An Acoustic Target Location and Recognition Method Based on Band Difference Degree

Abstract

The present invention discloses an acoustic target localization and recognition method based on band difference degree, which includes: in a distributed sensor scenario, all sensors synchronously sample a target to obtain N target signals; combine the N target signals, each time select two sensors to form a sensor pair, and form multiple sensor signal pairs; synchronously divide the N target signals according to a pre-determined frequency band, and respectively calculate the cross-correlation results of the target signals collected by each pair of sensors in each frequency band; based on the cross-correlation results, respectively calculate the steering response power distribution of each frequency band; based on the steering response power distributions of all frequency bands, obtain the localization and recognition results of the sound source. The present invention improves the efficiency of target resolution, localization and recognition.

Description

Technical Field

[0001] The present invention relates to the technical field of sensor signal processing, and particularly to an acoustic target positioning and recognition method based on frequency band difference degree. Background Art

[0002] Target positioning is a common application of acoustic sensor systems. Traditional distributed acoustic sensor positioning mechanisms include positioning based on signal strength / arrival time (RSSI / ToA), angle of arrival positioning (DoA), positioning based on time difference of arrival (TDoA), matching positioning, steering response power (SRP), etc. Under these mechanisms, various existing algorithms have been derived, as well as improved and corrected algorithms for various environments, which are applied to various positioning fields.

[0003] In actual use scenarios, the presence of multiple targets and the influence of interference signals / sound sources have a negative impact on the performance of positioning algorithms. Simply using preprocessing means to convert the signal into intermediate quantities such as intensity, time (difference), angle, etc. results in the loss of target characteristic information, and additional processing is required for signal recognition, discrimination, etc. To achieve synchronous implementation of positioning and recognition, target characteristic information needs to be introduced from the preprocessing stage. Summary of the Invention

[0004] In view of this, the present invention provides an acoustic target positioning and recognition method based on frequency band difference degree to solve the above technical problems.

[0005] The present invention discloses an acoustic target positioning and recognition method based on frequency band difference degree, which includes the following steps:

[0006] Step 1: In a distributed sensor scenario, all sensors synchronously sample the target to obtain N target signals; where N is the number of sensors; the target is a sound source;

[0007] Step 2: Combine the N target signals, and each time select two sensors to form a sensor pair to form multiple sensor signal pairs;

[0008] Step 3: Synchronously divide the N target signals according to the pre-determined frequency bands, and calculate the cross-correlation results of the target signals collected by each pair of sensors in each frequency band;

[0009] Step 4: Based on the cross-correlation results, calculate the steering response power distribution of each frequency band respectively;

[0010] Step 5: Based on the steering response power distributions of all frequency bands, obtain the positioning and recognition results of the sound source.

[0011] Further, the Step 1 includes:

[0012] In the distributed sensor scenario, all sensors simultaneously collect signals from sound sources within the coverage of all sensors at the same sampling frequency, obtaining N target signals.

[0013] Further, the step 2 includes:

[0014] Simultaneously divide the N target signals into frequency bands according to the same frequency band range, and at the same time pair up the sensors in the distributed sensor scenario in pairs, with a maximum of sensor pairs; in actual use, only part of the sensor pair data can be used, and the number is between 2 and inclusive.

[0015] Further, the step 3 includes:

[0016] Step 31: Calculate the cross-correlation result of the target signals collected by the first pair of sensors in the first frequency band through the following formula:

[0017]

[0018] where f is the frequency, F 1 is the first frequency band, Pr 1 is the first pair of sensors, Ψ(f) is the weighting function of the generalized cross-correlation function, S 1 (f) is the Fourier transform of the signal of the first sensor in the first pair of sensors, is the conjugate of the Fourier transform of the signal of the second sensor in the first pair of sensors, and τ is the time delay;

[0019] Step 32: By analogy with step 31, calculate the cross-correlation results of the target signals collected by each pair of sensors in all frequency bands respectively.

[0020] Further, the step 4 includes:

[0021] Step 41: Based on the cross-correlation results of the target signals collected by all sensors in the first frequency band F 1 where M is the number of selected sensor pairs; calculate the steering response power distribution in the first frequency band through the following formula:

[0022]

[0023] where, is the steering response power distribution corresponding to the first frequency band, p x,a , p x,b are the positions of the two sensors currently selected in Pr x ;

[0024] Step 42: By analogy with Step 41, obtain the steering response power distributions corresponding to all frequency bands respectively. After combination, a multi - frequency band distribution G(p) is formed, denoted as where K is the total number of frequency bands obtained by dividing the frequency bands of N target signals.

