58 results about "Cepstrum" patented technology

The invention relates to the field of biological medicine, and discloses a method and system for analyzing cough sound by using disease characteristics to diagnose respiratory diseases, and the method comprises the steps: deploying a six-microphone annular array to achieve the precise positioning and triggering of a sound source; self-adaptive spectral subtraction and Wiener filtering cascade are adopted to enhance the audio; segmenting a cough segment based on energy envelope; fusing the Mel-cepstrum, the linear prediction residual error, the harmonic energy ratio and the transient zero-crossing rate to construct a pathological feature matrix; extracting local, medium-range and global time sequence features through a three-branch parallel convolutional network; inputting a disease specific classifier to discriminate asthma, pneumonia and laryngitis respectively, and applying a feature decoupling regular term to improve interpretability. The system correspondingly realizes the modularized processing flow. According to the method, the cough sound collection quality and the disease subtype recognition accuracy in a complex environment are improved, meanwhile, the thermodynamic diagram is output to assist clinical decision making, and the diagnosis credibility and practicability are enhanced.

The invention discloses an Internet of Things anomaly detection method and system based on quaternion state space diffusion enhancement, and belongs to the technical field of network security and artificial intelligence. The method comprises the following steps: mapping a flow time sequence feature into a quaternion tensor to maintain an internal coupling relationship of a multi-dimensional feature; a double-flow encoder is designed, a quaternion selective state space model is adopted to extract continuous fluid features, and a dynamic hypergraph neural network is adopted to model discrete protocol features; carrying out self-supervised pre-training on a resistance pseudo-anomaly sample by utilizing potential diffusion model generation, and optimizing characterization by combining quaternion cepstrum distance loss; the injected learnable prompt vector is optimized in the small sample fine tuning stage, and a category prototype is corrected by using a semi-supervised expectation maximization algorithm; and calculating a sample anomaly score based on an energy model to realize known attack classification and unknown anomaly judgment. According to the method, the generalization ability of the model under the small sample condition and the unknown threat detection ability are remarkably improved.

The invention relates to the field of voice signal processing and voice recognition, and discloses a voice recognition enhancement method and system based on harmonic model fundamental frequency optimization and RNN noise suppression, and the method comprises the steps: collecting a voice signal in a noise environment in real time, carrying out the framing and windowing of the voice signal, and generating multi-dimensional time-frequency data; estimating the fundamental frequency of each frame of voice through a cepstrum analysis method, and screening effective fundamental frequency frames according to a confidence coefficient threshold to form a fundamental frequency characteristic matrix; dynamically adjusting the noise power spectrum of the Wiener filter under the drive of the fundamental frequency characteristic matrix, and carrying out dislocation fusion on the filtering output and the original spectrum to form an enhanced spectrum first draft; inputting the enhanced spectrum first draft into a recurrent neural network in a framing manner, predicting the gain of each frequency band, calculating an inhibition factor, and generating a multi-frame continuous enhanced spectrum sequence; multiple frames of continuous enhanced spectrum sequences are synthesized into voice signals through inverse short-time Fourier transform, an end-to-end enhanced recognition process is formed, and the method has the advantage of improving accuracy.

The invention discloses an earphone defect detection method and device based on voiceprint feature analysis and a storage medium, and relates to the technical field of electroacoustic quality control. According to the method, human ear hearing characteristics are simulated, Mel-cepstrum voiceprint characteristics are extracted from sound response signals, an adaptive weight mechanism based on an equal-loudness curve is introduced, and a voiceprint abnormal deviation index model conforming to human ear subjective perception is constructed; in order to solve the problem that tiny abnormal noise is difficult to recognize, high-order statistics such as kurtosis and skewness are adopted for feature fusion enhancement, and a support vector machine is combined to achieve accurate capture of transient impulse noise. According to the invention, various defect types such as voice coil friction, imbalance of balance degree, diaphragm damage and the like can be fully automatically distinguished, the problems that a traditional test means is disjointed with subjective hearing sense and the omission ratio of hidden defects is high are effectively solved, and the quality inspection efficiency and accuracy of earphone production lines are remarkably improved.

The invention discloses a fracturing-based water hammer wave velocity inversion method, a readable storage medium and a fracturing-based water hammer wave velocity inversion device, and belongs to the technical field of fracturing. Performing noise reduction processing to obtain a noise reduction signal; converting the noise reduction signal to a reverse frequency domain; establishing a relationship between the cepstrum amplitude and the liquid inlet response time, generating a two-dimensional curve graph, identifying a peak point, and correspondingly obtaining the liquid inlet response time; designing a water hammer wave velocity value, and converting to obtain a liquid inlet depth value; and solving the minimum value of the difference sum between the liquid inlet depth value and the designed depth value. The readable storage medium stores the method. The device is provided with a computer program for storing and executing the method. According to the method, the water hammer wave velocity can be effectively determined under the conditions that the bridge plug slides and the actual bridge plug position is not matched with the designed bridge plug position, the liquid inlet position of the fracturing liquid is promoted to be better matched with the designed perforation position, then fracturing liquid inlet point recognition is guided, a fracturing construction scheme is optimized, and the fracturing transformation effect is improved.

