Unmanned aerial vehicle acoustic identification method and device based on frequency band feature extraction

Abstract

The invention provides an unmanned aerial vehicle acoustic recognition method and device based on frequency band feature extraction, and the method comprises the steps: collecting an environment sound signal, carrying out the preprocessing, and generating a target sound signal; decomposing the target sound signal into an intrinsic mode component, filtering a noise component in the intrinsic mode component, and restoring the intrinsic mode component into a noise reduction sound signal; multiple acoustic features corresponding to the noise reduction sound signal are extracted, the acoustic features at least comprise frequency band filtering cepstrum coefficient features, and the multiple acoustic features are fused into a multi-dimensional feature vector; and inputting the multi-dimensional feature vector into a pre-trained acoustic classification identification model, and determining an unmanned aerial vehicle identification result. By performing deep learning modeling on the acoustic features of the unmanned aerial vehicle, high-precision target detection and classification in a complex environment are realized, the target unmanned aerial vehicle can be accurately identified under the condition of relatively large environmental noise, and the reliability and adaptability of acoustic detection are improved.

Description

Technical Field

[0001] The present disclosure relates to the technical field of drone target recognition, and in particular to a drone acoustic recognition method and device based on frequency band feature extraction. Background Art

[0002] In recent years, drone technology has developed rapidly and has been widely used in military, logistics, security, agriculture and other fields. However, with the rapid growth in the number of drones, the demand for detection and prevention of illegal intrusion, abuse and potential threats of drones is increasing. Traditional drone detection technology mainly relies on radar, optical imaging and radio frequency signal processing, but these methods have certain limitations in certain application scenarios.

[0003] Although radar detection can cover a large range, it is easily affected by multipath effects and low-flying obstacles in the environment. In addition, small drones may be difficult to accurately capture and identify due to their small radar cross-section. Optical imaging relies on high-resolution camera equipment and is prone to failure in complex weather conditions such as strong light, rain, fog and haze. In addition, this method usually requires the target to remain within the field of view, which has certain limitations on the monitoring range. The RF signal detection method relies on capturing the drone communication signal. However, some drones may adopt autonomous flight mode or encrypted communication methods, resulting in the risk of missed detection in RF signal detection. In addition, there may be a large false alarm rate in complex electromagnetic environments. Summary of the invention

[0004] The embodiments of the present disclosure at least provide a method and device for drone acoustic identification based on frequency band feature extraction, which achieves high-precision target detection and classification in complex environments by performing deep learning modeling on the acoustic features of drones, and can accurately identify target drones under conditions of high environmental noise, thereby improving the reliability and adaptability of acoustic detection.

[0005] The present disclosure provides a method for drone acoustic recognition based on frequency band feature extraction, including:

[0006] Collect environmental sound signals and pre-process them to generate target sound signals;

[0007] Decomposing the target sound signal into eigenmode components, filtering noise components in the eigenmode components, and restoring the eigenmode components to a noise-reduced sound signal;

[0008] Extracting a plurality of acoustic features corresponding to the noise reduction sound signal, wherein the acoustic features at least include frequency band filtering cepstral coefficient features, and fusing the plurality of acoustic features into a multi-dimensional feature vector;

[0009] The multi-dimensional feature vector is input into a pre-trained acoustic classification and recognition model to determine the drone recognition result.

[0010] In an optional implementation, collecting the ambient sound signal and preprocessing it to generate the target sound signal specifically includes:

[0011] The ambient sound signal is collected by using a microphone array, wherein the ambient sound signal includes at least a drone sound signal, ambient noise, and an interference signal;

[0012] The environmental sound signal is subjected to preprocessing including time-frequency filtering denoising, adaptive filtering denoising, gain adjustment and data normalization to generate the target sound signal.

[0013] In an optional implementation, decomposing the target sound signal into eigenmode components, filtering noise components in the eigenmode components, and restoring the eigenmode components to a noise-reduced sound signal specifically includes:

[0014] Using an eigenmode decomposition algorithm to process the target sound signal, decomposing the target sound signal into the eigenmode components;

[0015] Analyzing the intrinsic mode component by a fast independent component analysis algorithm, identifying the noise component and setting the noise component to zero;

[0016] The denoised intrinsic mode components are restored by using a reconstruction matrix, and the denoised intrinsic mode components are reconstructed into the denoised sound signal by using the eigenmode decomposition algorithm.

[0017] In an optional implementation, extracting multiple acoustic features corresponding to the noise reduction sound signal and fusing the multiple acoustic features into a multi-dimensional feature vector specifically includes:

[0018] Using a Mel filter to extract Mel-frequency cepstral coefficient features of the noise-reduced audio signal;

[0019] A Gammatone filter is used to replace the Mel filter to extract the Gammatone spectrum features corresponding to the noise reduction sound signal;

[0020] For the output of the Mel filter, Mel frequency cepstral coefficients are not extracted as Fbank features;

[0021] Using a filter group of a preset frequency band, extracting the frequency band filtering cepstrum coefficient characteristics corresponding to the noise reduction sound signal;

[0022] The Mel frequency cepstral coefficient feature, the Gammatone spectrum feature, the Fbank feature and the band filtering cepstral coefficient feature are fused to generate the multi-dimensional feature vector.

