Range Of Aspect Angles Research Articles

Contrastive Feature Disentangling for Partial Aspect Angles SAR Noncooperative Target Recognition

Deep learning algorithms have achieved state-of-the-art progress in synthetic aperture radar (SAR) automatic target recognition (ATR) tasks. They theoretically assume that training and test samples are independent and identically distributed (i.i.d.) for generalization, but it is intractable for practical ATR scenarios. In this paper, we propose a novel contrastive feature disentangling framework termed ConFeDent to learn features with improved generalization performance under a condition of a weaker distribution consistency. More specifically, ConFeDent aims to describe the semantic interactions between two arbitrary SAR training samples instead of treating them independently. It can implicitly disentangle features encoding the pose and identity knowledge from the whole samples with a semi-parametric geometric transformation model and a second-order energy model. In particular, except for the identity label, we use deductive-based geometry knowledge as supervision to teach the model to learn the concept of aspect angle variation. A progressively amortized inference scheme is constructed for efficient feature learning and recognition in an end-to-end manner. Finally, we further release a strengthened version, called ConFeDent+, which can explicitly utilize and learn more information from cross-category samples. Experimental results on the moving and stationary target acquisition and recognition (MSTAR) benchmark demonstrate the effectiveness of our proposed models in the SAR ATR. In particular, we validate the algorithms in a more challenging scenario where the range of aspect angles for training and testing samples is permitted to be disparate. Our model can achieve much higher recognition accuracy than other SAR ATR algorithms.

Optimal Design and Implementation of Inhomogeneous Planar Radome by Perforating the Host Material

An analytical expression for the reflection and transmission coefficients (TE and TM) for inhomogeneous planar layer (IPL) with variable inverse square relative permittivity is achieved. The correctness of derived expression is shown by comparison with Comsol numerical simulation. After choosing an appropriate material as the host medium, the transmission coefficient of inhomogeneous layer is calculated versus ${d} /{\lambda }_{{0}}$ . Then, the optimum profile of relative dielectric constant is obtained based on desired requirements. As an experimental prototype, IPL is discretized to seven different homogeneous layers. Equivalent medium theory (EMT) is used to realize necessary dielectric constant of each layer by using asymmetric Brugemans’s (ABG) formula. The radome layers are fabricated by perforating numerous holes in the host material with diameters sufficiently smaller than the wavelength. Properly sticking different layers together forms the final radome geometry. The designed radome is verified with free-space measurement setup. The measurement results are in good agreement with simulations and shows low insertion loss and high stability in wide aspect angle range for both TE and TM polarizations.

Learning Capsules for SAR Target Recognition

Deep learning has been successfully utilized in synthetic aperture radar (SAR) automatic target recognition tasks and obtained state-of-the-art results. However, current deep learning algorithms do not perform well when SAR images are occluded, noisy, or with a great depression angle variance. This article proposes a novel method, SAR capsule network, to achieve the accurate and robust classification of SAR images without significantly increasing network complexity. Specifically, we develop a convolutional neural network extension based on Hinton's capsule network to capture spatial relationships specialized in classification between different entities in a SAR image. The SAR capsules are learned by a vector-based full connection operation instead of the traditional routing process, which not only alleviates the computational burden but also improves recognition accuracy. For occlusion, additive noise, and multiplicative noise tests, SAR capsule network shows superior robustness compared with typical convolution neural networks. When missing training data in a certain aspect angle range or existing a large depression angle variance between training data and test data, the proposed network achieves better performance than the existing works and reveals some competitive advantages in several test scenarios.

Multilevel Physical Optics Algorithm for Near-Field Double-Bounce Scattering

A fast algorithm for the evaluation of the double-bounce (DB) contributions to the physical optics scattering integrals, over a range of aspect angles and frequencies, is presented. The work extends the preceding far-field algorithm, to encompass three-dimensional and near-field scenarios. The algorithm relies on multilevel sampling and interpolation of phase- and amplitude-compensated contributions of subdomain pairs. A particular design of the phase- and amplitude-compensation functions and sampling grids, tailored to the DB near-field case, is presented. The improved performance and error controllability are demonstrated via representative examples.

