Unmanned aerial vehicle sound processing method and device based on frequency importance feature filter, equipment and medium

Abstract

The unmanned aerial vehicle sound processing method, device and equipment based on a frequency importance feature filter and a medium, the method comprises the following steps: collecting target unmanned aerial vehicle sound signal samples and background noise samples, and constructing a training set; performing time-frequency analysis on the training samples, and calculating the power spectrum and energy distribution of each audio frame; calculating the F-ratio ratio of all frequency ranges as an index for measuring the importance of the frequency in distinguishing target sound and background noise; configuring a non-uniform filter bank according to the F-ratio ratio, so that the frequency resolution of the filter bank is proportional to the F-ratio; wherein the number of filters is configurable; for the sound signal to be identified, the log non-uniform spectrum features of the sound signal are extracted by using the filter bank, that is, the logarithm of the energy output by each filter. The present application can extract sound features with more key discriminant frequency bands, thereby improving the performance of unmanned aerial vehicle sound target recognition.

Description

Technical Field

[0001] This invention relates to the field of acoustic signal processing and pattern recognition technology, and in particular to a method, apparatus, device and medium for UAV acoustic processing based on a frequency importance feature filter. Background Technology

[0002] During flight, the power system of a drone (including electric motor, piston engine, rotor blades, etc.) generates acoustic radiation signals with specific spectral structures, and their acoustic characteristics vary significantly. Based on this, drone acoustic target recognition has important applications in low-altitude air defense, disaster early warning, agricultural sowing, and aerial logistics.

[0003] In existing acoustic target recognition technologies, the feature extraction stage widely employs fixed-structure filter banks to extract spectral features from acoustic signals, and then inputs these spectral features into a classifier to complete the recognition. Common filter banks include Mel-filter banks and Gammatone filter banks; however, the design of these filter banks is based on the characteristics of human hearing (such as logarithmic frequency scales), and is not an optimal design for specific acoustic targets. Directly applying them to UAV acoustic target recognition has the following shortcomings: The discriminative information of different acoustic targets (such as those with different rotor configurations or fixed rotor hybrids) is not uniformly distributed in the frequency domain. Some frequency bands are more important for distinguishing targets from background noise, while others are more distinguishable between targets. Traditional fixed filter banks cannot adaptively emphasize these key frequency bands.

[0004] When conducting acoustic detection on drones, the spectral distribution varies depending on the drone's power source and distance (fixed-wing, rotary-wing, or hybrid-wing). Fixed-wing drones exhibit harmonic relationships with the number of cylinders in a piston engine, while rotary-wing drones are related to the number of blades and the operating speed of the electric motor. Hybrid-wing drones represent a superposition of sound contributions from both power sources. In practice, the scenarios for drone detection are usually fixed, and the type of drone is typically also defined. While traditional Mel-triangle filters, using a fixed filter bank, can extract relatively consistent acoustic signatures, the drone's acoustic radiation characteristics, combined with the effects of the deployment environment, result in a strong non-uniform variation in the frequency response of its acoustic energy. Therefore, more adaptive feature filters need to be designed for specific targets.

[0005] Existing filter banks struggle to extract acoustic features from the more critical discrimination frequency bands, limiting the performance of UAV acoustic target recognition. Summary of the Invention

[0006] To address the shortcomings of the existing technology, this invention provides a method, apparatus, device, and medium for UAV acoustic processing based on frequency importance feature filters, employing the following technical solution: Firstly, a method for processing UAV acoustic signals based on frequency importance feature filters is provided, including: Collect acoustic signal samples and background noise samples of the target UAV and construct them into a training set; Perform time-frequency analysis on the training samples to calculate the power spectrum and energy distribution of each audio frame; Calculate the F-ratio across all frequency ranges as an indicator of the importance of that frequency in distinguishing target sound from background noise; Configure a non-uniform filter bank according to the F-ratio, such that the frequency resolution of the filter bank is proportional to the F-ratio; wherein the number of filters is configurable. For the acoustic signal to be identified, the logarithmic non-uniform spectral features are extracted using the filter bank, which is the logarithm of the output energy of each filter.