[0023] In an optional implementation, the acoustic classification recognition model is trained based on the following steps:

[0024] Constructing the acoustic classification recognition model including a convolutional layer, a pooling layer, a fully connected layer, and a Softmax layer, and using the ReLU function as an activation function;

[0025] Obtain a training sound signal data set including drone sound signals and background noise signals;

[0026] Annotating the training sound signal data set using the drone sound signal and the background noise signal as labels;

[0027] The training multidimensional feature vector corresponding to the training sound signal data set is extracted, the training multidimensional feature vector is input into the acoustic classification and recognition model for training, and the model parameters of the acoustic classification and recognition model are optimized using a cross entropy loss function and a gradient descent method.

[0028] The present disclosure also provides a drone acoustic recognition device based on frequency band feature extraction, including:

[0029] The sound signal acquisition module is used to collect environmental sound signals and pre-process them to generate target sound signals;

[0030] A noise reduction module, configured to decompose the target sound signal into eigenmode components, filter noise components in the eigenmode components, and restore the eigenmode components to a noise-reduced sound signal;

[0031] A feature extraction module, used to extract multiple acoustic features corresponding to the noise reduction sound signal, wherein the acoustic features at least include frequency band filtering cepstral coefficient features, and fuse the multiple acoustic features into a multi-dimensional feature vector;

[0032] The recognition module is used to input the multi-dimensional feature vector into a pre-trained acoustic classification recognition model to determine the drone recognition result.

[0033] In an optional implementation manner, the sound signal acquisition module is specifically used to:

[0034] The ambient sound signal is collected by using a microphone array, wherein the ambient sound signal includes at least a drone sound signal, ambient noise, and an interference signal;

[0035] The environmental sound signal is subjected to preprocessing including time-frequency filtering denoising, adaptive filtering denoising, gain adjustment and data normalization to generate the target sound signal.

[0036] In an optional implementation manner, the noise reduction module is specifically used to:

[0037] Using an eigenmode decomposition algorithm to process the target sound signal, decomposing the target sound signal into the eigenmode components;

[0038] Analyzing the intrinsic mode component by a fast independent component analysis algorithm, identifying the noise component and setting the noise component to zero;

[0039] The denoised intrinsic mode components are restored by using a reconstruction matrix, and the denoised intrinsic mode components are reconstructed into the denoised sound signal by using the eigenmode decomposition algorithm.

[0040] An embodiment of the present disclosure also provides an electronic device, including: a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor and the memory communicate through the bus, and when the machine-readable instructions are executed by the processor, the above-mentioned drone acoustic recognition method or the steps in any possible implementation of the above-mentioned drone acoustic recognition method based on frequency band feature extraction are performed.

[0041] The embodiments of the present disclosure also provide a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the computer program executes the steps of any possible implementation of the above-mentioned drone acoustic recognition method or the above-mentioned drone acoustic recognition method based on frequency band feature extraction.

[0042] The embodiments of the present disclosure also provide a computer program product, including a computer program / instruction, which, when executed by a processor, implements the above-mentioned drone acoustic identification method based on frequency band feature extraction, or the steps in any possible implementation of the above-mentioned drone acoustic identification method.

[0043] The disclosed embodiment provides a method and device for drone acoustic identification based on frequency band feature extraction, which collects environmental sound signals and performs preprocessing to generate target sound signals; decomposes the target sound signals into intrinsic mode components, filters the noise components in the intrinsic mode components, and restores the intrinsic mode components to noise-reduced sound signals; extracts multiple acoustic features corresponding to the noise-reduced sound signals, the acoustic features at least include frequency band filtering cepstral coefficient features, and fuses the multiple acoustic features into a multidimensional feature vector; inputs the multidimensional feature vector into a pre-trained acoustic classification and recognition model to determine the drone identification result. By performing deep learning modeling on the acoustic features of drones, high-precision target detection and classification in complex environments can be achieved, and target drones can be accurately identified under conditions of high environmental noise, thereby improving the reliability and adaptability of acoustic detection.

[0044] In order to make the above-mentioned objectives, features and advantages of the present disclosure more obvious and easy to understand, preferred embodiments are specifically cited below and described in detail with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the following is a brief introduction to the drawings required for use in the embodiments. The drawings herein are incorporated into the specification and constitute a part of the specification. These drawings illustrate embodiments consistent with the present disclosure and are used together with the specification to illustrate the technical solutions of the present disclosure. It should be understood that the following drawings only illustrate certain embodiments of the present disclosure and should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can also be obtained based on these drawings without creative work.