The invention relates to the technical field of rotating machinery fault monitoring, and discloses a method and system for monitoring a torsional vibration fault of a shaft system of a steam turbine generator unit, and the method comprises the steps: obtaining a single-point torsional vibration signal, and converting the single-point torsional vibration signal into an angular domain sequence; extracting a narrow-band modulation component of a specific order band and calculating a torsional vibration logarithmic response spectrum; performing inverse transformation on the logarithmic spectrum to obtain cepstrum data reflecting the periodicity of the shafting structure; delimiting a structure reflection propagation time delay interval according to the geometric distance from the sealing end to the rigid coupling and the material wave velocity; identifying a sealing-coupling boundary reflection peak value and a series structure in the interval, and constructing a series enhancement index; reversely deducing an equivalent reflection intensity and an equivalent structure distance, and generating a health index; and finally, mapping the health index into a binarization result through a step function, and outputting normal or abnormal judgment for the sealed end. According to the invention, the high-sensitivity and deterministic monitoring of the local abnormity of the sealing end is realized by using the echo structure information.

The application discloses an audio quality automatic scoring method and system combining MFCC and time domain statistical features, comprising the following steps: on the basis of resampling and time alignment of a reference signal and a to-be-detected signal, extracting cepstrum and time domain statistical features according to second segmentation and further 100 ms subsegment extraction, forming 170-dimensional second-level features through mean value+maximum value hierarchical aggregation of subsegment features, splicing the corresponding reference segment into 340-dimensional joint features, inputting the pre-trained support vector machine model to output a second-by-second discrete score of 0 to 5, and realizing high-time-resolution automatic quality evaluation and abnormal positioning under the conditions of light computing power and small samples.

The speech synthesis method, the speech synthesis device, the electronic equipment and the storage medium provided by the embodiment of the present application relate to the technical field of financial technology. The method comprises the following steps: performing spectrum feature extraction on original speech data to obtain original speech mel-frequency cepstrum and original speech linear spectrum; performing feature coding on the original speech mel-frequency cepstrum through a preset speaker encoder to obtain original speaker features; performing posterior coding on the original speech linear spectrum through a preset posterior encoder to obtain speech posterior features; performing content feature extraction according to the original speaker features and the speech posterior features to obtain target speech content features; and performing speech synthesis according to preset target speaker features and the target speech content features to obtain synthesized speech data; wherein the synthesized speech data is used to represent speech uttered by a target speaker object according to original speech content. The embodiment of the present application can guarantee the quality of speech synthesis and can perform speech conversion for any speaker.

The present application relates to a kind of data-independent anti-multipath radio frequency fingerprint extraction method and system, for extracting the radio frequency fingerprint of wireless device.The method implementation steps include: sampling and preprocessing to wireless signal, obtain received signal;The time-domain signal of received signal is transformed in frequency domain and logarithm operation to obtain cepstrum domain signal;Select two frames of signals experiencing channel incoherence to do cross-correlation operation and further remove the correlation caused by local signal, as radio frequency fingerprint expression;The cross-correlation value between multiple symbols is superimposed and averaged to enhance fingerprint, while eliminating noise.The method of the present application effectively decouples the radio frequency fingerprint component in wireless signal from wireless multipath channel component, varying transmission data component, while enhancing the representation of radio frequency fingerprint, improving the stability of fingerprint, and suppressing noise, which is an effective data-independent anti-multipath radio frequency fingerprint extraction method for any wireless transmission signal.

The application belongs to the technical field of speech recognition and semantic understanding, and discloses a high-precision speech recognition and semantic understanding method. The method comprises the following steps: collecting a far-field multi-sound-source original speech signal to generate a discrete speech sampling sequence; performing frequency domain transformation on the original speech signal to extract an amplitude-frequency distortion quantization parameter; segmenting the original speech signal to generate a speech frame sequence and calculating a non-steady-state speech speed quantization parameter; extracting an original mel-frequency cepstrum feature and performing distortion correction to obtain a weighted acoustic feature through dynamic weighting; constructing a preset semantic feature library and mapping to generate an initial semantic matching vector; and recursively iterating to calibrate the semantic confidence and screening an optimal term to output a result. The application solves the problem of low far-field speech recognition accuracy in the prior art, realizes high-precision speech recognition and semantic understanding through multi-link collaborative optimization, and improves recognition stability.