Correlation trends in Naval Surface Warfare Center Panama City Division’s database of simulated and collected target scattering responses focused on automated target recognition

In recent years, NSWC PCD has assembled a database of sonar scattering responses encompassing a variety of objects including UXO, cylindrical shapes, and other clutter-type objects deployed on underwater sand and mud sediments and inspected over a large range of aspect angles and frequencies. Data available on these objects consist of a simulated component generated with 3D finite element calculations coupled to a fast Helmholtz-equation-based propagation scheme, a well-controlled experimental component collected in NSWC PCD’s pond facilities, and a component of measurements in realistic underwater environments off Panama City, FL (TREX13 and BayEX14). The goal is to use the database to test schemes for automating reliable separation of these objects into desired classes. Here, we report trends observed in an on-going correlation analysis of the database projected onto the target aspect vs frequency plane to clarify the roles of the environment, the data collection process, and target characteristics in identifying suitable phenomena useful for classification. [Work supported by ONR and SERDP.]

The ability to recognize objects from bottlenose dolphin (Tursiops truncatus) echoes generalizes across multiple orientations in humans and neural networks

Object constancy, the ability to recognize objects despite changes in orientation, has not been well studied in the auditory modality. Dolphins use echolocation for object recognition, and objects ensonified by dolphins produce echoes that can vary significantly as a function of orientation. In this experiment, human listeners had to classify echoes from objects varying in material, shape, and size that were ensonified with dolphin signals. Participants were trained to discriminate among the objects using an 18-echo stimulus from a 10° range of aspect angles, then tested with novel aspect angles across a 60° range. Participants were typically successful recognizing the objects at all angles (M = 78 %). Artificial neural networks were trained and tested with the same stimuli with the purpose of identifying acoustic cues that enable object recognition. A multilayer perceptron performed similarly to the humans and revealed that recognition was enabled by both the amplitude and frequency of echoes, as well as the temporal dynamics of these features over the course of echo trains. These results provide insight into representational processes underlying echoic recognition in dolphins and suggest that object constancy perceived through the auditory modality is likely to parallel what has been found in the visual domain in studies with both humans and animals.

Recognizing objects from multiple orientations using dolphin echoes

Object constancy, the ability to recognize objects despite changes in orientation, has not been well studied in the auditory modality. Dolphins use echolocation for object recognition, and objects ensonified by dolphins produce echoes that can vary significantly as a function of orientation. In four experiments, human listeners had to classify echoes from objects ensonified with dolphin signals. Participants were trained to discriminate among the objects using an 18-echo stimulus from a 10 degree range of aspect angles, then tested with novel aspect angles across a 60 degree range. In the first two experiments, the three objects varied in material, size, and shape. Participants were typically successful recognizing the objects at all angles (M = 78%). In experiment 3, the three objects had the same material but different shapes. Participants were often unsuccessful recognizing the objects at all angles (M = 46%). In experiment 4, participants had to classify echoes from four fish species across a wider range of angles (330 degrees). Preliminary results show overall poor performance (M = 45%). These results suggest that object characteristics play a role in whether performance is more view-dependent or view-invariant. These studies can provide insight into the process dolphins use to identify objects.

Fast 3D-ISAR Image Simulation of Targets at Arbitrary Aspect Angles Through Nonuniform Fast Fourier Transform (NUFFT)

We present a method to generate three-dimensional (3D) inverse synthetic aperture radar (ISAR) images of a target at arbitrary aspect angles using the shooting and bouncing ray (SBR) method. We have derived a 3D-image-domain ray-tube integration formula based on the SBR technique. The imaging formula is in a form of convolution, then a nonuniform fast Fourier transform (NUFFT) is induced to achieve the convolution, which can be used to generate 3D ISAR images rapidly. Compared to the conventional algorithm where the ISAR images are obtained by inverse Fourier transforming the computed scattered fields over a range of frequencies and a range of aspect angles, the new formula needs only one-time ray tracing and physical optics (PO) integration over a small angle span. Hence the computational time is decreased tremendously. The ISAR images of a missile target and an armored car for several aspect angles are presented to demonstrate the efficiency and accuracy of the proposed method.

Boundary enhancement of a solid cylinder backscattering response: Experiments and modeling.