[0046] Figure 1 A flowchart of a method for drone acoustic recognition based on frequency band feature extraction provided by an embodiment of the present disclosure is shown;

[0047] Figure 2 A flowchart of another method for drone acoustic recognition based on frequency band feature extraction provided by an embodiment of the present disclosure is shown;

[0048] Figure 3 A schematic diagram of a drone acoustic identification device based on frequency band feature extraction provided by an embodiment of the present disclosure is shown;

[0049] Figure 4 A schematic diagram of an electronic device provided by an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical scheme and advantages of the embodiments of the present disclosure clearer, the technical scheme in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only part of the embodiments of the present disclosure, rather than all of the embodiments. The components of the embodiments of the present disclosure generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present disclosure provided in the drawings is not intended to limit the scope of the present disclosure for protection, but merely represents the selected embodiments of the present disclosure. Based on the embodiments of the present disclosure, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present disclosure.

[0051] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0052] The term "and / or" herein only describes an association relationship, indicating that three relationships may exist. For example, A and / or B may represent the following three situations: A exists alone, A and B exist at the same time, and B exists alone. In addition, the term "at least one" herein represents any combination of at least two of any one or more of a plurality of. For example, including at least one of A, B, and C may represent including any one or more elements selected from the set consisting of A, B, and C.

[0053] Research has found that traditional drone detection technology mainly relies on radar, optical imaging, and radio frequency signal processing, but these methods have certain limitations in certain application scenarios. Although radar detection can cover a large range, it is easily affected by multipath effects and low-altitude flying obstacles in the environment. In addition, small drones may be difficult to accurately capture and identify due to their small radar cross-section. Optical imaging relies on high-resolution camera equipment and is prone to failure under complex weather conditions such as strong light, rain, fog, and haze. In addition, this method usually requires the target to remain within the field of view, which has certain limitations on the monitoring range. The radio frequency signal detection method relies on capturing drone communication signals. However, some drones may adopt autonomous flight mode or encrypted communication methods, resulting in the risk of missed detection in radio frequency signal detection. In addition, there may be a large false alarm rate in complex electromagnetic environments.

[0054] Based on the above research, the present disclosure provides a method, device, electronic device and storage medium for drone acoustic identification based on frequency band feature extraction, which collects environmental sound signals and performs preprocessing to generate target sound signals; decomposes the target sound signals into intrinsic mode components, filters the noise components in the intrinsic mode components, and restores the intrinsic mode components to noise-reduced sound signals; extracts multiple acoustic features corresponding to the noise-reduced sound signals, the acoustic features at least include frequency band filtering cepstral coefficient features, and fuses multiple acoustic features into multidimensional feature vectors; inputs the multidimensional feature vectors into a pre-trained acoustic classification and recognition model to determine the drone identification results. By performing deep learning modeling on the acoustic features of drones, high-precision target detection and classification in complex environments can be achieved, and target drones can be accurately identified under conditions of high environmental noise, thereby improving the reliability and adaptability of acoustic detection.

[0055] To facilitate understanding of this embodiment, a method for drone acoustic identification based on frequency band feature extraction disclosed in the embodiment of the present disclosure is first introduced in detail. The execution subject of the method for drone acoustic identification based on frequency band feature extraction provided in the embodiment of the present disclosure is generally a computer device with certain computing capabilities, and the computer device includes, for example: a terminal device or a server or other processing device, and the terminal device can be a user equipment (User Equipment, UE), a mobile device, a user terminal, a terminal, a cellular phone, a cordless phone, a personal digital assistant (Personal Digital Assistant, PDA), a handheld device, a computing device, a vehicle-mounted device, a wearable device, etc. In some possible implementations, the method for drone acoustic identification based on frequency band feature extraction can be implemented by a processor calling a computer-readable instruction stored in a memory.

[0056] See also Figure 1 FIG. 1 is a flowchart of a method for acoustic identification of a drone based on frequency band feature extraction provided by an embodiment of the present disclosure, the method comprising steps S101 to S104, wherein:

[0057] S101: Collect environmental sound signals and perform preprocessing to generate target sound signals.

[0058] In the specific implementation, a microphone array is used to collect ambient sound signals, which include at least drone sound signals, ambient noise and interference signals. The ambient sound signals are preprocessed including time-frequency filtering denoising, adaptive filtering denoising, gain adjustment and data normalization to generate target sound signals.

[0059] Here, a microphone array is used as the acquisition device. The microphone array can capture sound information from different directions and has a high signal-to-noise ratio, which is convenient for subsequent signal processing. The environmental sound signals include: the drone sound signal is the sound in a specific frequency range emitted by the drone when flying, which is the identified target signal; environmental noise includes background noise such as wind, bird calls, traffic noise, etc., which may interfere with the extraction of target sound signals; interference signals such as non-related sound signals caused by other mechanical equipment or electromagnetic interference.

[0060] During acquisition, a suitable sampling rate (such as 44.1kHz or 48kHz) and quantization bit depth (such as 16 bits or 24 bits) can be set to ensure that the acquired audio signal has sufficient time domain and frequency domain resolution.

[0061] Furthermore, the purpose of preprocessing is to extract the target sound signal containing target information from the complex environmental sound signal. Time-frequency filtering denoising uses a variety of filters (such as low-pass filters, high-pass filters or band-pass filters) to remove noise within a specific frequency band. For example: low-pass filters remove high-frequency interference signals; band-pass filters retain signals in the drone sound frequency range; time domain smoothing: smoothing the amplitude of the audio signal to reduce random fluctuation interference in a short period of time. Adaptive filtering denoising uses an adaptive filtering algorithm (such as LMS or RLS algorithm) to dynamically adjust the parameters of the filter to adapt to the changing background noise environment, thereby improving the fidelity of the target signal.

