WO2001094976A1

WO2001094976A1 - Active ultrasound scanner

Info

Publication number: WO2001094976A1
Application number: PCT/EP2001/004917
Authority: WO
Inventors: Eckard Glaser; Janet Grassmann
Original assignee: Sonem Gmbh; Wrobel, Miroslaw
Priority date: 2000-06-05
Filing date: 2001-05-02
Publication date: 2001-12-13
Also published as: DE10027828A1

Abstract

The invention relates to an active ultrasound scanner, which renders objects in a medium recognisable, which are obscure, not visible through a curtain of smoke or mist, or are, for example, in dirty water (harbour basin, river water). The active ultrasound scanner comprises at least one transmitter for the broadcasting of a transmission signal with a known modulation function and a signal bandwidth dependent on the medium, at least one corresponding receiver in a known position, with a first signal processing unit for spectral analysis of the response signal received by the receiver from the medium and a memory shift register in series, a second signal processing unit for determining spatial co-ordinates of the reflection points and a display.

Description

Active ultrasonic viewing device

description

The invention relates to an active ultrasonic viewing device which makes objects recognizable in a medium, which are enveloped in darkness, are not visible through a curtain of smoke or fog, or e.g. in cloudy water (harbor basin, river water).

In technology, it is always necessary in different areas of application to have a viewing device available with which objects that are invisible to humans can be made visible. These objects can e.g. not be visible to the human eye in darkness, fog, smoke, in cloudy water or behind a temperature curtain. In many cases there is also the fact that despite the visibility of an object, its distance cannot be estimated. For example, when fighting a fire, this can lead to additional problems.

In the past, infrared cameras, in particular with low light intensification for the visualization of warm objects such as animals or people, have found widespread use. Modern infrared sensors can detect temperature differences of up to 0.5 ° C. Suitable electronics, which are connected downstream of these infrared cameras, can significantly increase the sensitivity of such devices, and even possible lighting in the visible range is of little or no interference. Extensive training is required to be able to evaluate the images exactly. However, these devices are only conditionally or unusable for use in the event of a fire, since the fire can completely shield the objects of interest.

Another known device is equipped with a bolometer. The environment in which a warm object is suspected is scanned with a bolometer and a thermal image is recorded in the mid-infrared (2-5μm). This picture shows temperature differences up to 0.1 ° C. This device can also make objects in fog, smoke or other gases visible, but not objects in a fire or under water.

Another method of making objects visible uses ultrasound. An ultrasound signal is emitted and from the runtime until the response pulse is received

Distance to an object determined. As is known, the ultrasound propagates in air at approx. 340 m / s, so an object at a distance of 300 m is recognized after approx. 2 s. Please note that only the closest object is recognized.

Another method sends signals that are similar to those of the bat. The signals called chirp contain a pattern of frequencies. The response signals are then compared with the transmission signals and the so-called “beat frequencies” are filtered out. With these devices, all objects in the field of vision can be recognized. These devices were preferably developed for blind people who, through intensive training on the “ Hearing "of obstacles.

The human brain is also in demand in ultrasound spectroscopy as it is used for non-destructive material testing. Here the spectra of the echo signals are displayed on the screen, and the viewer can see from his experience whether there is, for example, a crack or a corrosion site or the like in the sonicated material. While the transit times of the individual frequencies are evaluated in the case of the former, the frequency shifts of the signal pattern are detected in the latter, which can be found in the response signal when reflected on an object. The frequency shift is proportional to the delay between the transmitted and the received signal. The distance of the objects can also be determined in this way.

In order to use this procedure the character and form of a

To determine the object, one tries to take advantage of the fact that each object has a specific reaction to the signal. The characteristics of the object should then be determined from the differences in sound or the differences in frequency patterns of the response signals. Here, too, a complex learning system is required to “hear” the objects or to recognize them from typical patterns.