The invention relates to the technical field of speech processing, and particularly provides a disguise speech recognition method based on a Transform model and mixed features, which is carried out from two aspects of feature extraction and model establishment. A resonance peak parameter is calculated by adopting a cepstrum method, a cepstrum coefficient (GFCC) is obtained through a Gammatone filter bank, then the resonance peak, the GFCC and a difference coefficient of the GFCC are combined into a mixed characteristic parameter, and complementary correlation between mixed characteristics is mined. From the perspective of model establishment, the mixed features are used as the input of the model, and the Transform network model is used as the acoustic model of the voiceprint recognition system, so that the feature distribution is better fitted, the classification effect is remarkably improved, and the performance of the camouflage voice voiceprint recognition system is effectively improved. The problem of performance degradation caused by feature redundancy and modal noise in a traditional method is solved.

The application discloses an Internet of Things anomaly detection method and system based on quaternion state space diffusion enhancement, and belongs to the technical field of network security and artificial intelligence. The method comprises the following steps: mapping traffic time sequence features into a quaternion tensor to maintain the internal coupling relationship of multi-dimensional features; designing a double-flow encoder, extracting continuous flow features by using a quaternion selective state space model, and modeling discrete protocol features by using a dynamic hypergraph neural network; generating an adversarial pseudo-anomaly sample by using a latent diffusion model to perform self-supervised pre-training, and combining a quaternion cepstrum distance loss to optimize the representation; optimizing the injected learnable prompt vector in the small sample fine-tuning stage, and correcting the class prototype by using a semi-supervised expectation maximization algorithm; calculating sample anomaly scores based on an energy model to realize known attack classification and unknown anomaly determination. The application significantly improves the generalization ability of the model under the condition of small samples and the detection ability of unknown threats.

The application relates to the field of artificial intelligence and discloses an audio generation method, device and equipment based on artificial intelligence and a storage medium, the method comprising the following steps: obtaining to-be-converted audio data, and obtaining a target field identifier corresponding to expected sound quality and / or expected emotion; performing cepstrum feature conversion based on a speech spectrum on the to-be-converted audio data to obtain cepstrum feature data of the to-be-converted audio data; inputting the cepstrum feature data and the target field identifier into a speech conversion model to encode the cepstrum feature data at an encoding layer, randomly sample the encoded vector at a resampling layer, and decode and reconstruct the vector sampled at a decoding layer based on the target field identifier to obtain target audio data; and on the basis of effectively removing sound rhythm information of a sounder, the application effectively converts the audio data into audio data with expected rhythm style information, thereby improving an audio conversion effect.

This invention belongs to the field of oil and gas extraction technology and discloses a method for detecting the effect of temporary plugging of fracture openings in wellbore, comprising: S1, inducing water hammer phenomenon in the fracturing fluid in the wellbore; S2, detecting the real-time pressure signal value P(t) in the wellbore; S3, analyzing the real-time pressure signal value P(t) and plotting a cepstrum; S4, determining whether there are paired peaks based on the cepstrum; if not, stopping; if so, determining the water hammer period between the two peaks in each pair and proceeding to the next step; S5, calculating the average wave velocity C of the fracturing fluid when the water hammer phenomenon occurs in the wellbore, and multiplying the average wave velocity C by the water hammer period to obtain the specific location of the fracture opening. This method for detecting the effect of temporary plugging of fracture openings in wellbore improves the efficiency and accuracy of construction, achieves real-time feedback, and has low detection cost.

A keyword spotting method and apparatus based on magnetic tunnel junction arrays. The method comprises: using a statistically-aware training method to train a neural network, so as to obtain weight data of the neural network, mapping the weight data to high and low resistance state data of magnetic tunnel junction arrays, on the basis of an actual network structure of the neural network, selecting a magnetic tunnel junction array of a corresponding size, and pre-programming magnetic tunnel junctions into corresponding resistance states (S101); extracting a Mel-frequency cepstral coefficient (MFCC) feature from a speech keyword signal (S102); and during performing keyword spotting, continuously inputting binarized speech MFCC features into the magnetic tunnel junction array in the form of voltage amplitudes, acquiring a multiply-accumulate operation result by means of measuring an output current of a target column of the magnetic tunnel junction array, and finally outputting a hardware-based keyword spotting result after performing normalization and layer-by-layer inference (S103).