[0007] Secondly, a method for drone sound recognition based on frequency importance feature filters is provided, in which logarithmic non-uniform spectrum features as described in the first aspect are input into a preset recognition classifier for sound target recognition, so as to at least determine whether a drone is a non-drone and / or obtain drone model information.

[0008] Thirdly, a drone acoustic processing device based on a frequency importance feature filter is provided, comprising: The data acquisition module is used to collect acoustic signal samples and background noise samples of the target UAV and build them into a training set; The time-frequency analysis module is used to perform time-frequency analysis on the training samples and calculate the power spectrum and energy distribution of each audio frame. The frequency importance assessment module is used to calculate the F-ratio ratio across all frequency ranges, serving as an indicator of the frequency's importance in distinguishing target sound from background noise. A filter bank design module is used to configure a non-uniform filter bank according to the F-ratio, such that the frequency resolution of the filter bank is proportional to the F-ratio; wherein the number of filters can be selected. The feature extraction module is used to extract the logarithmic non-uniform spectral features of the acoustic signal to be identified using the filter bank, that is, the logarithm of the output energy of each filter.

[0009] Fourthly, an electronic device is provided, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, performs the steps of the method as described in the first aspect.

[0010] Fifthly, a readable storage medium is provided, on which a program or instructions are stored, which, when executed by a processor, implement the steps of the method as described in the first aspect.

[0011] This application has the following beneficial effects: This application proposes using the F-ratio statistical method to calculate the frequency importance of target data, using this as a key parameter for feature extraction filter design. Estimating the F-ratio provides a clear understanding of which frequency bands play a crucial role in recognition, aiding in understanding the physical characteristics of the target acoustic signal. By quantifying frequency importance through the F-ratio, the designed filter bank can highlight the frequency bands most discriminative for acoustic target recognition, capturing subtle features of the target signal better than fixed filter banks (such as Mel filter banks). The filter bank design is entirely based on training data, requiring no manual frequency band division, and is suitable for customized feature extraction and feature set data updates for different acoustic targets (such as drones, industrial machines, and specific equipment). By selecting the number of filters, feature dimensionality and computational complexity can be balanced, achieving controllable feature dimensionality. Simultaneously, it addresses the problem that inherent filters for acoustic recognition features cannot adapt to the target's frequency characteristics. Attached Figure Description

[0012] Figure 1 A schematic flowchart illustrating a UAV acoustic processing method based on a frequency importance feature filter, provided for an embodiment of this application; Figure 2 Another normalized F-ratio curve provided for embodiments of this application; Figure 3 A frequency distribution diagram of a non-uniform filter bank is provided for an embodiment of this application; Figure 4 A comparison diagram of extracted LNS features provided for an embodiment of this application; Figure 5 A comparison chart of the recognition rates of different feature extraction methods provided in the embodiments of this application. Detailed Implementation

[0013] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the technical solutions in the embodiments of this application are clearly described. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art are within the scope of protection of this application.

[0014] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such terms can be used interchangeably where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0015] The steps described in this application and the flowcharts in the accompanying drawings are not necessarily strictly executed according to the step numbers; the execution order of the steps can be changed. Furthermore, certain steps can be omitted, multiple steps can be combined into one step, and / or one step can be broken down into multiple steps.

[0016] This specification provides a UAV acoustic processing method based on a frequency importance feature filter, and also relates to a UAV acoustic recognition method based on a frequency importance feature filter, a UAV acoustic processing apparatus based on a frequency importance feature filter, a computer device, and a computer-readable storage medium, for extracting discriminative features for UAV acoustic target recognition, solving the problem that existing fixed filter banks cannot highlight key frequency bands of target signals. The following detailed description, in conjunction with the accompanying drawings and preferred embodiments, is provided in detail.

[0017] Please see Figure 1-5 This application provides a method for UAV acoustic processing based on a frequency importance feature filter, including: Step S1: Collect acoustic signal samples and background noise samples of the target UAV and construct them as a training set.

