Synthetic Aperture Sonar Images Research Articles

Inverse Synthetic Aperture Sonar Imaging of Underwater Vehicles Utilizing 3-D Rotations

Inverse synthetic aperture sonar (ISAS) can obtain the image of a moving underwater vehicle (UV) with both high range resolution and high cross-range resolution. Existing ISAS imaging methods are based on the translational motion of the UV. Since a real UV also does 3-D rotations, these methods suffer from defocusing especially in severe sea condition. Hence, a novel ISAS imaging method utilizing the 3-D rotations is proposed in this paper. Under the 3-D rotational model, the echo in slow time is simplified as an amplitude modulated-cubic phase signal (AM-CPS), and a high precision signal parameter estimation algorithm based on the Radon transform cubic phase function is presented to process the AM-CPS. Moreover, the image plane is analyzed. The result shows that the ISAS imaging method based on the 3-D rotations can obtain the target's images on more image planes compared with the translational motion based methods, and the image quality is improved compared with other signal parameter estimation algorithms. The simulation verifies the effectiveness of the proposed method.

Imaging Algorithm for Multireceiver Synthetic Aperture Sonar

This paper presents a range migration algorithm for the multireceiver synthetic aperture sonar (SAS) based on Loffeld’s bistatic formula, which consists of the quasi monostatic term and the bistatic deformation (BD) term. It is expected that the multireceiver data can be easily focused by the monostatic SAS imaging algorithm after compensating the BD term. In this paper, the BD term is first decomposed into the range variant phase and the range invariant phase. At this point, the BD term is compensated via two steps. One step is the compensation of the range variant phase through the range-dependent sub-block processing in the 2-D frequency domain. The other one is the arrangement of the multireceiver data in a sequential order in the 2-D time domain. After these operations, we obtain the monostatic SAS equivalent data, which is viewed as the input of the traditional range migration algorithm. The processing results of the simulated data and the real data show that the proposed method can work well for the focusing of the multireceiver SAS data. More importantly, the presented method dramatically improves the imaging efficiency without loss of performance since the complex multiplication and the data arrangement are only exploited by the presented method.

Grating lobe prediction and deconvolution for synthetic aperture sonar

Synthetic aperture sonar (SAS) arrays are typically undersampled in the sense that the array element spacing is much larger than the acoustic wavelength. Grating lobes are suppressed by making a judicious choice of beampattern nulls for the transmit and receive elements, such as a 3:2 ratio for the length of the transmit and receive elements. However, grating lobe artifacts can appear in SAS imagery when the target strength difference between a highly reflective object and the surrounding seabed exceeds the sidelobe level of the synthetic array. We present a theoretical model of the SAS point scattering function (PSF) that takes into account shaded element beampatterns for a multichannel SAS array. The PSF model is validated using experimental data from AquaPix, a wideband 300 kHz interferometric SAS. In practice, observed SAS images are described by the convolution of the seabed reflectivity with the PSF. Therefore, knowledge of the PSF facilitates the removal of grating lobe artifacts using deconvolution techniques such as the Richardson-Lucy algorithm. Conventional and deconvolved SAS images of a highly reflective target are presented to demonstrate the effectiveness of deconvolution based on the modeled PSF.

Wavenumber domain acceleration for stripmap fast factorized backprojection beamforming

Synthetic aperture sonar (SAS) imaging is frequently used in conjunction with autonomous underwater vehicles (AUV's) to perform high resolution wide-area surveys. Wave action and currents perturb the trajectories of AUV's performing these surveys and these perturbations have important ramifications for the SAS beamforming process. Generalized beamforming algorithms operating in the time domain are very flexible with regard to trajectory non-linearities, but they also tend to have a higher computational cost than alternative frequency domain algorithms. Of the generalized beamformers one of the more efficient is fast factorized backprojection (FFBP), which leverages sub-aperture processing and a series of coordinate transformations to improve computational efficiency. FFBP can be used to beamform SAS data, however many SAS systems make use of a real aperture to boost range potential. The existence of a real aperture implies that a frequency domain beamformer could be used to bypass the first few stages of FFBP without loss of generality; however, standard frequency domain SAS beamforming algorithms are not compatible with the FFBP framework. A wavenumber domain beamformer compatible with the early-stage FFBP framework is derived and beamforming results for the modified FFBP algorithm are compared with the standard approach that operates entirely in the time domain.

