Calibrated processing for improved synthetic aperture sonar imagery

 

Blair Bonnett, Holger Schmaljohann and Thomas Fickenscher

 

Citation: Proc. Mtgs. Acoust. 47, 070007 (2022); doi: 10.1121/2.0001599

 

View online: https://doi.org/10.1121/2.0001599

 

View Table of Contents: https://asa.scitation.org/toc/pma/47/1

 

Published by the Acoustical Society of America

 

ARTICLES YOU MAY BE INTERESTED IN

 

Experimental investigation of a virtual planar array for MIMO sonar systems

 

Proceedings of Meetings on Acoustics 47, 070005 (2022); https://doi.org/10.1121/2.0001591

 

Exploring the use of AI in marine acoustic sensor management

 

Proceedings of Meetings on Acoustics 47, 070008 (2022); https://doi.org/10.1121/2.0001601

 

Comparison of two different methods to include a beam pattern in parabolic equation models

 

Proceedings of Meetings on Acoustics 47, 070006 (2022); https://doi.org/10.1121/2.0001593

 

Acoustic attenuation of cohesive sediments (mud) at high ultrasound frequencies

 

Proceedings of Meetings on Acoustics 47, 070009 (2022); https://doi.org/10.1121/2.0001594

 

Improving the realistic rendering of artificial sonar images using Cycle Generative Adversarial Networks

 

Proceedings of Meetings on Acoustics 47, 070010 (2022); https://doi.org/10.1121/2.0001598

 

Utilizing imaging geometry meta-data in classification of synthetic aperture sonar images with deep learning

 

Proceedings of Meetings on Acoustics 47, 070011 (2022); https://doi.org/10.1121/2.0001607

 

 

 

Calibrated processing for improved synthetic aperture sonar imagery

 

Blair Bonnett

 

Hochfrequenztechnik, Fakultät für Elektrotechnik, Helmut-Schmidt-Universitat Universitat der Bundeswehr Hamburg, Hamburg, 22043, GERMANY; blair.bonnett@ieee.org

 

Holger Schmaljohann

 

Bundeswehr Wehrtechnische Dienststelle für Schiffe und Marinewaffen Maritime Technologie und Forschung: Bundeswehr Wehrtechnische Dienststelle fur Schiffe und Marinewaffen Maritime Technologie und Forschung Kiel, Schleswig-Holstein, 24148, GERMANY; HolgerSchmaljohann@bundeswehr.org

 

Thomas Fickenscher

 

Hochfrequenztechnik, Fakultät für Elektrotechnik, Helmut-Schmidt-Universitat Universitat der Bundeswehr Hamburg, Hamburg, 22043, GERMANY; tfi@hsu-hh.de

 

Uncalibrated synthetic aperture sonar (SAS) images provide the amount of energy backscattered by an area of the seafloor relative to the other areas of the seafloor. If the transmitters, receivers and processing chain can be calibrated, then the images will yield the actual scattering strength of each portion of the seafloor instead of the relative strength. This permits investigation of the physical properties of the seafloor such as the sediment type, composition, density etc and allows images from different missions or frequency bands to be easily compared. In a previous publication we discussed some initial work into achieving calibrated imaging. In this work, we demonstrate an improved calibration chain which improves the image quality from that previous work. Compensation of the three main physical effects causing losses (geometric spreading, acoustic attenuation and the transducer beampatterns) is validated with point target simulations. A SAS mission imaging the same patch of the seafloor from different tracks was performed. After applying our processing chain, the scattering strength of the seafloor shows close agreement between measurements from different tracks. Although this scattering strength does not yet match predictions from a standard model, the results show the benefits of a fully calibrated processing chain.

 

1. INTRODUCTION

 

Synthetic aperture sonar (SAS) systems allow high-resolution images of the seafloor to be captured. In their basic form, the amplitudes of the pixels provide locally relative information: the amount of energy scattered from one point within the imaged area can be measured relative to the amount of energy scattered from a different point within the imaged area. Due to spatially variant sources of energy loss during data collection, such a comparison can only be made within a spatial region close to the point of interest. If these effects that attenuate or otherwise impact the amplitude of the generated image can be compensated, then a calibrated image can be formed where the amplitudes yield the actual scattering strength of each point in the scene.

 

There are a number of effects that need to be compensated. Acoustic waves moving through the water spread out and are attenuated, and various properties of the system (beampatterns, source level, receiver sensitivity) impact the image. Most of these will vary with the frequency being used to collect the data. With multi-frequency SAS systems becoming more common, comparing data collected in different operating bands is an obvious step in image analysis. Having calibrated images available for such analysis is vital to ensure only the target properties are being compared, not system or medium properties.