[0062] Optionally, since the energy of the drone sound signal may be weak and easily masked by environmental noise, gain adjustment is required to increase the signal strength.

[0063] Furthermore, the audio signals collected at different times and in different environments are unified into a standard range to facilitate subsequent feature extraction and model input. The amplitude of the audio signal is normalized (such as scaling the signal amplitude to the range of [-1,1]), and short-time energy or signal-to-noise ratio is used as the normalization standard to balance the overall energy level of the audio signal.

[0064] Here, after the above preprocessing, the original ambient sound signal is processed into the target sound signal, most of the background noise and interference signals have been filtered out, the intensity and signal-to-noise ratio of the target sound signal are improved, and the normalized signal has a consistent amplitude range and frequency domain distribution, which is suitable for further analysis and processing.

[0065] As a possible implementation, assuming that the source sound signal is s i (t), where i represents the number of different sound sources, and the environmental noise can also be regarded as the source sound signal. i(t) is a continuous analog signal. The source sound signal reaches different sound sensors after passing through different paths and different distances in space with intensity attenuation. Let x i (t) is the sound analog signal received at the microphone, then:

[0066]

[0067] Among them, x i (t) represents the sound analog signal received at the microphone; a j (i) represents the equivalent weighted value of the j-th source sound signal after attenuation to the i-th microphone sound sensor; s i (t) represents the source sound signal.

[0068] Here, in practice, due to the electrical characteristics of the sound sensor itself, a tiny Gaussian white noise will be generated, and the signal output by the sensor is a digital signal, so in fact the signal received by the microphone array is a discrete digital signal: X(k) = X(n) n＝k +X noise , where X(k) represents the digital signal received by the kth microphone; X noise represents the noise signal received by the kth microphone; X(n) n＝k Represents the drone sound signal received by k microphones.

[0069] Among them, high-precision, high-sensitivity microphones and professional sound cards are used to collect environmental sounds. The sounds in the environment are collected in the form of an array, so that the algorithm can be used to restore the sound source later, which is conducive to reducing the impact of noise.

[0070] S102: Decompose the target sound signal into eigenmode components, filter noise components in the eigenmode components, and restore the eigenmode components to a noise-reduced sound signal.

[0071] In the specific implementation, the target sound signal is processed by the intrinsic mode decomposition algorithm to decompose the target sound signal into eigenmode components; the eigenmode components are analyzed by the fast independent component analysis algorithm to identify the noise components and set the noise components to zero; the reconstruction matrix is ​​used to restore the denoised eigenmode components, and the denoised eigenmode components are reconstructed into denoised sound signals by the intrinsic mode decomposition algorithm.

[0072] Here, Empirical Mode Decomposition (EMD) is a time-frequency analysis method used to decompose complex signals into several physically meaningful intrinsic mode functions (IMFs) and a residual. The EMD algorithm identifies all local maxima and minima in the target sound signal, interpolates the maxima and minima respectively, and obtains the upper envelope and the lower envelope. Calculate the mean of the upper envelope and the lower envelope, subtract the envelope mean from the current signal, and obtain the first IMF candidate component. Determine whether the first IMF candidate component meets the IMF condition. If not, repeat the above steps until the IMF component is obtained.

[0073] Among them, the extracted IMF is subtracted from the original signal, and the remaining signal is decomposed until the stop condition is met. Through the above steps, the target sound signal can be decomposed into multiple IMF components, each component represents the signal characteristics within a specific frequency range.

[0074] Furthermore, the noise components in the target sound signal usually exist in a specific IMF, and the IMF is processed by Fast Independent Component Analysis (FastICA) to identify and remove the noise components.

[0075] Here, FastICA is a blind signal separation method that separates the target signal from the noise signal by assuming that the components are statistically independent. Taking the IMF component as the input data matrix, each column of the matrix is ​​an IMF component. FastICA maximizes the non-Gaussianity, calculates the signal's unmixing matrix, decomposes the original signal into independent components, and uses energy analysis or characteristic statistical methods to identify the noise component and set it to zero or replace it.

[0076] The identified noise components are set to zero in the corresponding IMF to eliminate the noise impact of these frequency bands. The IMF components after noise filtering need to be recombined to generate the target sound signal after noise reduction.

[0077] Specifically, the processed independent component matrix is ​​recombined according to the demixing matrix to obtain the modified IMF matrix. All the denoised IMF components are added to the residual to reconstruct the denoised sound signal, and the reconstructed signal is smoothed to further improve the quality of the denoised signal and eliminate artifacts introduced by noise zeroing or other operations.

[0078] In this way, the process of decomposing the target sound signal into intrinsic mode components through EMD, filtering the noise components using FastICA, and reconstructing the denoised IMF components into denoised sound signals fully utilizes the advantages of signal decomposition and reconstruction technology in acoustic processing, which not only improves the quality of the target sound signal, but also lays the foundation for high-precision analysis of UAV acoustic recognition.