Effects of reverberation on estimates of synthetic aperture sonar multi-look coherence

Multi-look coherence makes use of phase information for target detection and classification by splitting the angle and frequency spectral bandwidth contained in a complex synthetic aperture sonar image into sub-bands and then estimating the coherence between the images formed from these sub-bands. Based on experimental results, it appears to be possible to separate man-made targets from interfering reverberation and clutter using coherence, as targets have features that scatter coherently as a function of angle or frequency. Characteristics of a target object may also be inferred, as sub-band coherence is a sensitive function of both angle and frequency. The expectation operation performed to obtain estimates of the complex correlation coefficient is evaluated using a spatial average, which lowers the spatial resolution of the resulting coherence map. The final resolution of a coherence image will be a function of the original image resolution, the number of sub-look images formed from the original image and the number of samples used in the expectation. In this work, we explore the effects of signal, reverberation, and noise levels on estimates of coherence using experimental data. Trade-offs in terms of signal-to-reverberation, expectation window size, and the number of sub-looks will be presented.

Utilization of Aspect Angle Information in Synthetic Aperture Images

Synthetic aperture sonar and synthetic aperture radar involve the creation of high-resolution images of a scene via scattered signals recorded at different locations. Each pixel of the reconstructed image includes information obtained from multiple aspects due to the changing position of the sources/receivers. In this paper, the aspect-dependent scattering at each pixel is exploited to provide additional information about the scene; this paper presents a framework for converting and utilizing multiaspect data, as well as several examples. For sonar data, as is presented here, the aspect dependence may be leveraged to separate objects of interest from the background, to understand the local bathymetry, or for visualizing acoustic shadowing in full circular synthetic aperture sonar images. Several examples of images of the seafloor containing objects of interest are presented for both circular and linear apertures. In addition, the aspect dependence of low-frequency elastic scattering from objects may be used to understand the underlying scattering physics, which is of potential use in fields such as target recognition and nondestructive testing; a laboratory example is presented.

Quantifying the effect of random roughness on synthetic aperture sonar image statistics

A perturbation-theory-based model has been developed to predict the effect of random seafloor roughness on synthetic aperture sonar (SAS) image statistics. The continuous variation in scattering strength produced by a random slope field is treated as an intensity scaling on image speckle produced by the SAS imaging process. Changes in image statistics caused by roughness are quantified in terms of the scintillation index (SI). Factors influencing the SI include slope variance, geo-acoustic properties of the seafloor, the probability density function describing the speckle and the signal-to-noise ratio. Example model-data comparisons will be shown for SAS images taken off the coast of Tellaro, Italy, by the NATO Undersea Research Centre, La Spezia, Italy (now the NATO Centre for Maritime Research and Experimentation) using the 300 kHz MUSCLE SAS system, and for data collected with the Norwegian Defence Research Establishment’s 100 kHz HISAS system off the coast of northwest coast of Elba Island, Italy, in ...

Simulating an object response with synthetic aperture sonar inverse imaging

Synthetic aperture sonar (SAS) algorithms were developed in order to produce high resolution imagery from wide-beam wideband data. The high data rates associated with SAS data have fostered the development and implementation of efficient beamforming algorithms, many of which were originally developed for synthetic aperture radar. Some of these beamforming algorithms are invertible, and inverse beamforming techniques have been developed for SAS data analysis, where imagery is inverted in order to produce aspect-dependent data. Since imagery is invertible, this process can also be used as a simulation to produce stave-level data from a simulated image. This paper describes a modeling approach where inverse imaging techniques are applied in order to generate simulated stave-level sonar data. Using this approach, an image is generated from assumed object geometry and reflectance and inverted in order to generate stave-level data. The stave level data is then beamformed specifically to produce frequency vs. aspect data products. Since the inverse imaging process assumes specular reflection, features present in the frequency vs. aspect data are solely due to the surface reflectance and object shape. Comparison of inverse image simulated data with physics based simulations or field-collected data are shown and the limitations of the comparison are discussed.