Solutions are also known in which an attempt is made to classify the object from the reflections on an unknown object. However, the classification of the objects is associated with a very high level of uncertainty, which is currently still given as 30-40%. These devices are intended to help blind people find their way. These devices are also unsuitable for fire fighting or underwater use, since at least one spatial presentation of the object must be provided.

It is therefore the object of the present invention to propose a device which delivers a dimensionally accurate image in 1D, 2D or 3D even in an opaque medium. The device should have a compact structure and be usable for fire fighting, underwater use or as a night vision device. Another task is to propose an active ultrasonic viewing device that provides a 3D representation of objects in real time.

The object is solved by the appended claims.

The active ultrasonic viewing device according to the present invention consists of at least one transmitter for transmitting a transmission signal with any known modulation function and a medium-dependent bandwidth of the signal, at least one receiver arranged in a known position, to each of which a first signal processing unit for n-channel spectral Decomposing the response signals received by the receivers from the medium and a memory shift register is arranged downstream of a second

Signal processing unit for determining the spatial coordinates, and a display. The necessary bandwidth is submerged about 1 kHz up to 10 MHz, in ^'air 20-500 kHz. In solid materials, it must be set so that the signals can penetrate the medium and are reflected or absorbed by any embedded structures. The number n of channels must be determined beforehand. With this number «(eg n = 1024) there is an upper limit for the

Given the resolution of the image.

In an advantageous embodiment, the receivers are arranged symmetrically to the transmitter.

The receiver or receivers is followed in a known manner by signal conditioning, which can consist, for example, of an amplifier and a filter bank. In addition, the display device also has a time control in a known manner. A mixer for generating the beat frequency between the transmission signal and the response signals can be arranged downstream of the receivers. This makes it possible to reduce the frequencies to be processed.

The response signals can be processed both analog and digital. For digital processing, an A / D converter is provided in front of the first signal processing units for spectral decomposition.

After determining the spatial coordinates of the reflection points in the second signal processing unit, they can be sent directly to a display and the gray values of the reflection points can be displayed. However, it is also possible to assign a classification unit to the memory shift register. This classification unit is a processor (for example a neural network), which first records the spectra of the individual channels in parallel from the memory shift register. The classification unit is thus supplied with n (for example n = 1024) values with each time step. The classification unit can consist of a preprocessing unit, the actual classifier and an analysis unit. The preprocessing unit has the task of summarizing the data to be classified in advance, normalizing, selecting, or the like. The classifier consists of a processor which converts the data coming in from the preprocessing unit on the basis of a calculation rule, that is to say a mathematical model. The calculation result indicates, for example, a probability that the data received belong to a structure of a certain class. The underlying calculation rule (i.e. the mathematical model) is optimized in the course of a so-called training phase so that the best possible automatic assignment of the data to specified classes is achieved. The analysis unit now has the task of collecting the received data (e.g. the n variables) to sort out according to their importance for the optimal class allocation.

In this classification unit, the properties of the reflection points can be evaluated and classified, or only the reflection points with the same spectra can be summarized and displayed.

For analog processing, the first signal processing unit for spectral decomposition can be an acousto-optical processor, a holographic wavelet processor or a similar analog processor for spectral decomposition.

For digital processing, the first signal processing unit for spectral decomposition can use a Fast Fourier

Processor, a digital wavelet processor or a similar digital processor for spectral decomposition.

The classification unit can represent a statistical classifier, such as a neural network, a support vector machine or the like.

The second signal processing unit for determining the spatial coordinates, in particular for digital processing, consists of an adaptive grid reconstruction processor, which is also the subject of the present invention. The adaptive grid reconstruction processor consists of a raster generator, a unit for determining the paths of the signals from the transmitter to the raster points and from the raster points to the individual receivers for each raster point, a memory line for the sums of all amplitude values belonging to the receiver the channels of the n-channel spectral decomposition, a sampling address generator for determining the address of the amplitude values within the memory line belonging to the receiver, and a unit for mathematical combination of the amplitude values taken from the memory lines.

In analog processing, a known triangulation or ellipsoidal back projection can also be used to determine the spatial coordinates. If only one receiver is used, the coordinates of the reflection points are immediately known.