[0018] In practice, drone sound exhibits a distinct harmonic structure, with the fundamental frequency typically ranging from tens to hundreds of hertz, and energy concentrated in specific frequency bands. Background noise includes environmental noise, wind noise, and human voices. For example, 2000 drone sound samples are collected as target class samples, and 2000 background noise samples are collected as non-target class samples, both at a sampling rate of 16kHz. All samples are 10 seconds long. In some possible implementations, they can be further divided into a training set (80%) and a test set (20%). After training, the test set data is input into the pre-designed filter bank and the trained classifier for testing and evaluation.

[0019] Step S2: Perform time-frequency analysis on the training samples and calculate the power spectrum and energy distribution of each audio frame.

[0020] In practice, a short-time Fourier transform is used with a window length of 1024 points and a frame shift of 512 points. A Hamming window is selected as the window function to obtain the power spectrum matrix for each sample. and frequency Energy on Where t is the frame index, Frequency index (513 in total) (corresponding to 0-8kHz).

[0021] Step S3: Calculate the F-ratio for all frequency ranges as an indicator of the importance of that frequency in distinguishing target sound from background noise.

[0022] This step iterates through and calculates the F-ratio statistic for all frequency ranges of each training sample, quantifying its importance in distinguishing target sound from background noise.

[0023] Furthermore, the formula for calculating the F-ratio is: in, 'c' is the frequency index, and 'c' is the category (target class or background class). For the i-th sample of category c, in terms of frequency Energy on the surface For category c in The average energy on, For samples of all categories The total average energy is given by N, where N is the number of samples in each class.

[0024] For each frequency Calculate the target class and background class at this frequency respectively. The mean and variance of energy are substituted into the F-ration formula to calculate F( ). F( Normalize to [0,1]. Figure 2 The normalized F-ratio curves are shown. As can be seen from the figure, peaks appear in 80-120Hz (blade fundamental frequency range), 240-300Hz (second harmonic range corresponding to the fundamental frequency), and 5-6kHz (higher harmonic range), indicating that these frequency bands are most important for distinguishing drones from noise.

[0025] Step S4: Configure a non-uniform filter bank according to the F-ratio ratio, so that the frequency resolution of the filter bank is proportional to the F-ratio; wherein the number of filters can be configured.

[0026] This step designs a non-uniform filter bank based on the distribution pattern of F-ratio at various frequencies, ensuring that the frequency resolution of the filter is proportional to the calculated F-ratio statistics at each frequency. Further, step S4: configuring a non-uniform filter bank according to the stated F-ratio, ensuring that the frequency resolution of the filter bank is proportional to the F-ratio, includes: Step S401: Normalize the F-ratio; Step S402: Calculate the weighting factors ,in Sampling rate; preferably, ; Step S403: Calculate the weighted cumulative sum ; Step S404: Select M points uniformly on the cumulative sum and scale, and obtain the corresponding linear frequency values ​​through interpolation, which will be used as the center frequencies of the M filters. ; Step S405: Determine the bandwidth of the m-th (1,2,…M) filter based on the difference between adjacent center frequencies. , ]; Step S406: Configure a triangular bandpass filter according to the obtained bandwidth parameters to form a non-uniform filter bank; its frequency response for: .

[0027] For example, if the number of filters is set to M=64, then 64 points are uniformly selected on the cumulative sum scale, and the corresponding linear frequency value is obtained through spline interpolation. The center frequency of the filter is used as the center frequency; the bandwidth of the filter is determined based on the difference between the center frequencies. Using this bandwidth as a parameter, configure a triangular bandpass filter to form a non-uniform filter bank; obtain the filter bank response matrix. .

[0028] Figure 3 The frequency distribution of the designed non-uniform filter bank shows that the filter distribution is denser and the bandwidth is narrower in the frequency band with a high F-ratio ratio (such as 80~120Hz, 5~6kHz).

[0029] Step S5: For the acoustic signal to be identified, extract its log non-uniform spectrum (LNS) features using the filter bank, which is the logarithm of the output energy of each filter.

[0030] The calculation formula for LNS feature extraction is: Where t is the frame index and m is the filter index. Power spectrum, Let m be the frequency response of the filter. It is a small constant.

[0031] Power spectrum for each sample Calculate subband energy Then take the logarithm to obtain the LNS features, with a dimension of (number of frames × 64).