Synthetic Aperture Sonar Track Registration Using SIFT Image Correspondences

Repeat–pass synthetic aperture sonar (SAS) imaging for change detection and interferometry relies on accurate image registration. SAS image registration is usually performed using a slow correlation-based method, whereas feature-based registration is becoming more prevalent for synthetic aperture radar due to its relative computational efficiency. In this paper we propose a sonar track registration method, from which the estimated track parameters can be used to compute an image registration. The proposed method uses the scale-invariant feature transform (SIFT), a popular algorithm for feature detection and matching, and is demonstrated on a simulated repeat–pass pair of SAS images. A model of the ideal sonar track geometry is presented. A least squares method is formulated for estimating the track registration from a set of noisy feature correspondences, where random sample consensus (RANSAC) is first used to reject outliers. The resulting track parameter estimate corresponds to an image registration aligned to within 0.03 pixels over the scene, which is within the 0.1-pixel resolution recommended for interferometric processing. The ground truth for the simulated ideal data is precisely known, which allowed for the subpixel localization accuracy of SIFT correspondences to be objectively evaluated. The interaction between the RANSAC error threshold and the misregistration error was also evaluated and found to be chaotic. Overall, our results indicate the potential for features to be used for faster coarse registration to reduce the computation time of a more accurate subsequent correlation-based registration.

Wideband Synthetic Aperture Sonar Backprojection With Maximization of Wave Number Domain Support

Wideband and widebeam synthetic aperture sonar (SAS) can provide information on the frequency- and aspect-dependent scattering in a scene. We suggest an approach to predict the quality of the sensor data over the available frequencies and aspect angles. We relate the typical spatial domain quality metrics to their wave number domain (WD) counterpart, and use these to map the data quality in WD. Because SAS arrays often are undersampled along-track, we pay particular attention to data degradation from aliasing. We use the proposed approach to examine how three SAS image formation algorithms based on time domain backprojection (TDBP) access data of different quality from wideband SAS systems. We illustrate the results with predictions for a generic SAS design and demonstrate the findings on two experimental systems. We observe that the maximum support of high-quality data is achieved through BP onto a high-resolution grid followed by WD filtering.

Deep convolutional neural networks for semi-supervised learning from synthetic aperture sonar (SAS) images

Advancements in deep neural networks for computer-vision tasks have the potential to improve automatic target recognition (ATR) in synthetic aperture sonar (SAS) imagery. Many of the recent improvements in computer vision have been made possible by densely labeled datasets such as ImageNet. In contrast, SAS datasets typically contain far fewer labeled samples than unlabeled samples—often by several orders of magnitude. Yet unlabeled SAS data contain information useful for both generative and discriminative tasks. Here results are shown from semi-supervised ladder networks for learning to classify and localize in SAS images from very few labels. We perform end-to-end training concurrently with unlabeled and labeled samples and find that the unsupervised-learning task improves classification accuracy. Ladder networks are employed to adapt fully convolutional networks used for pixelwise prediction based on supervised training to semi-supervised semantic segmentation and target localization by pixel-level classification of whole SAS images. Using this approach, we find improved segmentation and better generalization in new SAS environments compared to purely supervised learning. We hypothesize that utilizing large unsupervised data in conjunction with the supervised classification task helps the network generalize by learning more invariant hierarchical features. [Work supported by the Office of Naval Research.]

Automated change detection for synthetic aperture sonar

Change detection in seabed mapping is the process by which regions of interest are highlighted through the comparison of current scene data with historical data, and involves techniques that have application to a wide variety of disciplines and sensing modalities. Automated comparison of multi-temporal sidescan sonar images at the pixel (or parcel) level is challenging as the background pixels change due to a variety of factors. These include image resolution that degrades as a function of range, and temporal changes in the propagation medium and seafloor structure. With the advent of synthetic aperture sonars (SAS), which provide constant resolution over the swath and are deployed on actively navigated platforms capable of precise track repetition, the automated comparison of multi-temporal seabed images has in recent years been shown to be feasible. Change detection through the automated comparison of multi-temporal SAS imagery can be generally categorized as incoherent or coherent, where incoherent implies identifying changes in the amplitude, and coherent implies detecting changes in both amplitude and phase of the images. Utilizing the phase component may lead to detection of a physical disturbance in the scene. This presentation will provide an overview of the state-of-the-art of automated sonar seabed change detection.