The use of an adaptive grid reconstruction processor is therefore only advantageous if at least two receivers are available.

The adaptive grid reconstruction processor is also the subject of the present invention. The adaptive grid reconstruction processor consists of a raster generator, a unit for determining the paths of the signals from the transmitter to the raster points and from the raster points to the individual receivers for each raster point, a memory line for the sum of all belonging to the receiver Amplitude values along the channels of the spectral decomposition, a sampling address generator for determining the address of the amplitude values within the memory line belonging to the respective receiver, a unit for mathematically linking the amplitude values taken from the memory lines.

To generate an illustration of an invisible object, a signal that is modulated with any known modulation function that guarantees a corresponding bandwidth of the signal (preferably FM and / or binary-phase-coded, but no pulse (AM modulation)) is generated in Direction sent to the object and at the same time saved as a reference signal. At least one receiver receives the signals reflected on the object simultaneously. The Reflected signals are then optionally mixed with the transmission signal stored as a reference signal and the beat signal obtained is spectrally broken down in the n-channel. If there is no mixer, the response signal is directly spectrally broken down into the n-channel.

For example, an acousto-optical processor can be used for real-time spectral decomposition. However, if a digitized signal is present, it is also possible to use a Fast Fourier processor or another processor, e.g. to use a digital wavelet processor, with the aid of which a rapid spectral decomposition of the signals can be carried out.

The results of the spectral decomposition are first stored in a parallel memory shift register for each reflection point. At the same time, the sum of the amplitude values of each shift register can be formed and stored in a memory line of the adaptive grid reconstruction processor, if this is used.

As already mentioned, known methods such as the ellipsoidal back projection and triangulation methods or the like can be used to determine the coordinates of the reflection points. The basis for this is that the points of the same distance from the transmitter to the reflection points and to the receiver lie on an ellipsoid. A faster solution that offers significant advantages for display units with more than one receiver and digital signal processing and only makes them functional in real time offers one

Adaptive Grid Reconstruction processor. This processor is also part of the present invention.

As has already been indicated, it is possible to measure the signal paths d by measuring the beat frequencies Δ ^. with the help of Formula d. ~ - ' ^■ ——— (where c is the speed of sound, T _{M is} the M

Determine the duration of the transmission signal and M is the deviation of the transmission signal). The time course of the beat frequencies has been stored in the parallel memory shift register after the spectral decomposition of the beat signal. This means that for every Δ (i.e. every ellipsoid) a spectrum is obtained which contains the spectral components of all reflection points on this ellipsoid. The number n of channels in the spectral decomposition gives an upper limit for the resolution of the image according to the above formula. From the

Intersection of these ellipsoids results in a coordinate system that is not metric. In order to make these intersection points visible in a metric space, special additional optics (e.g. special correction lens) can be used in front of the screen or the HDM (Head Mounted Device). The space given by the points of intersection of the ellipsoids can also be clearly converted into a metric space that is favorable for visualization by means of a transformation. However, there is no uniform distribution of the points to be displayed in space.

When using the adaptive grid reconstruction processor in the digital process, a Euclidean grid is first placed over the sound volume, ie a grid with advantageously, but not necessarily, uniform distances (favorable in sampling distance) between the points in the horizontal and vertical directions. From the x, y, z coordinates of the grid points, the paths from the transmitter to the grid point (R0) and the paths from the grid point to the respective receiver (Rl, R2, R3 with 3 receivers) can now be derived using the Pythagorean theorem , The sums of these calculated paths R0 + R1, R0 + R2 and R0 + R3 do not necessarily correspond to a sampling point in the response signal. In addition, the sampling points of the response signals form each channel a separate space that cannot be mapped linearly to the space of the sampling points of another channel (see FIG. 6). This means that the receivers see different parts of the sound structure (similar to the RGB mask on color TV). This effect can also be described as a mask effect, with each receiver seeing the sonicated structure through a mask, which is clearly defined by the sampling frequency and the position of the receiver. As a result, the masks of the individual receivers cannot be covered by a linear image.