[0079] S103: extract multiple acoustic features corresponding to the noise reduction sound signal, where the acoustic features at least include band filtering cepstral coefficient features, and fuse the multiple acoustic features into a multi-dimensional feature vector.

[0080] In the specific implementation, a Mel filter is used to extract the Mel-frequency cepstral coefficient features of the noise-reduced sound signal; a Gammatone filter is used instead of the Mel filter to extract the Gammatone spectrum features corresponding to the noise-reduced sound signal; for the output of the Mel filter, the Mel-frequency cepstral coefficients are not extracted as Fbank features; a filter group with a preset frequency band is used to extract the frequency band filtering cepstral coefficient features corresponding to the noise-reduced sound signal; the Mel-frequency cepstral coefficient features, Gammatone spectrum features, Fbank features and frequency band filtering cepstral coefficient features are fused to generate a multi-dimensional feature vector.

[0081] Here, the Mel Frequency Cepstral Coefficient (MFCC) feature is a feature widely used in speech processing. It can capture the main acoustic features of drone sounds by simulating the nonlinear frequency distribution of human auditory perception and extracting the low-frequency characteristics of sounds. The noise-reduced sound signal is subjected to a short-time Fourier transform (STFT) to obtain the power spectrum, which is then mapped to the Mel frequency scale. The Mel filter bank is used to simulate the auditory characteristics of the human ear, the Mel frequency power spectrum is logarithmic, and then a discrete cosine transform (DCT) is performed to obtain the MFCC feature.

[0082] Here, the Gammatone filter is used for Gammatone spectrum feature extraction. The Gammatone filter is a filter based on the auditory model that can extract the frequency energy distribution of the signal. The Gammatone spectrum can capture the subtle spectrum changes of the signal, which is particularly suitable for the micro-features of drone sounds. The Gammatone filter group is used to process the noise reduction sound signal. Each filter corresponds to a center frequency, and the energy distribution of each filter is calculated to generate the Gammatone spectrum diagram.

[0083] Here, the Fbank feature is the power spectrum output by the Mel filter, which directly retains the spectrum energy information without DCT conversion. The noise-reduced audio signal is subjected to short-time Fourier transform, and the frequency band energy is obtained through the Mel filter bank. The Fbank feature retains the complete spectrum energy distribution of the signal and is suitable for fusion with other features to improve the feature expression ability of the model.

[0084] Here, the band-filtered cepstral coefficient is an extended cepstral feature. The filter bank is designed based on the custom band division to extract the signal characteristics of a specific band. The noise reduction sound signal is band-segmented using the filter bank with preset band division. The power spectrum is calculated for each band signal and the cepstral features are extracted. The band-filtered cepstral coefficient feature can capture the energy distribution of a specific frequency band in the drone sound and provide more refined spectral information for the model.

[0085] Furthermore, since different acoustic features have different dimensions and ranges, they need to be processed uniformly. This process includes alignment and normalization. The alignment process ensures that the number of time steps for each feature is consistent. The normalization process performs maximum value normalization or zero mean unit variance normalization on all features to eliminate the numerical differences between features.

[0086] Here, the MFCC, Gammatone spectrum, Fbank features and band filter cepstral coefficient features extracted above are spliced ​​according to the time dimension to generate a unified feature matrix. The fused feature matrix is ​​expanded into feature vectors by column, and the feature splicing results of each time slice are spliced ​​to generate a multi-dimensional feature vector.

[0087] Among them, different features focus on different dimensions of the signal (such as time domain features, frequency domain features, and cepstrum features). Fusion can improve the model's ability to express complex signals. Multi-feature fusion can improve the system's robustness to noise and interference and enhance the recognition accuracy of drone sounds. The fused multi-dimensional feature vector contains rich information and can adapt to the drone acoustic recognition needs of different scenarios.

[0088] In this way, by extracting multiple acoustic features of the noise reduction sound signal (MFCC, Gammatone spectrum, Fbank and band filter cepstral coefficients), and fusing these features into a multi-dimensional feature vector, high-dimensional and high-resolution input data is provided for the subsequent acoustic classification and recognition model. The fusion process makes full use of the advantages of various features and improves the accuracy and robustness of the drone acoustic recognition method.

[0089] S104: input the multi-dimensional feature vector into a pre-trained acoustic classification and recognition model to determine the drone recognition result.

[0090] In the specific implementation, the model framework of the acoustic classification and recognition model is based on a convolutional neural network (CNN). The convolution layer is used to extract local patterns of input features, such as correlations in frequency and time dimensions. The pooling layer is used to reduce the resolution of the features and the computational complexity. The convolution layer and the pooling layer are stacked multiple times to capture higher-level features.

[0091] Here, the fully connected layer maps the high-dimensional features output by the convolution module to a low-dimensional space, compressing the multi-dimensional features into a specific dimensional vector suitable for classification. The classification layer (Softmax layer) outputs the probability distribution of different categories, each category corresponds to a probability value, and the category with the highest probability is considered to be the final classification result. The activation function uses the ReLU function for nonlinear transformation to increase the expressive power of the model.