[0025] Further, the said Step 5 includes:

[0026] Step 51: Based on the steering response powers of all frequency bands at each position, obtain the power variance at this position, and then obtain the variance distribution over all positions. After finding the peak of the variance distribution, obtain the position estimate of the sound source.

[0027] Step 52: Based on the similarity between the steering response power of the reference target and the steering response powers of all frequency bands at other positions, obtain the category estimate of the sound source; where the category of the reference target is known.

[0028] Step 53: Fuse the category of the sound source and the position estimate of the sound source, that is, obtain the positioning and recognition result of the sound source.

[0029] Further, the said Step 51 includes:

[0030] Based on G(p), calculate the variance distribution through the following formula:

[0031]

[0032] where σ 2 (p) is the variance distribution;

[0033] After obtaining σ 2 (p) and identifying the peak, the estimate of the target position is obtained.

[0034] Further, the said Step 52 includes:

[0035] Analyze the similarity between each position in the search space and the feature vector; where the similarity is described by cosine similarity, that is:

[0036] M(,p ref )=G T (p)G( ref ) / ‖G(p)‖‖G(p ref )‖

[0037] where M(p,p ref ) is the similarity between the position p in the search space and the reference position p ref , G(p ref ) is the steering response power distribution corresponding to the reference position p ref ;

[0038] M(,pref ) The greater the similarity, the higher the similarity between the target at the corresponding search space position p and the reference target.

[0039] Further, the step 53 includes:

[0040] Using σ 2 (p) and the Hadamard product of M(p, p ref ) as the fusion result, that is The fusion result is the positioning and recognition result of the sound source.

[0041] Due to the adoption of the above technical solutions, the present invention has the following advantages: This method adopts a positioning method based on the band difference degree, and can better locate the target in a stationary noise scene. The present invention proposes a new processing method and mode, which improves the target resolution, positioning, and recognition capabilities of the system. The present invention can improve the target classification ability in a multi-target scene. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments recorded in the embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0043] Figure 1 It is a schematic flowchart of an acoustic target positioning and recognition method based on the band difference degree according to an embodiment of the present invention;

[0044] Figure 2 It is a schematic diagram of a distributed sensor scene containing two sound sources according to an embodiment of the present invention;

[0045] Figure 3 It is a schematic diagram of the comparison of signals of each channel in a two-target scene according to an embodiment of the present invention;

[0046] FIG. 4(a) and FIG. 4(b) are respectively schematic diagrams of the SRP distribution result and the band variance distribution result in a two-target scene according to an embodiment of the present invention;

[0047] Figure 5 It is a schematic diagram of the band difference degree distribution when the peak value (3, 6) is used as a reference according to an embodiment of the present invention;

[0048] FIG. 6(a) and FIG. 6(b) are the top view and the perspective view of the fusion result of the difference degree and the band variance in a two-target scene according to an embodiment of the present invention;

[0049] FIG. 7(a) and FIG. 7(b) are respectively schematic diagrams of the comparison between the traditional SRP distribution result and the fusion result in a three-target scene according to an embodiment of the present invention;

[0050] Figures 8(a), 8(b), and 8(c) are distribution diagrams of three groups of difference degrees (three hues) in the target scenario of the third embodiment of the present invention. Detailed implementation manners

[0051] The present invention will be further described in conjunction with the accompanying drawings and embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art shall fall within the protection scope of the embodiments of the present invention.

[0052] See Figure 1 , the present invention provides an embodiment of an acoustic target positioning and recognition method based on band difference degree, which includes the following steps:

[0053] Step 1: In a distributed sensor scenario, all sensors synchronously sample the target to obtain N target signals; where N is the number of sensors; the target is a sound source.

[0054] Step 2: Combine the N target signals. Each time, two sensors are selected to form a sensor pair, and multiple sensor signal pairs are formed.

[0055] Step 3: Synchronously divide the N target signals according to the pre-determined frequency bands, and calculate the cross-correlation results of the target signals collected by each pair of sensors in each frequency band.

[0056] Step 4: Based on the cross-correlation results, calculate the steering response power distribution of each frequency band respectively.

[0057] Step 5: Based on the steering response power distributions of all frequency bands, obtain the positioning and recognition results of the sound source.