The presence of a reflecting boundary modifies the free‐field backscattering response of an object. Boundary reflections increase the number of available paths for scattered sound to return to a source‐receiver. In the present work, high‐frequency tank experiments on solid aluminum cylinders demonstrate a strong boundary related backscattering response that is present for a broad range of aspect angles. To study the effects of close proximity to a flat reflecting boundary, cylinders were suspended in water near the tank’s air‐water interface. Targets were insonified from below at grazing incidence. Monostatic measurements were made as the cylinders were rotated in a plane parallel to the boundary. Some aspects of this response could be understood by modeling the region of water bounded by the cylinder top and the air‐water interface as a quasi‐2‐D waveguide. Sound incident on this region of decreasing water thickness can be reflected as from a seamount. The backscattered amplitude is strongly affected by the threshold frequency for propagation in the gap between the cylinder and the adjacent flat surface. Experiments show that the behavior of this response, as a function of target depth and aspect angle, is in qualitative agreement with this waveguide model. [Work supported by ONR.]

Generalized Framework for Real Aperture, Synthetic Aperture, and Tomographic Sonar Imaging

When an underwater object is insonified over a range of aspect angles, the signals that are backscattered from the object can be sampled in space and time by an array of sensors, which can form either a real or a synthetic aperture. The spatial information contained in the backscattered signals can be processed using inverse transform methods to form an image of the 2-D spatial distribution of the object's acoustic reflectivity function. The signal measurement space (or data space) defines a sector of the polar wave number spectrum that quantifies the variation with aspect angle of the spatial frequency components that contribute to the backscattered signal. The temporal bandwidth of the incident sonar pulse determines the <i xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">radial</i> extent of the sector, while the range of aspect angles over which the object is insonified defines the <i xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">angular</i> extent of the sector. Equivalently, the extent of the aperture that is formed by the sensor positions relative to the object determines the angular extent of the sector. This concept provides a generalized framework that unifies sonar imaging techniques such as reconstructive tomography (image reconstruction from projections), synthetic aperture sonar, and real aperture sidescan sonar. The difference in these techniques is simply due to the azimuthal angle subtended by the aperture of the sensing array, which is typically a fraction of a degree for a real aperture sonar, several degrees for a strip map synthetic aperture sonar, tens of degrees for a spotlight synthetic aperture sonar, and 360deg for a tomographic imaging sonar. Experimental results are presented for a real aperture sidescan sonar, a synthetic aperture sonar, and a tomographic sonar that demonstrate the development of an acoustic image from a single blurred point to a clearly identifiable object as the azimuthal extent of the angle subtended by the sonar aperture is increased.

Radar Signatures of a Passenger Car

Upcoming new synthetic aperture radar (SAR) satellites such as TerraSAR-X and Radarsat-2 offer high spatial image resolution and dual receive antenna capabilities, which open new opportunities for worldwide traffic monitoring applications. If the radar cross section (RCS) of the vehicles is strong enough, they can be detected in the SAR data, and their speed can be measured. For system performance prediction and algorithm development, it is therefore indispensable to know the RCS of typical passenger cars. The geometry parameters that have to be considered are the radar look direction, incidence angle, and vehicle orientation. In this letter, the radar signatures of nonmoving or parking cars are presented. They are measured experimentally from airborne experimental SAR (E-SAR) data, which have been collected during flight campaigns in 2005 and 2006 with multiple overflights at different aircraft headings. The radar signatures could be measured for the whole range of aspect angles from 0 to 180 and with high angular resolution due to the large synthetic aperture length of the E-SAR radar sensor. The analysis for one type of passenger car and particular incidence angles showed that the largest radar cross-sectional values and, thus, the greatest chance of detection of the vehicles appear when the car is seen from the front, back, and side. Radar cross-sectional values for slanted views are much lower and are therefore less suitable for car detection. The measurements have been performed in the -band (9.6 GHz) with VV-polarization, and at incidence angles of 41.5deg and 42.5deg. The derived radar signature profile can also be used for the verification of radar cross-sectional simulation studies.