[0092] The input of the acoustic classification and recognition model is a multi-dimensional feature vector, which contains the time and spectrum information of the drone signal. The output is a category label (such as "drone A", "drone B", "background noise").

[0093] As a possible implementation, see Figure 2 FIG. 2 is a flowchart of another method for drone acoustic identification based on frequency band feature extraction provided by an embodiment of the present disclosure, wherein the method includes steps S201 to S204, wherein:

[0094] S201, constructing the acoustic classification recognition model including a convolutional layer, a pooling layer, a fully connected layer and a Softmax layer, and using the ReLU function as the activation function.

[0095] S202: Acquire a training sound signal data set including drone sound signals and background noise signals.

[0096] S203: labeling the training sound signal data set using the drone sound signal and the background noise signal as labels.

[0097] S204, extracting the training multidimensional feature vector corresponding to the training sound signal data set, inputting the training multidimensional feature vector into the acoustic classification and recognition model for training, and optimizing the model parameters of the acoustic classification and recognition model using a cross entropy loss function and a gradient descent method.

[0098] In the specific implementation, the data sources of the training data include the collected drone sound signals and various environmental noise signals. The sound signals are divided into multiple categories, such as "drone A sound", "drone B sound", "background noise", etc., and the multi-dimensional feature vectors corresponding to the training data are extracted as the input of the model.

[0099] Here, cross entropy is used to quantify the deviation between the class probability predicted by the model and the true class label, and the gradient descent method (such as Adam optimizer) is used to update the model parameters and minimize the loss function. In order to improve the adaptability of the model to different environmental noises, data enhancement techniques are used to add random noise, adjust gain, change time-frequency characteristics, etc.

[0100] Exemplarily, a high-precision microphone is used to record drone sound source files of different distances (1 meter to 250 meters), different models, and different flight postures (approaching, moving away, turning, suspending, rising, and falling). The specific recording method is: record a certain model of drone taking off from the microphone, flying to a distance of 40 meters, and then moving forward 20 meters after a set of actions including approaching, principle, turning, suspending, rising, and falling are completed every 20 meters, and the operation is repeated. Classify according to the drone model, and record non-drone environmental noise of similar sample volume at the same time. Complete the preparation of drone multi-classification original training data. Perform data preprocessing on the original training data, unify the sampling rate, gain adjustment and normalization of all signal data, and then pre-emphasize the signal data to increase the gain of high-frequency components. In order to unify the input dimension of the convolutional neural network, all pre-processed sound signals need to be segmented according to certain rules (every 0.5-3.5 seconds), such as segmenting the sound signal into small samples of 3.0 seconds. Use Matlab to build a convolutional neural network model and use Matlab for model training.

[0101] Furthermore, after the training is completed, the multi-dimensional feature vector obtained by preprocessing is input into the model for drone recognition reasoning. The multi-dimensional feature vector is input into the model frame by frame. Each frame contains information in the time domain and frequency domain. The convolution layer extracts the local pattern of the feature vector through a sliding window to generate a feature map. The pooling layer downsamples the feature map to retain key information and reduce the amount of calculation. The output of the convolution module is flattened into a vector form, and the features are combined through the fully connected layer to extract high-level patterns. The probability distribution of each category is calculated through the Softmax layer, and the category corresponding to the maximum probability is selected from the probability distribution of the Softmax output as the recognition result.

[0102] Here, the recognition results are output in the form of category labels, for example: "Drone A": indicates that a specific model of drone signal is detected; "Background noise": indicates that no drone sound is detected.

[0103] The above solution will be described below in conjunction with specific implementation methods.

[0104] The same data acquisition equipment and method are used to collect sound data in the environment in real time as the original data for recognition. After the same preprocessing, the segmentation part is carried out in a feedback manner. When the prediction of the previous sample data is completed, the completion signal is fed back to the signal segmentation part, and the sound signal is segmented based on the time point when the completion signal is received and the sound signal in the previous certain period of time, and then input into the feature extraction and convolutional neural network recognition for recognition. It can make full use of computing resources to achieve real-time drone recognition. The recognition accuracy can be maintained between 92% and 99%, and the real-time performance is to recognize the sound in the environment more than 10 times per second. The final recognition result is expressed as a set of recognition result vectors, including the prediction results of different drone model categories. The prediction result of 1 means that the drone of this category has been identified, and the prediction result of 0 means that the sound does not belong to this type.

[0105] The disclosed embodiment provides a method for drone acoustic identification based on frequency band feature extraction, which collects environmental sound signals and performs preprocessing to generate target sound signals; decomposes the target sound signals into intrinsic mode components, filters the noise components in the intrinsic mode components, and restores the intrinsic mode components to noise-reduced sound signals; extracts multiple acoustic features corresponding to the noise-reduced sound signals, the acoustic features at least include frequency band filtering cepstral coefficient features, and fuses multiple acoustic features into a multidimensional feature vector; inputs the multidimensional feature vector into a pre-trained acoustic classification and recognition model to determine the drone identification result. By performing deep learning modeling on the acoustic features of drones, high-precision target detection and classification in complex environments can be achieved, and target drones can be accurately identified under conditions of high environmental noise, thereby improving the reliability and adaptability of acoustic detection.