Depth estimation of water column objects using interferometric synthetic aperture sonar

Synthetic Aperture Sonar (SAS) is a technique which delivers sonar images of the seabed with both high resolution and large area coverage rate. SAS is therefore a well suited sensor for search of small objects on the seafloor, and is an important tool in many emerging mine countermeasures (MCM) systems. An interferometric SAS system can also resolve the angle of arrival in the vertical plane, and thus estimate the depth of an object. In this talk, we present techniques using SAS imaging for depth estimation of small objects, not on the seafloor, but located in the water column. We consider the effect of geometry, sensor settings and processing parameters. For objects of interest, we present a belief propagation inspired method for estimating the depth of the objects. This method is CPU intensive, but avoids the phase ambiguity problem encountered in standard SAS interferometry. We compare this method to a coarse depth estimate acquired from the multipath response in the SAS images as well as to ground truth. We show example images and depth estimates from the Kongsberg HISAS interferometric SAS collected by a HUGIN autonomous underwater vehicle.

Circular synthetic aperture sonar image resolution theory

The introduction of SAS (Synthetic Aperture Sonar) systems has been a game changer for underwater surveys. The gain in resolution, compared to traditional sidescan systems, created a paradigm shift as the information contained in a SAS image switches from shadows to highlights. SAS systems traditionally perform lawnmower type surveys, but the need for multiple views in MCM (Mine-Counter Measure) tasks, for example, opened the interesting problem of target re-acquisition patterns. In particular, circular patterns maximize the aperture, thus the overall image resolution of such system. The capability of CSAS (Circular SAS) has been demonstrated on the field, but the derivation of CSAS processing has not been fully developed. The non-uniform sampling of the circular pattern in particular introduces aberrations within the field of view and a non uniform PSF (Point Spread Function). In this talk, we propose a new spatial sampling scheme which makes the CSAS PSF perfectly uniform. The theoretical closed form solution of the PSF is then derived both in time and Fourier domain. The PSF derivation naturally leads to redefine the image resolution as an energy leakage problem. Thanks to the new sampling scheme and the uniform PSF, we also propose a deconvolution method based on atom waves which increases the CSAS resolution.

Computing navigation corrections for co-registration of repeat-pass synthetic aperture sonar images

Interferometric processing of repeat-pass Synthetic Aperture Sonar (SAS) data requires a precise co-registration of the images in order to exploit the phase information for applications such as differential interferometry or coherent change detection. Co-registration of SAS images is made difficult by changing environmental conditions and positioning ambiguities that give rise to warping functions which are not well modeled by affine or polynomial transformations, resulting in non-negligible residual misregistration errors. The objective of this presentation will be to examine these warping functions in greater detail to gain a better physical understanding of the estimated along- and across-track displacements. A non-linear least-squares optimization is used to determine the corrections to be applied to the repeat-pass data that will yield a repeat-pass image that is co-registered with the original. These corrections include sway, heave and yaw as well as sound speed errors and element-level surge adjustments. Repeat-pass SAS data collected a Kongsberg HISAS 1020 system is used to demonstrate this optimization, showing good agreement with modeled and real data for most parameters, while determining surge estimates are shown to be particularly challenging. An examination of the resulting repeat-pass coherence will also be performed.

On using multi-look synthetic aperture sonar analysis for the investigation of scattering mechanisms

Multi-look coherence explores the information content of synthetic aperture sonar (SAS) images by splitting the total angle and frequency spectral bandwidth of a complex SAS image into sub-bands. In the presence of a random distribution of surface or volume scatterers, the spectral coherence is proportional to the degree of overlap in sub-bands of the SAS image. Consequently, for a perfectly random distribution of scatterers, the spectral coherence should be zero when the sub-bands do not overlap. If the scatterer distribution deviates from this basic random distribution, the spectral coherence may be non-zero for special cases that would be related to scattering from specific structures such as points or facets. For that reason it appears to be possible to separate coherent features from the random scattered background signal using spectral coherence, suggesting possibilities for detecting and/or characterizing objects (especially in low signal-to-reverberation situations). In this talk, we will discuss the concept and the utility of using the complex coherence between a series of sub-look SAS images to investigate different scattering mechanisms as well as show examples of the technique using real and modeled SAS data. [Work performed under ONR Grants N00014-16-1-2469 and N00014-16-1-2313.]