[0032] Based on the above technical solution, this application has the following outstanding substantive effects and significant progress: 1. By quantifying the importance of frequencies using the F-ratio, the designed filter bank can highlight the frequency bands that are most discriminative for acoustic target identification, adaptively focusing on key frequencies, and capturing subtle features of target signals better than fixed filter banks (such as Mel filter banks).

[0033] 2. Visualizing the F-ratio can intuitively show which frequency bands play a key role in identification, which helps to understand the physical characteristics of the target acoustic signal and has strong interpretability for feature classification and identification.

[0034] 3. The filter bank is designed entirely based on training data, without the need for manual frequency band division, and is suitable for customized feature extraction of different acoustic targets (such as drones, industrial machines, and specific equipment).

[0035] 4. By selecting the number of filters, the feature dimension and computational complexity can be balanced.

[0036] It is worth noting that some related technologies have attempted to design filter banks using data-driven methods, such as frequency band selection based on linear discriminant analysis or mutual information criteria. However, existing methods usually employ independent frequency band selection strategies, ignoring the continuity and smoothness between adjacent frequency bands. Moreover, most of them require pre-dividing a fixed set of frequency bands, failing to achieve a continuous match between the frequency resolution and frequency importance of the filter bank. They still struggle to highlight key discrimination frequency bands, limiting further improvements in recognition accuracy and robustness.

[0037] In contrast, this application proposes using the F-ratio to statistically calculate the frequency importance of target data, using this as a key parameter for feature extraction filter design. Estimating the F-ratio provides a clear understanding of which frequency bands play a crucial role in recognition, aiding in understanding the physical characteristics of the target acoustic signal. By quantifying frequency importance through the F-ratio, the designed filter bank can highlight the frequency bands most discriminative for acoustic target recognition, capturing subtle features of the target signal better than fixed filter banks (such as Mel filter banks). The filter bank design is entirely based on training data, requiring no manual frequency band division, and is suitable for customized feature extraction and feature set data updates for different acoustic targets (such as drones, industrial machines, and specific equipment). By selecting the number of filters, feature dimensionality and computational complexity can be balanced, achieving controllable feature dimensionality. Simultaneously, it addresses the problem that inherent filters for acoustic recognition features cannot adapt to the target's frequency characteristics.

[0038] This application has great potential value for the identification of acoustic targets from close-range UAVs and can also be applied to the false alarm suppression processing of airborne acoustic target detection.

[0039] Please see Figure 1-2 This application provides a method for drone acoustic recognition based on a frequency importance feature filter, including: The logarithmic non-uniform spectral features are input into a preset recognition classifier for acoustic target recognition, so as to at least determine whether a drone is a non-drone and / or obtain drone model information.

[0040] The LNS features are input into a classifier for acoustic target recognition. Specifically, the recognition classifier can employ Gaussian mixture model, support vector machine, deep neural network, autoencoder, etc. Preferably, the LNS features are input into a Gaussian mixture model classifier (GMM, 16 mixtures), and the maximum likelihood criterion is used to output the recognition result. Simultaneously, Mel-spectral features of the same frame length are extracted, with 64 filters, and input into the GMM for classification, serving as a comparison.

[0041] Figure 5 A comparison of the recognition accuracy of the two features on the test set is presented. The LNS feature of this application achieves an average accuracy of 96.5% on the test set, while the Mel spectrum feature achieves 91.2%. This demonstrates that the non-uniform filter bank designed based on the F-ratio ratio can effectively improve the performance of UAV acoustic recognition.

[0042] Based on this, this application also has the following outstanding substantive effects and significant progress: 5. The extracted LNS features can be seamlessly integrated with various existing classifiers and are easy to integrate into existing recognition systems.

[0043] Corresponding to the above-described UAV acoustic processing method embodiment based on frequency importance feature filters, this application embodiment provides a UAV acoustic processing apparatus based on frequency importance feature filters, comprising: The data acquisition module is used to collect acoustic signal samples and background noise samples of the target UAV and build them into a training set; The time-frequency analysis module is used to perform time-frequency analysis on the training samples and calculate the power spectrum and energy distribution of each audio frame. The frequency importance assessment module is used to calculate the F-ratio ratio across all frequency ranges, serving as an indicator of the frequency's importance in distinguishing target sound from background noise. A filter bank design module is used to configure a non-uniform filter bank according to the F-ratio, such that the frequency resolution of the filter bank is proportional to the F-ratio; wherein the number of filters can be selected. The feature extraction module is used to extract the logarithmic non-uniform spectral features of the acoustic signal to be identified using the filter bank, that is, the logarithm of the output energy of each filter.