[0058] In this embodiment, Step 1 includes:

[0059] All sensors in the distributed sensor scenario simultaneously collect signals from the sound source within the coverage range of all sensors at the same sampling frequency to obtain N target signals.

[0060] In this embodiment, Step 2 includes:

[0061] Simultaneously divide the N target signals according to the same frequency band range, and at the same time pair up the sensors in the distributed sensor scenario two by two. There are at most sensor pairs; in specific use, only part of the sensor pair data can be used, and the number is between 2 and .

[0062] In this embodiment, Step 3 includes:

[0063] Step 31: Calculate the cross-correlation result of the target signals collected by the first pair of sensors in the first frequency band through the following formula:

[0064]

[0065] where f is the frequency, F 1 is the first frequency band, Pr 1 is the first pair of sensors, Ψ(f) is the weighting function of the generalized cross-correlation function, S 1 (f) is the Fourier transform of the signal of the first sensor in the first pair of sensors, is the conjugate of the Fourier transform of the signal of the second sensor in the first pair of sensors, and τ is the time delay;

[0066] Step 32: By analogy with Step 31, calculate the cross-correlation results of the target signals collected by each pair of sensors in all frequency bands respectively.

[0067] In this embodiment, Step 4 includes:

[0068] Step 41: Based on the cross-correlation results of the target signals collected by all sensors in the first frequency band F 1 where M is the number of pairs of sensors selected; calculate the steering response power distribution in the first frequency band through the following formula:

[0069]

[0070] where, is the steering response power distribution corresponding to the first frequency band, p x,a , p x,b are the positions of the two sensors currently selected in Pr x x x

[0071] Step 42: By analogy with Step 41, obtain the steering response power distributions corresponding to all frequency bands respectively After combination, form the multi-frequency band distribution G(p), denoted as where K is the total number of frequency bands for dividing the N target signals into frequency bands.

[0072] In this embodiment, Step 5 includes:

[0073] Step 51: Based on the steering response power of all frequency bands at each position, obtain the power variance at that position, and then obtain the variance distribution at all positions. After finding the peak of the variance distribution, obtain the position estimate of the sound source;

[0074] Step 52: Based on the similarity between the steering response power of the reference target and the steering response power of all frequency bands at other positions, obtain the category estimate of the sound source; where the category of the reference target is known;

[0075] Step 53: Integrate the category of the sound source and the estimated position of the sound source, that is, obtain the positioning and recognition result of the sound source.

[0076] In this embodiment, step 51 includes:

[0077] Based on G(p), calculate the variance distribution through the following formula:

[0078]

[0079] where σ 2 (p) is the variance distribution;

[0080] Obtain σ 2 (p) and identify the peak, that is, obtain the estimation of the target position.

[0081] In this embodiment, step 52 includes:

[0082] Analyze the similarity between each position in the search space and the feature vector; among them, the similarity is described by cosine similarity, that is:

[0083] M(,p ref ) = G T (p)G( ref ) / ‖G(p)‖‖G(p ref )‖

[0084] where M(p,p ref ) is the similarity between the search space position p and the reference position p ref , G(p ref ) is the steering response power distribution corresponding to the reference position p ref ;

[0085] M(,p ref ) The greater the similarity, the higher the similarity between the target at the corresponding search space position p and the reference target.

[0086] In this embodiment, step 53 includes:

[0087] Use the Hadamard product of σ 2 (p) and M(p,p ref ) as the fusion result, that is The fusion result is the positioning and recognition result of the sound source.

[0088] For the convenience of understanding, the present invention gives a more specific embodiment:

[0089] Here, the effect of this method is demonstrated by combining simulation examples. As Figure 2As shown in the figure, a 20m * 20m experimental site is preset, and acoustic sensors are deployed at the four corners. Two sound sources are respectively deployed at positions (3, 6) and (12, 15). Each will emit a short - time signal at a random time. Each signal consists of three frequency components randomly selected from 1 - 5kHz, and white noise is superimposed. The powers of the three frequency components are also random, but the total noise - free power of the two signals is equal, and the signal - to - noise ratio after signal superposition with noise is - 5dB. Under such settings, all four acoustic sensors receive the signals of the two sound sources.

[0090] Step 1: Collect a 1024 - point (102.4ms duration) signal segment at a sampling rate of 10kHz respectively, and the signal is as Figure 3 shown. Although the specific signal frequency is unknown, since the sampling rate is 10kHz, signals below 5kHz can be processed. Divide [1Hz, 5kHz] evenly into 10 frequency bands, F 1 : 1 - 500Hz, F2: 501 - 1000Hz, …, until F 10 : 4501 - 5000Hz.