A £ environment (δ> grid spacing) is therefore defined for each grid point. The amplitude values that were stored in the memory register of the adaptive grid reconstruction processor are averaged for all sampling points that lie within this δ environment. If δ is large

Is Ϊ3, you get smoothing. If δ is less than, / - times

V2

the grid spacing in the 3D grid (times the grid spacing in

2D grid), information is lost. This results in a simple way for the "zoom-out", with which storage space and computing time can be saved. For this purpose, the grid can be chosen coarser from the outset. Then only the paths R0 + R1, R0 + R2 and R0 + R3 is calculated. Depending on the grid size and the determination of δ, a correspondingly higher number of sampling points is then averaged. Analogously, one can proceed if a higher level of detail is to be achieved. Assuming that the sampling frequency is lower than the Nyquist frequency of the received signal. Then According to the sampling theorem, aliasing effects occur. However, if the filtering is skillful and the grid spacing is selected, this should not be regarded as a disruptive factor. Then the grid size is limited by this sampling spacing. If the sampling frequency is higher, eg by Oversampling, then an even greater level of detail can be achieved here, since the corresponding smaller grid section (voxel) still contains sampling points that contain information.

As here, as in the analog case, the spectra of all points lying on an ellipsoid are superimposed, backprojections must be carried out for each channel of the spectrum, as in computer tomography, which provide the spectrum of each individual point to be displayed ,

The choice of the grid spacing can also be done locally differently, thus also the determination of δ, which should be adapted to the grid spacing. This enables local zooming (such as bird's eye). You can e.g. show the near area in more detail than the far area or show the far area in detail. The actual detailed information is always contained in the data, only the display becomes particularly user-friendly, and computing time is saved.

In order to be able to classify each spatial point in 1D, 2D or 3D with a given spectrum, the spectra are read out in parallel from the memory shift register and a statistical classification method, e.g. nearest neighbors, neural network, support vector machine and the same applied to it.

The selection of the statistical classification method essentially depends on the ability to generalize and the speed of the computing technology. The ability to generalize means the ability to correctly classify a point of reflection whose class membership is not known beforehand. The method can be expanded in such a way that exactly the frequency ranges in the signal are selected with the aid of which a "best" classification is possible for the given structure. A transmit signal can then be transmitted again, but this time with a non-constant modulation function, but with a modulation function that mainly contains the frequency ranges that are required for an optimal classification ("color assignment") of the reflection points. These "optimal" frequency ranges can be selected, for example, with the aid of the analysis unit which is connected to the classifier. It is precisely these ranges which are then generated and transmitted by the generator in the next run.

With this device, therefore, exactly the frequency range, i.e. the Spectrogra m, are determined in which a maximum

Reflection of the transmission signals is caused. An acoustic impedance, by which the surface of an object is characterized, can thus be set “sharply” or its “acoustic color” can be determined more precisely. Transitions of the same "acoustic color" have the same properties and are determined as the same objects. All reflection points with the same properties can be represented in a closed manner. Since the resulting spectrum is a property ("color") of the surface of an object on which the reflection the experience of a viewer now plays a subordinate role.

With this method it is therefore possible to perceive objects which are hidden from the human eye by smoke, fog, darkness, a temperature curtain, by cloudy water or the like. With this device, all information contained in the reflections of the arbitrarily modulated ultrasound about the object surface is used. If several display devices are used at the same time, different modulations can be used for each device of the transmission signals are made to distinguish the signals.

The invention will be explained in more detail below using an exemplary embodiment. In the drawings, the same reference numerals mean the same or similar parts.

1A and 1B show block diagrams of an active ultrasonic viewing device according to the present invention for 1D and 3D space, respectively;

Fig. 2 shows a graphical representation to explain the

Frequency shift between the transmission signal and the

Response signal;

3 shows a graphical representation of the spectrally decomposed reflection signals;

Fig. 4 shows a memory shift register as used in the invention;

5 shows the basic structure of an adaptive grid reconstruction processor; Fig. 6 is an illustration of the different spaces formed by the sampling points of different receivers; and

Figure 7 is an illustration of the grid formed by the adaptive grid reconstruction processor.