[0044] Furthermore, the filter bank design module is specifically used for: Normalize the F-ratio; Calculate weighting factors ,in Sampling rate; Calculate the weighted cumulative sum ; M points are uniformly selected on the cumulative and scale sides, and the corresponding linear frequency values ​​are obtained by interpolation, which are then used as the center frequencies of the M filters. ; The bandwidth of the m-th filter is determined based on the difference between adjacent center frequencies. , ]; Based on the obtained bandwidth parameters, a triangular bandpass filter is configured to form a non-uniform filter bank; its frequency response... for: .

[0045] Furthermore, the formula for calculating the F-ratio is: in, 'c' is the frequency index, and 'c' is the category (target class or background class). For the i-th sample of category c, in terms of frequency Energy on the surface For category c in The average energy on, For samples of all categories The total average energy is given by N, where N is the number of samples in each class.

[0046] Furthermore, the calculation formula for LNS feature extraction of logarithmic nonuniform spectrum is as follows: Where t is the frame index and m is the filter index. Power spectrum, Let m be the frequency response of the filter. It is a small constant.

[0047] The above-described UAV acoustic processing device based on frequency importance feature filters implements the steps and processes of the above-described UAV acoustic processing method based on frequency importance feature filters, and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0048] Corresponding to the above embodiments of the UAV sound processing method based on frequency importance feature filters, this application provides an electronic device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the steps and processes of the above embodiments of the UAV sound processing method based on frequency importance feature filters, and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0049] Memory can be used to store software programs and various data. Memory can primarily include a first storage area for storing programs or instructions and a second storage area for storing data. The first storage area can store the operating system, application programs or instructions required for at least one function (such as sound playback, image playback, etc.). Furthermore, memory can include volatile memory or non-volatile memory, or both. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory can be random access memory (RAM), static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (Synchlink DRAM, SLDRAM), and direct memory bus RAM (DRRAM). The memory in the embodiments of this application includes, but is not limited to, these and any other suitable types of memory.

[0050] The processor may include one or more processing units; optionally, the processor integrates an application processor and a modem processor, wherein the application processor mainly handles operations related to the operating system, user interface, and applications, while the modem processor mainly handles wireless communication signals, such as a baseband processor. It is understood that the aforementioned modem processor may also not be integrated into the processor.

[0051] Corresponding to the above-described embodiment of the UAV acoustic processing method based on frequency importance feature filters, this application embodiment also provides a readable storage medium storing a program or instructions. When the program or instructions are executed by a processor, they implement the steps and processes of the above-described embodiment of the UAV acoustic processing method based on frequency importance feature filters, and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0052] The processor is the processor in the electronic device described in the above embodiments of this application. The readable storage medium includes a computer-readable storage medium, such as a computer read-only memory (ROM), random access memory (RAM), a magnetic disk, or an optical disk.

[0053] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0054] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a computer software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0055] It is understood that the embodiments of this application have been described above in conjunction with the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. As those skilled in the art will know, various changes or equivalent substitutions can be made to these features and embodiments without departing from the spirit and scope of the invention. Furthermore, those skilled in the art, under the guidance or instruction of this application, can modify these features and embodiments to adapt to specific situations and materials without departing from the spirit and scope of the invention. Therefore, this invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the protection scope of this invention.

Claims

1. A method for processing UAV acoustic signals based on frequency importance feature filters, characterized in that, include: Collect acoustic signal samples and background noise samples of the target UAV and construct them into a training set; Perform time-frequency analysis on the training samples to calculate the power spectrum and energy distribution of each audio frame; Calculate the F-ratio across all frequency ranges as an indicator of the importance of that frequency in distinguishing target sound from background noise; Configure a non-uniform filter bank according to the F-ratio, such that the frequency resolution of the filter bank is proportional to the F-ratio; wherein the number of filters is configurable. For the acoustic signal to be identified, the logarithmic non-uniform spectral features are extracted using the filter bank, which is the logarithm of the output energy of each filter.