[0091] Step 2: Divide the four sensors into 4 pairs, Pr 1 : [1, 2], Pr 2 : [2, 3], Pr 3 : [3, 4], Pr 4 : [4, 1]. Calculate the cross - correlation results of each pair of signals. The number of sensor pairs should be large enough to ensure system performance. Here, more sensor pairs can also be introduced to improve system redundancy, such as [1, 3], [2, 4], but choosing 4 pairs of sensors already covers all nodes. The specific number of sensor pairs used can be adjusted between 2 and according to the actual situation during use. Here, basic cross - correlation processing is adopted, that is, Ψ(f) = 1. At this time, the cross - correlation result of the signals of the first pair of sensors Pr 1 in the frequency band F 1 can be obtained:

[0092]

[0093] And so on, the cross - correlation results of Pr 1 in all frequency bands can be obtained and similarly extended to Pr 2 , Pr 3 and Pr 4 .

[0094] Step 3: Generate the steering response power distribution for each frequency band. After having the cross - correlation results of each pair of signals in each frequency band, the formula mentioned above can be used to obtain the distribution of frequency band k, where px,a , p x,b the positions of two sensors in the currently selected Pr x .

[0095] Since the cross - correlation result is discretely represented in actual processing, all of them may not be calculated first in Step 2 but only the discrete Fourier transform result of the signal is prepared. When the result corresponding to the time delay τ is used in Step 3, the calculation is carried out and stored for later use. For the case where there is no exact value in the discrete scenario, interpolation methods such as nearest neighbor or linear interpolation can be used to obtain the fitting result.

[0096] After obtaining the distribution of all frequency bands, Next, the following can be calculated:

[0097]

[0098] The target resolution effect of the above - mentioned variance distribution can be seen in Fig. 4(a) and Fig. 4(b), and compared with the ordinary cross - correlation - based steering response power result, that is, ∑G(p), the performance advantages in peak recognition and non - target area suppression can be seen. After obtaining σ 2 (p) and identifying the peak, the estimation of the target position is basically obtained.

[0099] Step 4: Analyze the target difference characteristics for target recognition. Here, the peak position of σ 2 (p), that is, (3, 6), is defined as the reference point position p ref , and the corresponding G(p ref ) is the reference feature vector. In terms of meaning, such a division is equivalent to analyzing the similarity between each position in the search space and the feature vector without considering the intensity. Here, the similarity is described by cosine similarity, that is:

[0100] M(p, p ref ) = G T (p)G(p ref ) / ‖G(p)‖‖G(p ref )‖

[0101] The similarity distribution of the example in the experiment is as Figure 5 shown. The normalized result is shown in the figure, and the dark color represents high color similarity. It can be seen that there is the highest similarity near the reference point (3, 6), and the similarity in other regions is relatively low. Other peaks appearing in other regions (low - similarity regions) should be determined as different targets.

[0102] Step 5: Integrate the results of positioning and recognition. Using σ 2 (p) and M(p, p ref) is taken as the fusion result, namely To show the fused result, the fused result is used as the control variable of the green channel. The display result is shown in Figure 6(a). It can be seen that two dark blocks appear in the search space, one is the (3,6) area and the other is the (12,15) area. The colors of the two areas are different, indicating the difference in their signal characteristics. Figure 6(b) shows the difference between σ 2 The result of combining the (p) distribution with the color block distribution shows the clear positions and color differences of the two peaks, achieving preliminary positioning and identification.

[0103] The method can also handle scenarios with more targets. A target at position (17,9) is added to the above scenario for a three-target scenario test. To distinguish multiple targets, the frequency band is divided into F 1 ~F 4 ,F 5 ~F 7 , and F 8 ~F 10 Three groups, each group calculates M(p,p) on the frequency band within each group. ref ), and finally get three groups: H 1 ,H 2 ,H 3 Using it as the corresponding parameter of the RGB channel, the fusion positioning result of the target can be obtained. After combining it with the σ^2(p) distribution, the result is shown in Figure 7(b). Compared with the ordinary guide response power distribution result in Figure 7(a), in addition to the hint of target distinction from the peak color, the overall peak shape is also smooth compared with the noise signal without target. Figures 8(a) to 8(c) In the figure, we can see the distribution of different difference results of the three channels. It can be seen that the similarity characteristics of the targets are different in different frequency bands. By adopting this patented method, the target classification capability in multi-target scenarios can be improved.