Figure 1A shows a block diagram of an active ultrasound display device in accordance with the present invention for one-dimensional use. In this case, the active ultrasonic viewing device consists of a transmitter 1, which has a modulator 4, a generator 5 and a device for

Signal conditioning 3 is coupled, and a receiver 2 with a device for signal conditioning 3. The active ultrasonic viewing device also consists of a time control 10, the signal processing from the start of the transmission signal to the display of a Timed ultrasound image. On the receiver side, the device for signal conditioning 3 is followed by a mixer 6 which mixes the response signals with the transmission signal in order to obtain the beat frequencies which indicate the reflections. A mixer for spectral decomposition 9 follows the mixer 6

Beat frequencies, e.g. an acousto-optical processor. In this case, the unit for spectral decomposition 9 can also be carried out by a holographic wavelet processor or another unit for fast analog spectral decomposition. For each reflection point, a signal is generated which contains the information about the path of the signal from the transmitter via the reflection point to the receiver and the spectrum of the reflection point and which is stored in a memory shift register 12. The memory shift register 12 is followed by two circuits which select or evaluate the information contained in the signals, a circuit for determining the coordinate 15 of the reflection point and a circuit for classifying 14 the reflection point, which are shown on a display 16.

1B shows a block diagram of an active ultrasound viewing device according to the present invention for 3D space and with digital signal processing. A three-dimensional representation of ultrasound images can thus take place here. In this case, the active ultrasound viewing device consists of a transmitter 1 and three receivers 2. The active ultrasound viewing device also consists of a time control 10, which controls the signal processing starting with the start of the transmission signal until the display of an ultrasound image. One per

Device for signal conditioning 3 is arranged after each receiver 2. The signal conditioning 3 balances the non-uniformities of the reflection signals that are received by the receivers 2. The device for signal conditioning 3 is followed by a mixer 6, which is the Reflection signals each mixed with the transmission signal to obtain the beat signals that indicate the reflections. The mixer 6 is followed in each case by an A / D converter 7 and a unit for spectral decomposition 9, which spectrally decompose the beat signals. The units for spectral decomposition 9 can in this case be carried out by a digital wavelet processor, a Fast Fourier Transform (FFT) or another unit for fast digital spectral decomposition.

Each channel of the spectral decomposition contains the information about the path of the signal from the transmitter 1 via all reflection points on the ellipsoid, which is determined by this channel, to the receiver 2 and the superimposed spectrum of these reflection points. These signals are stored in a shift register 12. The sums of the cells of the shift register 12 are stored in a memory register 13, which is part of the adaptive grid reconstruction processor. The memory shift register 12, 13 is followed by two circuits which select and evaluate the information contained in the signals, a circuit for determining the coordinate 15 of the reflection points and a circuit for classifying 14 the reflection points, which are shown on a display 16.

2 and 3 serve to explain the detection of the reflection signals. 2 shows the time on the x-axis and the frequency on the y-axis. The transmission signal has the frequency f _s , the reflection signal has the frequency f _E and the frequency shift is Δf. f _s has a duration of T _M. M = f _max - f _min is the deviation of the

Signal. The reflection signal with the frequencies f _E comes back after a time Δt. The following applies: .delta.t

M

d c =

.delta.t

d ^Δ ^• τ _M - ^c

M

If a constant speed of sound, as would occur in a homogeneous medium, is assumed, then Δf is constant over time. If Δf can be measured, the path d from the transmitter via the reflection point to the receiver can be calculated using the known variables T _M , M, c and Δf

T _M dt. In general, the above formula can be represented by - ^J - = -

M df to be replaced.

If the beat signal is now spectrally broken down, one obtains FIG. 3. These are the signals that are in each receiver channel

AWeg can be processed separately. The straight line Δt = - results

separated in that the response signal arrives from different depths of the structure at different times at the receiver.