 

In a previous publication, we presented our initial work into generating calibrated SAS images. This pa- per details some subsequent improvements to this compensation method. Section 2 compares our previously published results to the latest processing outputs to demonstrate recent improvements. The types of compensation that are applied within our processing chain are detailed in Section 3, including some verification of their operation with simulated data. Section 4 presents the results from a mission designed to evaluate the current performance of the calibration. These results are discussed in Section 5 before concluding remarks in Section 6.

 

2. RECENT IMPROVEMENTS

 

In an earlier publication, 1 we detailed our initial attempts to produce calibrated SAS imagery. As part of this, we presented some images generated from data collected on a mission in the Bay of Lübeck near the town of Neustadt in Northern Germany. The mission used an Atlas Elektronik SeaOtter autonomous underwater vehicle (AUV) with two Atlas Elektronik Vision1200 SAS sensors, one imaging scenes on the port side of the AUV and the other imaging scenes on the starboard side. The Vision1200 has three operating bands, low frequency (LF, 20 kHz), medium frequency (MF, 75 kHz) and high frequency (HF, 150 kHz); any two of these bands can be used simultaneously. Each band has a vertically separated pair of receiver arrays allowing interferometry to be performed. On this mission, data was collected with the LF and MF bands.

 

After quality control checks, motion compensation using the standard displaced phase centre antenna (DPCA) algorithm was performed to correct the navigation data, and then the hydrophone data was beamformed and backprojected onto the imaging grid with various compensating factors applied during the back-projection. The images from the previous publication are shown in Fig. 1a and Fig. 1c for the LF and MF operating bands, respectively.

 

After some improvements to our processing chain, the data from this mission has been reprocessed; the new images are presented in Fig. 1b and Fig. 1d for comparison. The most obvious improvement can be seen in the MF imagery, where the bands of low intensity observed in the previous publication have been largely corrected. As proposed in that publication, these were due to incorrect compensation of the beampattern; while mapping the beamformed data onto the imaging grid, an incorrect angle was being used to estimate the compensation factor. The other major change is the conversion of the raw hydrophone data into Pascals prior to processing. Previously, the processing was performed in the arbitrary units used by the system when recording the data. Atlas Elektronik provided a set of receiver sensitivity measurements which now allow us to convert these into the pressure incident on the face of the transducers. This is performed in the frequency domain to compensate for the frequency-dependent sensitivity of the receivers. Some other minor changes to the compensation scheme have been implemented, including increased upsampling of data prior to interpolation to reduce errors introduced by the interpolation.

 

 

Figure 1: SAS intensity images of the seafloor at Neustadt captured with the LF and MF bands of the Vision1200, generated with the old and new compensation schemes. The vehicle was travelling up the page with the starboard SAS used to capture the data. The image is approximately 900 m in along track and 80 m in across track. The seafloor is mostly muddy with the bright patch in the middle assumed to be a sandbank. The bright target in the LF band is buried, and hence is significantly less obvious at MF where less energy penetrates into the sediment.

 

 

3. CALIBRATION CHAIN

 

The receiver sensitivity is taken into account when loading the hydrophone data for processing. After an image has been reconstructed, it can be divided by the source level (transmitter strength) to convert the values into the fraction of incident energy that was scattered back to the sonar. The remaining effects that need to be compensated during image formation are the geometric spreading (Section 3.A), the acoustic attenuation (Section 3.B), and the beampatterns of the transmitter and receiver (Section 3.C).

 

To validate each type of compensation, data exhibiting the corresponding loss was generated by a simple point-scatterer simulator. A number of point targets were placed 10 m below the sonar at different across track positions. The scattering strength of each target was set to −10 dB. After processing with the compensation method under test, the scattering strengths measured from the image at the target locations can then be easily checked against this expected strength. The results of these tests are shown in Fig. 3, and the behaviour of each type of compensation will be discussed in the corresponding section.

 

A. GEOMETRIC LOSSES

 

The intensity of an acoustic wave decreases as the size of the wavefront increases while moving away from the source. A common assumption in imaging sonars is that the wavefront spreads spherically, leading to a one-way reduction in amplitude proportional to 1 /r where r is the distance from the source. After scattering off a target, the reflected wave also spreads on its return journey, leading to a two-way spreading loss of 1 /r_tx r_rx ≈ 1/r² in amplitude, where r_tx and r_rx are the ranges from the transmitter to the target and from the target to the receiver, respectively.