Measurement of radar signatures of passenger cars: airborne SAR multi-frequency and polarimetric experiment

The launch of SAR satellites with high-resolution and dual-receive antenna capabilities opens new possibilities for traffic-monitoring applications on a global scale. Thus, it will be possible to detect cars and measure their speed from the acquired along track interferometric data. The development of vehicle-detection algorithms requires the knowledge of the radar signatures of vehicles, especially under consideration of the geometry of the radar look direction and the vehicle orientation. The radar signatures of the non-moving cars are presented. They are estimated experimentally from airborne E-SAR multi-frequency and polarimetric data, which have been collected during a flight campaign in 2003. Radar signatures are estimated for a considerable part of aspect angles ranging from 0° to 180°. The large synthetic aperture length of the E-SAR radar sensor allows the look processing of data and therefore allows an increase of the aspect angle resolution. The radar signature analysis for one type of passenger cars showed that the largest radar cross-section values and thus the greatest chance for high probability of detection are for cars standing in rear and front views of radar beam direction. This holds true for all frequencies and co-polarisations. Radar cross-section values for cross-polarisations and diagonal views are much lower and are therefore less suitable for car detection. The radar signature profile over a considerable range of aspect angles in fine resolution can be used further for the verification of simulation studies and for the performance prediction for traffic monitoring with a coming German TerraSAR-X satellite.

Object representation in echolocating bats: Recognition of targets from different orientations

Big brown bats (Eptesicus fuscus) can discriminate among objects using wideband FM sonar sounds. Bats could perceive and represent only the acoustic dimensions of echoes (e.g., delay, frequency, amplitude), or they could use those dimensions to reconstruct object features (e.g., distance, shape, size). To investigate this question, bats were presented with a two-alternative (left/right) forced-choice sonar discrimination task. The stimuli were a one-cylinder monopole target and a two-cylinder dipole target that was presented at all aspect angles. Acoustic dimensions of the dipole echoes change depending on aspect angle. If the bats represent only acoustic dimensions, they should have difficulty selecting the dipole. If the bats represent object features, they should be able to select the dipole independent of aspect angle. Bats can select the dipole over a wide range of aspect angles, suggesting that they construct object features from echo acoustic dimensions. The bats’ error patterns may indicate the echo dimensions they use to construct object features. [Work supported by NIH and ONR.]

Real time electromagnetic target classification using a novel feature extraction technique with PCA-based fusion

This work introduces an efficient technique to design an electromagnetic target classifier whose reference database is constructed using scattered data at only a few aspects. The suggested technique combines a natural-resonance related feature extraction process with a novel, multiaspect feature fusion scheme. First, moderately aspect-variant late-time features are extracted from scattered field of a given candidate target at several different reference aspects using the Wigner transformation to characterize the target's scattered energy distribution over a selected late-time segment of the joint time-frequency plane. Then, these features are fused using the principal component analysis to obtain a single characteristic feature vector that can effectively represent the target of concern over a broad range of aspect angles. The suggested technique is demonstrated to design a classifier that is verified to be highly accurate and robust even in the presence of excessive noise. Due to the computational efficiency of the technique, the classifier needs very small memory space to store the reference information and quite fast lending itself suitable for real-time target classification.

A Fast Physical Optics (FPO) Algorithm for High Frequency Scattering

A novel algorithm referred as the fast physical optics (FPO) for computing the back-scattered field over a range of aspect angles and frequencies is presented. The computation is performed in the framework of the conventional physical optics approximation appropriate for the high frequency scattering regime. The proposed algorithm is, also, directly applicable to fixed angle bistatic configurations and a variety of single scattering formulations. The method comprises two steps. First, a decomposition of the scatterer into subdomains and computation of the pertinent scattering characteristics of each subdomain. Second, interpolation, phase-correction and aggregation of the scattering patterns of the subdomains into the final pattern of the whole body. A multilevel algorithm is formulated via a recursive application of the domain decomposition and aggregation steps. The computational structure of the multilevel algorithm resembles that of the FFT. The proposed method is especially suited for generation of synthetic data for radar imaging simulation.

Backscattering of transients by tilted truncated cylindrical shells: time-frequency identification of ray contributions from measurements.

