessel.
Typically, α may be at least ±1°, up to or at least ±2°, up to or at least ±5°, up to or at least ±10° and/or up to or at least ±15°.
Typically, β may be at up to or at least 10 m, up to or at least 20 m and/or up to or at least 50 m.
Typically, γ may be of the order of a few seconds or a few minutes. For instance, γ may be at least 1 s, up to or at least 5 s, up to or at least 10 s, up to or at least 30 s, up to or at least 60 seconds, up to or at least 2 minutes and/or up to or at least 5 minutes.
It will be understood that the invention is not limited to the embodiments above-described and various modifications and improvements can be made without departing from the concepts herein. Except where mutually exclusive, any of the features may be employed separately or in combination with any other features and the disclosure extends to and includes all combinations and sub-combinations of one or more features described herein.
| Number | Date | Country | Kind |
|---|---|---|---|
| 1801613.9 | Jan 2018 | GB | national |
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/GB2019/050258 | 1/30/2019 | WO | 00 |