 

The backprojection algorithm commonly used to generate SAS images has an inherent one-way compensation for spreading losses due to the integration over the aperture. In practice, the aperture is discretely sampled and the integration is replaced by a summation. As a result, the inherent compensation is not smooth. An example of the number of pings used to form the synthetic aperture to reconstruct different points within an HF image is shown in Fig. 2. The diamond patterns are caused by this discrete sampling: one point may appear in N pings, but the point at the next along track position may be out of the main lobe of one of these pings, and so only appear in N − 1 pings despite being at the same broadside range to the sonar. The motion of the vehicle may also cause some inconsistencies in this compensation. In particular, changes in orientation, such as yaw, may cause points at the same broadside range to appear in different numbers of pings. To avoid this, our processing applies a two-way multiplicative range-varying compensation factor

 

 

to the data being backprojected. After image reconstruction is complete, the number of pings forming the synthetic aperture for each pixel is counted, and the image is normalised by these values to remove the inherent discrete compensation described above. Figure 3a shows the affects of geometric spreading on the test scene, and demonstrates that the compensation scheme corrects for these losses.

 

B. ACOUSTIC ATTENUATION

 

As the pulse travels through the water, some of its energy is attenuated. A number of models of varying complexity describing this attenuation have been published. We use the model developed by Ainslie and McColm² which considers three components: viscous drag, the chemical relaxation of boric acid, and the chemical relaxation of magnesium sulphate. The latter occurs with molecules which can exist in more than one stable state in water.

 

 

Figure 2: The number of pings used to form the Vision1200 HF synthetic aperture as a function of position within the final image for one track of the harbour mission described in Section 4.

 

Table 1: Attenuation coefficients and relaxation frequencies as a function of salinity S (ppt), temperature T (°C), pH and depth D (km).

 

 

The change in pressure from an acoustic signal changes the equilibrium point between the states, and the resulting change of molecules between states absorbs some of the energy. This mechanism dominates at low frequencies, characterised by a relaxation frequency or ‘knee point’ where its attenuation transitions from proportional to f² to proportional to f⁰. At higher frequencies, the pressure cycles too rapidly for the molecules to react further, and the viscous drag dominates. In this model, the total attenuation at a frequency f is given by

 

 

where the coefficients and relaxation frequencies are given in Table 1, and with all frequencies being given in kHz. This total attenuation is proportional to the square of the frequency. To compensate for the acoustic attenuation, a multiplicative range-varying compensation factor

 

 

is applied to the data before backprojection. In our standard operating environments, only the frequency and salinity have any noticable effect on the overall attenuation, and so in our compensation scheme the other parameters used to calculate the values in Table 1 are treated as constants, specifically T = 10 °C, D = 0 m and pH = 8 . The frequency is set to the centre frequency of the operating band. The behaviour of this compensation is shown in Fig. 3b. As with the geometric spreading, the acoustic attenuation can be fully compensated if the model parameters are accurate.

 

 

Figure 3: Demonstration of the compensation of various effects using simulated data with discrete point scatterers. The true target strength was −10 dB, i.e., the compensated peaks (drawn in blue) will be at −10 dB if the effect is fully compensated.

 

C. BEAMPATTERN

 

In the far field, the directivity of a transducer can be described by its beampattern. This is proportional to the Fourier transform of the shape of the transducer. For a rectangular transducer sitting in the yz plane with dimensions D_y × D_z , the beampattern is given by

 

 

where    is a unit observation vector pointing from the centre of the transducer in the direction of evaluation and λ is the wavelength. A naive method of compensation would be simply to invert the estimated beampattern and to use that as a multiplicative compensation factor. However, as can be seen from Eq. (4), the beampattern will drop to zero for some observation vectors. Around these nulls, the naive compensation factor would be very large, and, since the collected data will be dominated by internal noise at these points, would lead to a degradation in the signal-to-noise ratio (SNR). Instead, we clip the beampattern to a lower threshold ϵ before inverting, giving a compensation factor

 

 

which has an upper bound on the amplification around the nulls. The wavelength corresponding to the centre frequency of the operating band is used in calculating this factor.