[0104] It should be noted that the three-target scenario has a higher signal-to-noise ratio than the two-target scenario (0 dB here, -5 dB for the two-target scenario), but false targets still appear. This problem comes from insufficient information rather than processing errors. Only four sensors are used in the simulation, divided into four pairs, and the amount of information is relatively small. In practice, this needs to be solved by increasing the number of nodes.

[0105] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. An acoustic target localization and recognition method based on band difference degree, characterized in that, it includes the following steps: Step 1: In a distributed sensor scenario, all sensors synchronously sample the target to obtain N target signals; where N is the number of sensors; the target is a sound source; Step 2: Combine the N target signals. Each time, select two sensors to form a sensor pair, and form multiple sensor signal pairs; Step 3: Synchronously divide the N target signals according to the pre-determined frequency bands, and calculate the cross-correlation results of the target signals collected by each pair of sensors in each frequency band respectively; Step 4: Based on the cross-correlation results, calculate the steering response power distribution of each frequency band respectively; Step 5: Based on the steering response power distributions of all frequency bands, obtain the localization and recognition result of the sound source; The said Step 3 includes: Step 31: Calculate the cross-correlation result of the target signals collected by the first pair of sensors in the first frequency band through the following formula: where f is the frequency, F 1 is the first frequency band, Pr 1 is the first pair of sensors, Ψ(f) is the weighting function of the generalized cross-correlation function, S 1 (f) is the Fourier transform of the signal of the first sensor in the first pair of sensors, is the conjugate of the Fourier transform of the signal of the second sensor in the first pair of sensors, and τ is the time delay; Step 32: By analogy with Step 31, calculate the cross-correlation results of the target signals collected by each pair of sensors in all frequency bands respectively; The said Step 4 includes: Step 41: Based on the first frequency band F 1 The cross-correlation results of the target signals collected by all sensors on where M is the number of pairs of sensors selected; through the following formula, calculate the steering response power distribution on the first frequency band: Among them, is the steering response power distribution corresponding to the first frequency band, p x,a , p x,b are the positions of two sensors in the currently selected Pr x ; Step 42: By analogy with Step 41, the steering response power distributions corresponding to all frequency bands are obtained respectively. After combination, a multi-band distribution G(p) is formed, denoted as where K is the total number of frequency bands obtained by dividing the frequency bands of N target signals; The said Step 5 includes: Step 51: Based on the steering response power of all frequency bands at each position, obtain the power variance at this position, and then obtain the variance distribution at all positions. After finding the peak of the variance distribution, obtain the position estimate of the sound source; Step 52: Based on the similarity between the steering response power of the reference target and the steering response power of all frequency bands at other positions, obtain the category estimate of the sound source; where the category of the reference target is known; Step 53: Fuse the category of the sound source and the position estimate of the sound source, that is, obtain the localization and recognition result of the sound source; The said Step 51 includes: Based on G(p), calculate the variance distribution through the following formula: Among them, σ 2 (p) is the variance distribution; Obtain σ 2 (p) and a peak is recognized, i.e., an estimation of the target position is obtained; The said Step 52 includes: Analyze the similarity between each position in the search space and the feature vector; where the similarity is described by cosine similarity, that is: M(p,p ref ) = G T (p)G(p ref ) / ‖G(p)‖‖G(p ref )‖ Among them, M(p, p ref ) is the similarity between the search space position p and the reference position p ref , and G(p ref ) is the steering response power distribution corresponding to the reference position p ref ; The greater the similarity of M(p, p ref ), the higher the similarity between the target at the search space position p and the reference target.

2. The method according to claim 1, characterized in that, the said Step 1 includes: All sensors in the distributed sensor scenario simultaneously collect signals from the sound sources within the coverage of all sensors at the same sampling frequency to obtain N target signals.

3. The method according to claim 1, characterized in that, the said Step 2 includes: Simultaneously perform frequency band division on N target signals according to the same frequency band range, and at the same time pair up the sensors in the distributed sensor scenario two by two, with a maximum of sensor pairs; in actual use, only part of the sensor pair data can be used, and the quantity is between 2 and .

4. The method according to claim 1, characterized in that, the said Step 53 includes: Using σ 2 (p) and M(p, p ref )'s Hadamard product as the fusion result, that is The said fusion result is the localization and recognition result of the sound source.

Images