Impulse backscattering measurements by a thick-walled finite cylindrical shell are examined in the time-frequency domain to identify and characterize individual ray contributions from generalized Lamb waves excited on the shell. Previous experiments and analysis in the frequency-aspect angle domain [S. F. Morse et al., J. Acoust. Soc. Am. 103, 785-794 (1998)] indicate that large backscattering enhancements occur in the midfrequency region for the shell tilted at large angles. Presently this experimental data is examined in the time-frequency domain for selected angles of incidence. Individual ray contributions are evident and their evolution over aspect angle is discussed. The most prominent contribution is due to the meridional ray of the a0 leaky Lamb wave. This feature distinctly highlights the truncation of the shell and is found over a range of aspect angles spanning 200 degrees for the frequencies examined. Also observed are periodic features corresponding to end-reflected helical waves of the a0-. These scattering features are significantly different from those reported for thin-walled finite cylinders at low frequencies. The present results may be useful for target identification and localization and as a comparison tool for high-frequency computational scattering models.

Radar target identification using a combined early-time/late-time E-pulse technique

The E-pulse technique has been applied in the past to both the early- and the late-time components of a transient radar response. While the late-time E-pulse technique uses aspect-independent waveforms, the early-time E-pulse technique requires a separate waveform for each target aspect angle and thus significantly more storage and processing time. This paper discusses a combination of the two techniques that employs the early-time technique to remove ambiguities generated from application of the late-time method. By narrowing the possible range of aspect angles of the potential targets, the early-time technique can be employed more efficiently.

Polarimetric calibration for wideband synthetic aperture radar imaging

Multi-channel techniques are introduced to calibrate for effects of antenna shading and crosstalk in wideband, wide-angle full polarisation radar imaging. The system is modelled as a two-dimensional integral operator which includes the transmit pulse function, receive and transmit antenna transfer functions, and response from scattering objects. Existing imaging algorithms provide an approximate inversion of this integral operator, without compensation for the effect of the multi-channel antenna transfer functions. Thus, standard processing yields poor image quality, due to the inherent variation of the antenna response (in magnitude, phase, and polarisation) across the large band of frequencies and wide range of aspect angles. Two techniques are proposed for regularised inversion of the system model. Both techniques require measurements of the antenna transfer function over the look angles and frequency band of interest. The first technique adopts a local approximation to process multi-channel range lines separately and is suitable for imaging small regions. The second technique achieves multi-channel deconvolution of the antenna response in the transform (Doppler) domain and is suitable for large data records. The result is imagery with improved polarisation purity and a more localised point spread function.

Time-domain imaging of radar targets using algorithms for reconstruction from projections

A time domain bistatic inverse scattering identity based on the Radon transform and the space-time magnetic integral equation is derived for real-time use in short pulse radar systems. It is shown that using a sine-modulated exponential pulse (SILIEP) and algorithms for reconstruction from projections provides an edge-enhanced image. A simulation based on stepped frequency, multi-aspect measurements of aircraft models produces clear images with highly defined edges. Images using data from a restricted range of aspect angles are also demonstrated.

An ambiguity‐function‐based method for analysis of Doppler decompressed radar signals applied to EISCAT measurements of oblique UHF‐VHF meteor echoes

Binary phase modulation using Barker codes as the modulating sequences is often used as a pulse compression technique in incoherent scatter radar measurements of the ionospheric E region. Demodulation of the compressed signal is usually accomplished by matched filtering techniques. If the target has appreciable velocity, the resulting Doppler shift detunes the scattered signal relative to the filter, and the target response spreads over a large range interval. The paper describes the development of a new data analysis method, based on an ambiguity function description of radar total‐power measurements, which uses the characteristic Doppler decompression signatures to identify, extract, and analyze high‐velocity events from conventional Barker‐coded power profile type incoherent scatter (IS) measurements. Data sets recorded with the European Incoherent Scatter UHF (931 MHz) and VHF (224 MHz) radars during the 1990‐1991 Geminid and 1993 Perseid showers are shown to contain many events which are associated with individual meteors crossing the radar beams over a wide range of aspect angles, including a few near‐radial cases. Line‐of‐sight velocities and effective cross sections are derived for some of the best dual radar events. The cross sections are very small and increase with decreasing wavelength, something not observed before. It is suggested that this is indicative of a scattering mechanism different from the one operating at near‐normal incidence at VHF, and Rayleigh scatter from compact “balls” of plasma contained within the meteor coma is proposed as a possible candidate, in qualitative agreement with the observations. Finally, some signal‐processing related aspects of the method are discussed. It is shown that most IS radars should be able to apply it to their existing programs with very little effort and without having to compromise the performance in normal E region applications. The method could therefore become a powerful tool for studies of meteor statistics at UHF frequencies.