Detection and classification of targets in sidescan and other sonar imagery is well advanced.Standard image processing techniques operate on intensity data and usually sonar returns are match filtered to improve echo localisation and signal-to-noise ratio (SNR).These methods allow large amounts of data to be processed quickly and produce good results.Typically detection rates in excess of 90% are attainable with false alarm rates around 5%, though performance declines rapidly over heavily cluttered seabeds.To further increase effectiveness of target identification from sonar data we must find new approaches.This paper considers the contribution of detailed echo analysis using bio-inspired wideband sonar pulses.

In the MCM (Mine Counter Measure) context, a lot of effort has been put in imagery systems and especially in high frequency sonar. The latest generation of sonar, SAS (Synthetic Aperture Sonar) systems, has been developed in the last 15 years and provides a new powerful tool for mine detection, identification and classification. The main advantage of SAS systems is a resolution close to the wavelength even at long range. A high frequency SAS system has been developed at NURC, the MUSCLE vehicle (300kHz centre frequency) but, despite the extremely good quality of the SAS images, ambiguities between mine-like objects cannot always be resolved. This study investigates how low frequency SAS systems (LF-SAS) could address the two main limitations of the present systems linked essentially to the sound absorption at high frequencies: providing imagery inside the target and detecting buried targets. Synthetic SAS images have been generated and studied in order to demonstrate the capability of LF-SAS to provide information of the inside of targets. We demonstrate that the inner resonances of objects are visible and exploitable in these LF-SAS images. Most of the time the reverberation level (RL) of the seafloor is the main limiting factor in the interpretation of sonar images. A computation of the RL has been done using the small perturbation theory on a rough seabed and we demonstrate that the first resonant echoes are still visible relative to the RL.

The sonar of dolphins has developed over millions of years of evolution and has achieved excellent performance levels. Using the excellent performance of the dolphins sonar as an inspiration, bio-inspired wideband acoustic sensing methods for underwater target detection and tracking are being developed. In this study the authors explore what they expect to gain from such a wideband sonar. The systems wideband sensors are based on bottlenose dolphins sonar, covering a frequency band from around 30 to 150 kHz and having a frequency dependent beamwidth considerably larger than that of conventional imaging sonars. The system can be made relatively compact and is suitable for mounting on a variety of platforms including small-scale autonomous underwater vehicles to allow the sonar to operate in a similar way to that used by dolphins. In contrast to high-resolution image processing techniques, detection and classification in the wideband system are based on pattern recognition methods applied to the echo spectra. The authors highlight the properties of the transducers needed for such a system indicating how these properties may be exploited to give improved sonar performance. The results obtained from such a system are presented to illustrate the effectiveness of this approach.

The Ocean Systems Laboratory is developing bio-inspired wideband acoustic sensing methods for underwater target detection and tracking. The wideband sensors themselves are based on bottlenose dolphin sonar, covering a frequency band from around 30kHz to 150kHz and having a frequency dependent beamwidth considerably larger than conventional imaging sonars. The entire system is relatively compact and is suitable for mounting on a variety of platforms including small scale autonomous underwater vehicles (AUVs). In this paper we overview recent efforts applying the sonar to the detection and tracking of various underwater cables, and to the detection and classification of these cables in shallow burial, based on their midwater responses.

New generations of sonars appeared in the last decade. The major interest in SAS systems and high frequency sonars is in the improvement of the sonar resolution and the reduction of noise level. Sonar images are distance-images but at high resolution they tends to appear visually as optical images. Usually the algorithms developed for sidescan were specific for sonar images due to the poor resolution essentially. With high resolution sonars, algorithms developed in the image processing field for natural images became applicable. In this paper we present a real-time and realistic sidescan simulator, and test image-based classification algorithms (such as PCA and eigenface algorithms) with synthetic images in order to characterize the precision necessary for these image-based algorithm to work.

This paper presents a framework for registering and fusing classified sidescan sonar data. It builds on recent advances in navigation and registration for improved mosaicing, applying novel fusion algorithms to integrate data from overlapping sidescan survey lines to produce large scale classified mosaics. While typical mine-counter-measures (MCM) and rapid environmental assessment (REA) missions provide various over-lapping views of the same region of seafloor, research on sidescan image analysis has traditionally concentrated on the analysis of individual images. The available information from the other images, relating to the same region of seafloor, is generally not considered. The image registration and mosaicing process allows this complementary data to be fused, producing an improved final classification result. The sidescan imagery is first pre-processed through the application of advanced radiosity correction algorithms. Following radiosity correction, texture segmentation for the data presented in this paper is achieved using features derived from the averaged normalised power spectral density. The individual classification maps are georeferenced and coregistered using a Concurrent Mapping and Localisation Rauch-Tung-Striebel (CML-RTS) procedure. This uses local landmarks within the individual images and the AUVs navigation data to generate a more accurate and smooth navigation trajectory. This trajectory is used to produce the registered classification mosaics. The coregistered classification results are then fused to produce an improved class mosaic for the entire survey region. The fusion model uses a voting scheme to initialize the seafloor map after which a Markov random field (MRF) model is used to produce the final fused classification mosaic. The entire process (classification, registration and fusion) is demonstrated on real sidescan data taken at the Saclant Centre, La Spezia, Italy.