However, before performing an FFT or other fast spectral decomposition of the beat signal, the number n of channels must be determined. With this number «(e.g. n = 1024) there is an upper limit for the resolution of the image. For every beat frequency Af. , / = (!, ..., «), you now get a series of amplitudes over the length of the signal or from the time of transmission until the reception of the last signal from the most distant reflector. In channel Af, you get a superposition of parts of this beat frequency Af. in all response signals with a

Running time Δt _; , , ie of all reflectors that lie on an imaginary ellipsoid with the transmitter and receiver as focal points. With a 1024-channel spectral decomposition (eg FFT), all 1024 channels are stored in parallel in the memory, ie 1024 values are read in with each time step that is controlled by the time control. Since the reflection signals come back from deeper areas later, the shifting Δt increasing from left to right is obtained. Signals are transferred from the memory shift register 12, 13 to the reconstruction unit 15, each of which contains the path covered for each individual response signal and the superimposed spectrum of all of the reflection points which lie on the surface of the ellipsoid corresponding to this path.

After the spectral decomposition 9, the memory register of the adaptive grid reconstruction processor contains e.g. n = 1024 (n = number of channels in spectral decomposition 9) values which in turn result from the sum of these over time, i.e. n = 1024 amplitude values (attributes).

For the determination of the exact 3D positions, i.e. the spatial coordinates of the reflection points can be known

Reconstruction algorithms such as ellipsoid back projection or triangulation can be used. However, using an adaptive grid reconstruction processor offers one decisive time advantage over the other methods mentioned.

5 shows the adaptive grid reconstruction processor. The adaptive grid reconstruction processor consists of a raster generator 19, a unit 18 for determining the paths of the signals from the transmitter 1 to the raster points and units 17 for determining the paths of the signals from the raster points to the individual receivers 2 for each raster point , the already mentioned memory register 13 for the sum of the amplitude values of the reflection points belonging to the receiver 2, a sampling address generator 21 for determining the address of the memory register 13 belonging to the respective receiver 2 and a unit for mathematical linking 22 of the memory registers 13 withdrawn values, and a unit 23 for transferring the δ environment and the sampling frequency to the sampling address generator 21.

The adaptive grid reconstruction processor first creates a virtual Euclidean grid over the

PA volume — laid out, whereby a grid with uniform distances (optimally in the sampling distance) between the points in the horizontal and vertical direction is advantageous. For this purpose, transmitter 1 and receiver 2 are coupled to a raster generator 19. 7 shows such a raster for the sake of simplicity in 2D.

From the x, y, z coordinates of the grid points, the paths from transmitter 1 to grid point (R0) and the paths from grid point to the respective receiver 2 (Rl, R2, R3 with 3 receivers) can be derived using the Pythagorean theorem , The sums of these calculated paths R0 + R1, R0 + R2 and R0 + R3 do not necessarily correspond to a sampling point in the response signal. As already described, the Sampling points of the response signals of each channel have their own space, which cannot be mapped linearly to the space of the sampling points of another channel. FIG. 6 serves to illustrate this effect. Therefore, a unit for determining a δ environment is provided for each raster point 23. For all sampling points that lie within this δ environment, the amplitude values are averaged with the aid of the mathematical link 22. If δ is large, smoothing is obtained. If δ is smaller

[3 1 as - times the grid spacing in the 3D grid (- == • times the V2

Grid spacing in the 2D grid, information is lost.

This results in a simple way for the "zoom-out", in which storage space can be saved. For this, the grid can be chosen coarser from the outset. Then only the paths RO + Rl, R0 + R2 and R0 + belonging to these grid points R3 is calculated and then averaged over a correspondingly higher number of sampling points, depending on the grid size.