2. The UAV acoustic processing method based on a frequency importance feature filter according to claim 1, characterized in that, The formula for calculating the F-ratio is: in, 'c' is the frequency index, and 'c' is the category (target class or background class). For the i-th sample of category c, in terms of frequency Energy on the surface For category c in The average energy on, For samples of all categories The total average energy is given by N, where N is the number of samples in each class.

3. The UAV acoustic processing method based on a frequency importance feature filter according to claim 1, characterized in that, Configure a non-uniform filter bank according to the F-ratio, such that the frequency resolution of the filter bank is proportional to the F-ratio, including: Normalize the F-ratio; Calculate weighting factors ,in Sampling rate; Calculate the weighted cumulative sum ; M points are uniformly selected on the cumulative and scale sides, and the corresponding linear frequency values ​​are obtained by interpolation, which are then used as the center frequencies of the M filters. ; The bandwidth of the m-th filter is determined based on the difference between adjacent center frequencies. , ]; Based on the obtained bandwidth parameters, a triangular bandpass filter is configured to form a non-uniform filter bank; its frequency response... for: 。 4. The UAV acoustic processing method based on a frequency importance feature filter according to claim 1, characterized in that, The calculation formula for LNS feature extraction of logarithmic nonuniform spectrum is as follows: Where t is the frame index and m is the filter index. Power spectrum, Let m be the frequency response of the filter. It is a small constant.

5. A method for UAV acoustic recognition based on frequency importance feature filters, characterized in that, include: The logarithmic non-uniform spectrum features as described in any one of claims 1-4 are input into a preset recognition classifier for acoustic target recognition, so as to at least determine whether a drone is a non-drone and / or obtain drone model information.

6. A drone acoustic processing device based on a frequency importance feature filter, characterized in that, include: The data acquisition module is used to collect acoustic signal samples and background noise samples of the target UAV and build them into a training set; The time-frequency analysis module is used to perform time-frequency analysis on the training samples and calculate the power spectrum and energy distribution of each audio frame. The frequency importance assessment module is used to calculate the F-ratio ratio across all frequency ranges, serving as an indicator of the frequency's importance in distinguishing target sound from background noise. A filter bank design module is used to configure a non-uniform filter bank according to the F-ratio, such that the frequency resolution of the filter bank is proportional to the F-ratio; wherein the number of filters can be selected. The feature extraction module is used to extract the logarithmic non-uniform spectral features of the acoustic signal to be identified using the filter bank, that is, the logarithm of the output energy of each filter.

7. The UAV acoustic processing device based on a frequency importance feature filter according to claim 6, characterized in that, The filter bank design module is specifically used for: Normalize the F-ratio; Calculate weighting factors ,in Sampling rate; Calculate the weighted cumulative sum ; M points are uniformly selected on the cumulative and scale sides, and the corresponding linear frequency values ​​are obtained by interpolation, which are then used as the center frequencies of the M filters. ; The bandwidth of the m-th filter is determined based on the difference between adjacent center frequencies. , ]; Based on the obtained bandwidth parameters, a triangular bandpass filter is configured to form a non-uniform filter bank; its frequency response... for: 。 8. The UAV acoustic processing device based on a frequency importance feature filter according to claim 6, characterized in that, The formula for calculating the F-ratio is: in, 'c' is the frequency index, and 'c' is the category (target class or background class). For the i-th sample of category c, in terms of frequency Energy on the surface For category c in The average energy on, For samples of all categories The total average energy over the class, where N is the number of samples in each class; The calculation formula for LNS feature extraction of logarithmic nonuniform spectrum is as follows: Where t is the frame index and m is the filter index. Power spectrum, Let m be the frequency response of the filter. It is a small constant.

9. An electronic device, characterized in that, The electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the UAV acoustic processing method based on a frequency importance feature filter as described in any one of claims 1 to 4.

10. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the steps of the UAV acoustic processing method based on a frequency importance feature filter as described in any one of claims 1 to 4.

Images