[0106] Those skilled in the art will appreciate that, in the above method of specific implementation, the order in which the steps are written does not imply a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of the steps should be determined by their functions and possible internal logic.

[0107] Based on the same inventive concept, the embodiments of the present disclosure also provide a drone acoustic identification device based on frequency band feature extraction corresponding to the drone acoustic identification method based on frequency band feature extraction. Since the principle of solving the problem by the device in the embodiments of the present disclosure is similar to the above-mentioned drone acoustic identification method in the embodiments of the present disclosure, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be repeated.

[0108] See also Figure 3 , Figure 3 A schematic diagram of a drone acoustic recognition device based on frequency band feature extraction provided by an embodiment of the present disclosure. Figure 3As shown in , the drone acoustic recognition device 300 based on frequency band feature extraction provided by the embodiment of the present disclosure includes:

[0109] The sound signal collection module 310 is used to collect environmental sound signals and perform preprocessing to generate target sound signals.

[0110] The noise reduction module 320 is used to decompose the target sound signal into eigenmode components, filter the noise components in the eigenmode components, and restore the eigenmode components to a noise-reduced sound signal.

[0111] The feature extraction module 330 is used to extract multiple acoustic features corresponding to the noise reduction sound signal, wherein the acoustic features at least include frequency band filtering cepstral coefficient features, and fuse the multiple acoustic features into a multi-dimensional feature vector.

[0112] The recognition module 340 is used to input the multi-dimensional feature vector into a pre-trained acoustic classification recognition model to determine the drone recognition result.

[0113] For descriptions of the processing flow of each module in the device and the interaction flow between each module, reference may be made to the relevant descriptions in the above method embodiment, which will not be described in detail here.

[0114] The disclosed embodiment provides a drone acoustic identification device based on frequency band feature extraction, which collects environmental sound signals and performs preprocessing to generate target sound signals; decomposes the target sound signals into intrinsic mode components, filters the noise components in the intrinsic mode components, and restores the intrinsic mode components to noise-reduced sound signals; extracts multiple acoustic features corresponding to the noise-reduced sound signals, the acoustic features at least include frequency band filtering cepstral coefficient features, and fuses multiple acoustic features into a multidimensional feature vector; inputs the multidimensional feature vector into a pre-trained acoustic classification and recognition model to determine the drone recognition result. By performing deep learning modeling on the acoustic features of drones, high-precision target detection and classification in complex environments can be achieved, and target drones can be accurately identified under conditions of high environmental noise, thereby improving the reliability and adaptability of acoustic detection.

[0115] Corresponds to Figure 1 The drone acoustic recognition method based on frequency band feature extraction in the present disclosure also provides an electronic device 400, such as Figure 4 FIG. 4 is a schematic diagram of the structure of an electronic device 400 provided in an embodiment of the present disclosure, including:

[0116] Processor 41, memory 42, and bus 43; memory 42 is used to store execution instructions, including memory 421 and external memory 422; memory 421 here is also called internal memory, which is used to temporarily store operation data in processor 41 and data exchanged with external memory 422 such as hard disk. Processor 41 exchanges data with external memory 422 through memory 421. When the electronic device 400 is running, the processor 41 communicates with the memory 42 through bus 43, so that the processor 41 executes Figure 1 and Figure 2 The steps of the drone acoustic identification method based on frequency band feature extraction.

[0117] The present disclosure also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the drone acoustic identification method based on frequency band feature extraction described in the above method embodiment are executed. The storage medium can be a volatile or non-volatile computer-readable storage medium.

[0118] The embodiments of the present disclosure also provide a computer program product, which includes computer instructions. When the computer instructions are executed by a processor, the steps of the drone acoustic identification method based on frequency band feature extraction described in the above method embodiment can be executed. For details, please refer to the above method embodiment, which will not be repeated here.

[0119] The computer program product may be implemented in hardware, software or a combination thereof. In one optional embodiment, the computer program product is implemented as a computer storage medium. In another optional embodiment, the computer program product is implemented as a software product, such as a software development kit (SDK).

[0120] Those skilled in the art can clearly understand that, for the convenience and simplicity of description, the specific working process of the device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here. In the several embodiments provided in the present disclosure, it should be understood that the disclosed device and method can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of the device or unit can be electrical, mechanical or other forms.

[0121] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0122] In addition, each functional unit in each embodiment of the present disclosure may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0123] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium that is executable by a processor. Based on this understanding, the technical solution of the present disclosure, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present disclosure. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0124] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present disclosure, which are used to illustrate the technical solutions of the present disclosure, rather than to limit them. The protection scope of the present disclosure is not limited thereto. Although the present disclosure is described in detail with reference to the above-described embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-described embodiments within the technical scope disclosed in the present disclosure, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present disclosure, and should be included in the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure should be based on the protection scope of the claims.