Sidescan sonar images characteristically display across-track intensity variations resulting from the transducers' beam pattern. Along-track these changes are complicated by intensity variations introduced by changing sonar altitude. The effects are particularly apparent in low altitude surveys displaying strong sidelobe returns from regions closest to the transducers. The fall off in intensity with range for main lobe returns is also more noticeable in these missions. A time-varying gain (TVG) is usually applied to sonar signals to compensate for this range-dependent effect, but cannot adequately compensate for beam pattern or altitude variations. In this paper methods are presented for correction of beam pattern and residual TVG effects in the presence of altitude change. Estimation of correction factors is performed directly from the data. It is preferable to seek an estimate for these factors over a seabed region with relatively even, small-scale texture, a few hundred lines of data being sufficient. The approach to the beam pattern problem uses time-scaled representations of each line of data to account for geometrical variations introduced by altitude change. The applied TVG function is assumed to be dominant towards the end of each line of recorded data and can usually be approximated using a smooth function. Multiplicative correction factors are calculated for the TVG function and scaled beam pattern. These are applied to the sonar data to produce a representation of the ensonified seafloor topography with intensities adjusted relative to some user defined datum point.

The fractional Fourier transform (FrFT) provides an important extension to conventional Fourier theory for the analysis and synthesis of linear chirp signals. It is a parameterised transform which can be used to provide extremely compact representations. The representation is maximally compressed when the transform parameter, alpha, is matched to the chirp rate of the input signal. Existing proofs are extended to demonstrate that the fractional Fourier transform of the Gaussian function also has Gaussian support. Furthermore, expressions are developed which allow calculation of the spread of the signal representation for a Gaussian windowed linear chirp signal in any fractional domain. Both continuous and discrete cases are considered. The fractional domains exhibiting minimum and maximum support for a given signal define the limit on joint time-frequency resolution available under the FrFT. This is equated with a restatement of the uncertainty principle for linear chirp signals and the fractional Fourier domains. The calculated values for the fractional domain support are tested empirically through comparison with the discrete transform output for a synthetic signal with known parameters. It is shown that the same expressions are appropriate for predicting the support of the ordinary Fourier transform of a Gaussian windowed linear chirp signal.

The fractional Fourier transform (FrFT) provides a valuable tool for the analysis of linear chirp signals. This paper develops two short-time FrFT variants which are suited to the analysis of multicomponent and nonlinear chirp signals. Outputs have similar properties to the short-time Fourier transform (STFT) but show improved time-frequency resolution. The FrFT is a parameterized transform with parameter, a, related to chirp rate. The two short-time implementations differ in how the value of a is chosen. In the first, a global optimization procedure selects one value of a with reference to the entire signal. In the second, a values are selected independently for each windowed section. Comparative variance measures based on the Gaussian function are given and are shown to be consistent with the uncertainty principle in fractional domains. For appropriately chosen FrFT orders, the derived fractional domain uncertainty relationship is minimized for Gaussian windowed linear chirp signals. The two short-time FrFT algorithms have complementary strengths demonstrated by time-frequency representations for a multicomponent bat chirp, a highly nonlinear quadratic chirp, and an output pulse from a finite-difference sonar model with dispersive change. These representations illustrate the improvements obtained in using FrFT based algorithms compared to the STFT.

ó Polar regions, especially sensitive to small changes in temperature, play a key role in global climate change. Scientists are interested in evaluating the decline in local ice production. We will attempt to classify automatically First Year (FY) ice, Multi Year (MY) ice and Deformed ice using Sidescan sonar images of the Arctic ice-shelf. We use 4-bit data (16 grey levels) ground truth images provided by an expert as a starting point of the study. These images have been obtained using a Sidescan sonar pointing upward. Our methods use rst order statistics extracted from local areas of the image. The local histogram is tted to three pdfs (Rayleigh, Log-normal and Gaussian distributions) whose parameters are extracted. A Chi-square test is used to evaluate the quality of the t. The parameters are then used to classify the regions. The results obtained show that FY and MY ice follow Rayleigh or Log-normal distributions whereas Deformed ice is more like a Gaussian distribution. We have created a new method to classify ice types selecting the best tting for each region. A classication of three classes (FY, MY and Deformed ice) is achieved with rst order statistics. In future work we will investigate the potential of second order statistics to improve the classication. I. INTRODUCTION