One can proceed analogously if a higher level of detail is to be achieved. Assume that the sampling frequency is lower than the Nyquist frequency of the received signal. Then aliasing effects occur according to the sampling theorem. With clever filtering and the choice of the grid spacing, however, this should not be regarded as a disruptive factor. Then the grid size is limited downwards by this sampling distance. If the sampling frequency is higher, e.g. by oversa pling, an even greater level of detail can be achieved here, since the correspondingly smaller grid section (voxel) still contains sampling points that contain information.

Since the spectra of all points lying on an ellipsoid are superimposed here, just as in the analog case, here too, as with computer tomography, for everyone Channel of the spectrum back projections are carried out, which provide the spectrum of each individual point to be displayed.

The classification unit 14 is coupled to the reconstruction circuit 15 and evaluates the spectra of the individual reflection points, which are stored in the memory shift register 12, 13 according to FIG. 4. Depending on the properties of the reflection points, changes occur in the reflection signals. If these changes and thus the character of the reflection points are already known, the properties for the visualization can be marked accordingly. The coordinates of the individual reflection points and their properties are then brought together on a display 16.

There are two main options for visualization: first, stereoscopic visualization (e.g. for people or robots) or direct three-dimensional representation (e.g. for cartography), which is two-dimensional.

In order to reduce the energy transmitted, a transmission signal can be transmitted again, but this time it is modulated with a non-constant modulation function, which mainly contains only the frequency ranges that are required for an optimal classification ("color assignment") of the reflection points.

List of the reference symbols used

1 transmitter

2 receivers

3 Signal conditioning

4 modulator

5 generator

6 mixers

7 A / D converter 8 9 spectral decomposition

10 time control

11

12 shift registers

13 memory registers

14 Classification circuit

15 Circuit for determining the coordinates

16 display

Claims

Active ultrasonic viewing device

1. Active ultrasound viewing device with at least one transmitter for transmitting a transmission signal with any known modulation function and a medium-dependent bandwidth of the signal, at least one receiver arranged in a known position thereto, each of which has a first signal processing unit for n-channel spectral decomposition of the Received response signals received from the medium and a memory shift register is arranged downstream, a second signal processing unit for determining the spatial coordinates of the reflection points and a display.

2. Active ultrasonic viewing device according to claim 1, wherein the receivers are arranged symmetrically to the transmitter.

3. Active ultrasonic viewing device according to claim 1 or 2, in which the receivers are each followed by a mixer for generating the beat frequency between the transmission signal and the response signals.

4. Active ultrasonic viewing device according to one of claims 1 to 3, in each case in front of the first

A / D converter is provided for signal processing units for n-channel spectral decomposition.

5. Active ultrasonic viewing device according to one of claims 1 to 4, in which the memory shift register is followed by a classifier.

6. Active ultrasound viewing device according to claim 5, in which the classifier is preceded by a preprocessing unit.

7. Active ultrasonic viewing device according to claim 5 or 6, wherein the classifier is followed by an analysis unit.

8. Active ultrasound viewing device according to one of claims 5 to 7, wherein the classifier is a statistical

Classifier, such as a neural network, a support vector machine or the like.

9. Active ultrasonic viewing device according to one of claims 1 to 3, in which the first signal processing unit for n-channel spectral decomposition is an acousto-optical

Processor, a holographic wavelet processor or a similar analog processor for spectral decomposition.

10. Active ultrasound viewing device according to claim 4, wherein the first signal processing unit for spectral

Decomposes a Fast Fourier processor, a digital wavelet processor or a similar digital processor for spectral decomposition.

11. An active ultrasonic viewing device according to claim 4, wherein the second signal processing unit for determining the

Space coordinates consists of an adaptive grid reconstruction processor.

12. Adaptive grid reconstruction processor consisting of a raster generator, a unit for determining the paths of the signals from the transmitter to the raster points and from the halftone dots to the individual receivers for each halftone dot, a memory line for the sums of all amplitude values belonging to the receiver via the channels of the n-channel spectral decomposition, a sampling address generator for determining the address of the amplitude values within the memory line belonging to the receiver, and a unit for mathematically linking the amplitude values taken from the memory lines.