 

The behaviour of this compensation is demonstrated in Fig. 3c. The dashed line shows the combined vertical beampatterns of the transmitter and receiver clipped to an ϵ = 20 dB threshold and projected onto the target positions. The shape of the uncompensated target strengths is a close fit to this as the impact of the horizontal beampatterns is limited due to averaging from the integration over the aperture during processing. It can be seen that the compensated target strengths reach the correct −10 dB value within the central part of the beam, and are limited to a 20 dB increase where the thresholding takes affect. Lowering the threshold would allow a full correction of the losses over a wider swath, but at the cost of a decreased SNR. The best threshold will depend on the system used to collect the data and the application of the images. In many cases, image formation and analysis is restricted to some portion of the main lobe, and so the compensation threshold could be chosen to fully correct the beampattern within this region.

 

4. CALIBRATION MISSION

 

To further validate the compensation, a second mission with the same SeaOtter AUV and Vision1200 SAS was performed. This took place in the harbour of the Marine Arsenal in Kiel with the mission geometry shown in Fig. 4. An area of relatively flat seafloor was selected and 21 tracks past this scene were planned at a variety of across track ranges and heights above ground. Each track was followed twice, once in either direction so that both the port and starboard sensors were used. For this mission, the MF and HF operating bands were used. Regular conductivity-temperature-depth (CTD) readings were performed to determine the environmental parameters for processing.

 

After motion compensation with the DPCA algorithm, the hydrophone data from each track was beamformed and then backprojected onto a common imaging plane applying all the previously described compensation methods. Examples of the SAS intensity images are shown in Figs. 5a and 5b for the HF and MF bands respectively. Two different regions are marked on these images which will be used for subsequent analysis. Sidescan images were formed from the HF hydrophone data with interferometry used to generate an estimate of the bathymetry of the scene, shown in Fig. 5c. This confirms that the seafloor is relatively flat, especially in the selected smaller regions.

 

 

Figure 4: The cross-sectional geometry of the calibration mission drawn to scale. Each of the 21 tracks followed by the vehicle is marked with a dot and is perpendicular to the page. The across track extent of the full reconstructed images is shown with the horizontal sand-coloured line; the across track coordinates are relative to the central track. The patch of the scene marked in red is used for closer analysis. The null-to-null vertical beampatterns of the HF and MF transmitters are shaded in blue and orange respectively for one of the tracks.

 

 

Figure 5: (a, b) SAS intensity images of the seafloor insonified during the calibration mission and (c) bathymetry estimated from interferometry performed on corresponding sidescan images. This example was imaged with the starboard sensors with the vehicle 10 m to the left of the scene and 7 m above ground. The scene is approximately 380 m in along track and 65 m in across track. The seafloor is muddy and largely flat with a few scattered objects. The red rectangles mark patches used for Figs. 6 and 7, and the yellow rectangles mark areas used for Fig. 8.

 

The rectangular patch marked in red in Figs. 5a and 5b, with an extent of 20 m in along track and 13 m in across track, was extracted from the reconstructed images. The patches from the starboard sensors, annotated with the grazing angle and slant range to the centre of the patch, are shown in Figs. 6 and 7 for the HF and MF bands, respectively. The MF band has a stronger response at short ranges due to the additional tilt of the corresponding transmitter which can be seen in Fig. 4. The MF patches suffer from multipath effects at the higher ranges, manifesting as blurring and duplication of the target. Apart from the patches outside the corresponding main lobes, the intensity of the images follows the expected trend of decreasing as the grazing angle becomes shallower. Patches observed with similar grazing angles but different ranges have similar intensities, indicating good compensation for range-varying effects (geometrical spreading and acoustic attenuation).

 

To evaluate the scattering strength given by the reconstructed images, the slice marked in yellow in Figs. 5a and 5b was extracted, the along track mean was calculated and the results were remapped as a function of grazing angle. These results are shown in Figs. 8a and 8b for the HF and MF bands, respectively. Note that, due to a fault with the MF transmitter on the port side, only traces from the starboard side are shown in Fig. 8b.

 

The expected scattering strength of the sediment was estimated using the monostatic small-roughness perturbation approximation with a fluid sediment model.³ From the CTD readings, the mean speed of sound in the centre of the water column, c = 1450.7 m/s , was calculated and used in the model. The other parameters of the model were set to typical values for a muddy sediment,⁴ namely speed ratio ν_p = 0.98 , density ratio a_ρ =1.5 , loss parameter δ_s = 0.01 , and a power law roughness spectrum with spectral strength w₂ = 0 . 0001 and spectral exponent γ₂ = 3. These estimates are plotted for both operating bands in Fig. 8; in both cases, a significant offset can be seen between the measured values and those predicted by the model.