Deep neural networks for learning classification features and generative models from synthetic aperture sonar big data

Autonomous synthetic aperture sonar (SAS) imaging by unmanned underwater vehicles (UUVs) provides an abundance of high-resolution acoustic imagery useful for studying the seafloor and identifying targets of interest (e.g., unexploded ordnance or mines). Unaided manual processing is cumbersome as the amount of data gathered by UUVs can be enormous. Computer-vision and machine-learning techniques have helped to automate classification and object-recognition tasks, but often rely on hand-built features that fail to generalize. Deep-learning algorithms facilitated by emergence of graphics-processing unit (GPU) hardware and highly optimized neural-network implementations have recently enabled great improvements in computer vision. Autoencoders allow for deep unsupervised learning of features based on a reconstruction objective. Here, we present unsupervised feature learning applied to seafloor classification of SAS images. Deep architectures are also capable generative models. We illustrate this with generative networks that are capable of generating realistic SAS images of different seafloor bottom types. Deep models allow us to construct algorithms that learn hierarchical and higher-order SAS features, which promise to improve automatic target recognition (ATR) and aid operators in processing the large data volumes generated by UUV based SAS imaging. [Work supported by the Office of Naval Research.]

Three-dimensional image reconstruction of objects using synthetic aperture sonar

Synthetic aperture sonar imaging is typically considered for the case of linear or circular scans, where the reconstructed acoustic image is a two-dimensional representation of the true three-dimensional scattering from some object. However, a multi-dimensional aperture may be used to create a 3-D or volume image reconstruction of the acoustic scattering by that object. The 3-D image reconstruction may be performed using a delay-and-sum algorithm, or using comparatively fast wavenumber based methods. Simulated point scatterers will be used to initially demonstrate the algorithms, followed by a demonstration involving a simple object near a reflecting boundary. For objects that are located close to a reflecting boundary there exists a rich 3-D scattering structure. When imaged in two dimensions, object features becomes distorted due to the projection of scattering loci (e.g. the broadside reflection from a cylindrical target) that may be located outside of the image plane [Plotnick et al., J. Acoust. Soc. ...

Sonar image reconstruction of objects near a reflective boundary

The properties of reconstructed acoustic images, created from data obtained using backscattering synthetic aperture sonar (SAS) systems, have been studied for objects in the free field. The presence of a nearby reflecting boundary, such as the seafloor or surface, greatly increases the number of mechanisms or raypaths by which sound may be backscattered to the sonar system. Some of these raypaths involve one or more interactions with the reflecting boundary (multipaths), and others involve sound scattering multiple times from the same target (multiple scattering). Experiments were conducted using an air-water interface and simple cylindrical targets in order to examine the properties of the reconstructed circular SAS (CSAS) images, including effects due to vertical obliquity of the target [D. Plotnick et al., J. Acoust. Soc. Am. 137, 470-480 (2015)]. The CSAS image is by definition a two-dimensional representation of three-dimensional scattering; the image is thus highly dependent on the object’s 3-D shap...

Circular synthetic aperture sonar imaging of simple objects illuminated by an evanescent wavefield.

This paper is motivated by the case where an underwater object located within the sediment is illuminated by a grazing acoustic beam below the critical angle. The included experimental work uses a liquid-liquid interface and vertically inverted geometry as a stand-in for the water-sediment boundary. In the super-critical regime sound in the water column refracts into the sediment before scattering. However, for sub-critical illumination a rapidly decaying evanescent wavefield is generated in the sediment near the water-sediment interface. For compact objects located in the sediment near the interface this can result in strong backscattering signals suitable for acoustic image reconstruction using synthetic aperture sonar techniques. Certain properties of the evanescent wavefield such as the vertical phase-locking behavior, the rapid amplitude decay with distance from the interface, and the low-pass filter effect have understandable ramifications for the image formation process and for characteristics of the reconstructed image. In particular, circular imaging techniques require correct placement of the imaging plane to properly focus an object; however, for backscattering (monostatic) evanescent image formation the imaging plane may be placed at the interface and the target will remain in focus regardless of burial depth. A laboratory experiment using simple scatterers is presented.