Claims

1. A method for drone acoustic recognition based on frequency band feature extraction, characterized in that: include: Collect environmental sound signals and pre-process them to generate target sound signals; Decomposing the target sound signal into eigenmode components, filtering noise components in the eigenmode components, and restoring the eigenmode components to a noise-reduced sound signal; Extracting a plurality of acoustic features corresponding to the noise reduction sound signal, wherein the acoustic features at least include frequency band filtering cepstral coefficient features, and fusing the plurality of acoustic features into a multi-dimensional feature vector; The multi-dimensional feature vector is input into a pre-trained acoustic classification and recognition model to determine the drone recognition result.

2. The method according to claim 1, characterized in that Collect environmental sound signals and pre-process them to generate target sound signals, including: The ambient sound signal is collected by using a microphone array, wherein the ambient sound signal includes at least a drone sound signal, ambient noise, and an interference signal; The environmental sound signal is subjected to preprocessing including time-frequency filtering denoising, adaptive filtering denoising, gain adjustment and data normalization to generate the target sound signal.

3. The method according to claim 1, characterized in that Decomposing the target sound signal into eigenmode components, filtering noise components in the eigenmode components, and restoring the eigenmode components to a noise-reduced sound signal specifically includes: Using an eigenmode decomposition algorithm to process the target sound signal, decomposing the target sound signal into the eigenmode components; Analyzing the intrinsic mode component by a fast independent component analysis algorithm, identifying the noise component and setting the noise component to zero; The denoised intrinsic mode components are restored by using a reconstruction matrix, and the denoised intrinsic mode components are reconstructed into the denoised sound signal by using the eigenmode decomposition algorithm.

4. The method according to claim 1, characterized in that: Extracting multiple acoustic features corresponding to the noise reduction sound signal, the acoustic features at least including band filtering cepstral coefficient features, and fusing the multiple acoustic features into a multidimensional feature vector, specifically including: Using a Mel filter to extract Mel-frequency cepstral coefficient features of the noise-reduced audio signal; A Gammatone filter is used to replace the Mel filter to extract the Gammatone spectrum features corresponding to the noise reduction sound signal; For the output of the Mel filter, Mel frequency cepstral coefficients are not extracted as Fbank features; Using a filter group of a preset frequency band, extracting the frequency band filtering cepstrum coefficient characteristics corresponding to the noise reduction sound signal; The Mel frequency cepstral coefficient feature, the Gammatone spectrum feature, the Fbank feature and the band filtering cepstral coefficient feature are fused to generate the multi-dimensional feature vector.

5. The method according to claim 1, characterized in that The acoustic classification recognition model is trained based on the following steps: Constructing the acoustic classification recognition model including a convolutional layer, a pooling layer, a fully connected layer, and a Softmax layer, and using the ReLU function as an activation function; Obtain a training sound signal data set including drone sound signals and background noise signals; Annotating the training sound signal data set using the drone sound signal and the background noise signal as labels; The training multidimensional feature vector corresponding to the training sound signal data set is extracted, the training multidimensional feature vector is input into the acoustic classification and recognition model for training, and the model parameters of the acoustic classification and recognition model are optimized using a cross entropy loss function and a gradient descent method.

6. A drone acoustic recognition device based on frequency band feature extraction, characterized in that: include: The sound signal acquisition module is used to collect environmental sound signals and pre-process them to generate target sound signals; A noise reduction module, configured to decompose the target sound signal into eigenmode components, filter noise components in the eigenmode components, and restore the eigenmode components to a noise-reduced sound signal; A feature extraction module, used to extract multiple acoustic features corresponding to the noise reduction sound signal, wherein the acoustic features at least include frequency band filtering cepstral coefficient features, and fuse the multiple acoustic features into a multi-dimensional feature vector; The recognition module is used to input the multi-dimensional feature vector into a pre-trained acoustic classification recognition model to determine the drone recognition result.

7. The device according to claim 6, characterized in that The sound signal acquisition module is specifically used for: The ambient sound signal is collected by using a microphone array, wherein the ambient sound signal includes at least a drone sound signal, ambient noise, and an interference signal; The environmental sound signal is subjected to preprocessing including time-frequency filtering denoising, adaptive filtering denoising, gain adjustment and data normalization to generate the target sound signal.

8. The device according to claim 6, characterized in that The noise reduction module is specifically used for: Using an eigenmode decomposition algorithm to process the target sound signal, decomposing the target sound signal into the eigenmode components; Analyzing the intrinsic mode component by a fast independent component analysis algorithm, identifying the noise component and setting the noise component to zero; The denoised intrinsic mode components are restored by using a reconstruction matrix, and the denoised intrinsic mode components are reconstructed into the denoised sound signal by using the eigenmode decomposition algorithm.

9. An electronic device, characterized in that: include: A processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor and the memory communicate through the bus, and when the machine-readable instructions are executed by the processor, the steps of the drone acoustic identification method based on frequency band feature extraction as described in any one of claims 1 to 6 are performed.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, executes the steps of the drone acoustic identification method based on frequency band feature extraction as described in any one of claims 1 to 6.

Images