 

5. DISCUSSION

 

The changes in the compensation scheme have resulted in a clear improvement to reconstructed SAS images compared to our previous paper. Particularly noticable is the correction of the beampattern compensation which caused the intensity banding in Fig. 1c. Although some small areas with lower intensity can still be observed at short ranges, these areas are largely from data collected outside the mainlobe of the system. The individual factors being compensated have been shown to work using point-scatterer simulations exhibiting each effect independently. In all three cases (geometric spreading, acoustic attenuation and transducer beampattern), the described compensation schemes correctly recover the target strength. In the case of geometric spreading, only the range to the position being reconstructed is required, and thus the limiting factors are the accuracy of the navigation and the bathymetry. The acoustic attenuation additionally requires knowledge of the environmental parameters to estimate the attenuation factor. To compensate the beampattern, accurate measurements of the transducer beampatterns is required as is a suitable threshold value. When applied to the real-world data collected in the described calibration mission, the compensation does a good job of relatively compensating the images: the same patch imaged from different tracks has a similar intensity when the expected differences due to the incident grazing angle are taken into account. However, the scattering strengths differ significantly from standard scattering models. Although the exact parameters of the sediment are not known, to make the model fit the measured values would require using extremely unrealistic values. This suggests there are still some uncompensated effects, which could be either physical effects related to the movement and scattering of the acoustic energy, or due to the way the returning wave is captured, digitised and stored. A difference between the scattering strengths observed with the port and starboard sensors can be seen in Fig. 8a, suggesting there is still some uncertainty in the system parameters used during compensation.

 

 

Figure 6: SAS intensity images of a patch of extent 20 m in along track and 13 m in across track reconstructed from data captured with the starboard HF sensor on the different missions. Each image is annotated with the grazing angle and slant range of the centre of the patch. Note that the navigation data from different missions was not co-registered, and so the position of the object appears to vary slightly between missions.

 

 

Figure 7: SAS intensity images of the same patch as Fig. 6 reconstructed from data captured with the starboard MF sensor. Note that the intensity scale differs from Fig. 6 as the HF and MF transmitters emit signals of different strengths.

 

 

Figure 8: Measured scattering strength as a function of grazing angle. The solid lines are from data recorded with the starboard SAS, and the dotted lines from data recorded with the port SAS. The red line is the strength predicted by a small-roughness perturbation model. Only grazing angles within the 6 dB vertical beamwidth of the system are shown — note the different values on the horizontal axes of the two sub-figures due to the HF and MF transmitters being tilted at different angles.

 

Another topic that requires further investigation is the frequency dependence of the various loss mechanisms and the corresponding compensation. At higher operating frequencies, the transmitted signals are relatively narrowband and effects like acoustic attenuation will be relatively constant over the operating bandwidth. For low frequency imaging, systems tend to be wideband (the Vision1200 LF band has a Q factor close to 1, for example) and so the attenuation, beamwidth etc. will vary significantly over the bandwidth. Further investigation of the impact of this frequency dependence, and how to efficiently compensate it, is planned. It should also be noted that the current processing only considers far-field effects. Although this is likely to be sufficient for most systems in real-world operation, the impact of near-field effects, in particular on the beampattern compensation, warrants further study.

 

6. CONCLUSION

 

Fully compensating a SAS processing chain allows images with the actual scattering strength of the scene to be generated. An immediate benefit is the ability to estimate target parameters such as material composition. This also allows more valid comparisons between images of a scene captured in different operating bands where the system behaviour (source level, beampattern etc.) and propagation of the acoustic energy differ. Without compensation, images typically need to be normalised in some way before comparing, e.g., by subtracting the local mean, which may introduce further errors or cause a loss of information.

 

In this paper, we have shown some improvements to our previously reported compensation scheme and tested their performance with some real-world data, showing good relative compensation between images of the same scene captured from different tracks. Further investigation is required to determine the remaining sources of energy loss and how to compensate them.

 

REFERENCES

 

1. B. Bonnett, H. Schmaljohann, T. Fickenscher, and U. Herter, “Seafloor segmentation using multi-band SAS: initial results,” Proc. Meet. Acoust. 44 (1), 070015 (2021).
2. M. A. Ainslie and J. G. McColm, “A simplified formula for viscous and chemical absorption in sea water,” J. Acoust. Soc. Am. 103 (3), 1671–1672 (1998).
3. D. R. Jackson and K. B. Briggs, “High-frequency bottom backscattering: Roughness versus sediment volume scattering,” J. Acoust. Soc. Am. 92 (2), 962–977 (1992).
4. D. R. Jackson and M. D. Richardson, High-frequency seafloor acoustics (Springer, 2007).