The invention relates to an unmanned aerial vehicle radar detection and identification method and system based on background contrast attention and double-flow micro-Doppler fusion, and belongs to the technical field of radar signal processing and intelligent target identification. According to the method, the problems of low unmanned aerial vehicle detection rate, high false alarm rate, and poor recognition robustness caused by observation angle change and sample scarcity in a low signal-to-noise ratio environment are solved. According to the technical scheme, the method comprises the steps of performing pulse compression on a radar signal to construct a distance Doppler map, realizing rapid positioning of a target by using a full convolution detection network integrated with background contrast attention, generating a time Doppler map and a time cepstrum map, and performing few-sample learning recognition through a double-flow feature fusion network based on Transform. According to the method, the detection precision can be remarkably improved, the recognition robustness is enhanced, the real-time efficiency is optimized, and the method adapts to a few-sample condition.

The application discloses a kind of electric fish signal detection methods based on mel cepstrum and multilayer perception machine, belong to electric fish signal detection field, comprising: first, the sound emission signal generated by underwater discharge is collected, pre-processing is carried out by pre-emphasis, framing and windowing to improve signal-to-noise ratio;Subsequently, the mel frequency cepstrum coefficient feature of signal is extracted, and is optimized by first-order and second-order difference calculation, and is fused to form a feature parameter set with static and dynamic characteristics;Finally, the feature parameter and detection voltage value are input into multilayer perception machine neural network for nonlinear fitting, and the high-precision electric signal strength is reconstructed.The application effectively utilizes the sound emission signal feature, and significantly improves the sensitivity and confidence of underwater electric fish signal detection.

The application relates to the technical field of building health monitoring, and discloses a building health state monitoring method based on vibration signals. The method collects vibration data of different monitoring points of a building, and obtains external influence data such as meteorological data, traffic data, underground water level data, construction activity data and earthquake activity data; after uniform time calibration and preprocessing of the above data, natural frequency, vibration energy, cepstrum, three-dimensional vibration trajectory, ground reflection wave, elastic wave propagation and external influence characteristics are extracted; the extracted characteristics are input into a building evaluation model to obtain health state evaluation results including a health index, an abnormal type and a risk level; further combined with evaluation results and spatial position relationships of buildings in a target region, a group abnormal linkage relationship is identified, regional synchronous abnormality, single-body abnormality or conduction abnormality is determined, and regional safety evaluation results and risk early warning information are generated. The method improves the accuracy and early warning timeliness of building and building group health monitoring.

The present application belongs to the technical field of radio spectrum monitoring, and provides a wide-frequency unmanned aerial vehicle spectrum detection and rapid identification method based on deep learning, which comprises the following steps: collecting wide-frequency radio signals to form a composite signal sequence and converting the composite signal sequence into a time-frequency feature map, performing cepstrum transformation on the original signal to extract a cepstrum time delay peak, constructing an environment compensation parameter vector, identifying frequency selective fading notches in the time-frequency feature map, dividing to form a spectrum fragment set, constructing a fragment association model in combination with a frequency domain spacing and an inverse frequency position matching rule, and completing physical constraints, inputting the spectrum fragment spatial features and the environment compensation parameter vector into the model, calculating a coherence weight through a deep fragment association model based on a Transform architecture, completing logical reorganization of the spectrum fragments in a vector space and generating a logical reorganization vector, inputting the vector into a double-branch parallel classification network, and realizing accurate identification of unmanned aerial vehicle target models and communication protocols.

The present application relates to the technical field of deep learning voiceprint recognition, and particularly relates to a voiceprint recognition method based on a double-feature branch structure, which comprises extracting Mel cepstrum features and wavelet transform features from original speech signals respectively and forming two branches; inputting the features into a self-attention network and a convolution TDNN network respectively for multi-scale feature modeling, and fusing the two outputs; further calculating a multi-level discriminant loss for the fused voiceprint representation to enhance the speaker distinguishability in a noisy or mismatched environment; performing decoding or upsampling operation on the fused output and taking the output as the input of the next level processing, and finally generating multi-resolution and more robust voiceprint features through a cascaded encoder-decoder structure. The method aims to overcome the shortcomings of single-path feature extraction in complex environments, and significantly improves the capture and recognition ability of multi-resolution speech features by combining the advantages of self-attention and convolution TDNN.

This invention relates to a method for separating vibration and flow noise in the underwater noise spectrum of a ship propeller, comprising: constructing a "frequency label" for vibration noise by fabricating two propeller models with identical geometry but significantly different material properties; a dual-model difference stage: calculating the absolute difference in the noise spectra of the two models and generating a vibration noise mask based on modal analysis results; a blind source separation stage: for the residual signal after difference, using an improved Independent Component Analysis (ICA) algorithm to further separate residual interference by utilizing the statistical independence of vibration noise; introducing a convolution kernel function in this stage to enhance the algorithm's adaptability to time-delayed signals; and introducing Cepstrum analysis technology to perform phase correction on the complex spectrum of the replacement frequency band. This method solves the significant shortcomings of existing noise control technologies in the coupling separation of vibration and flow noise, enabling the separation of the components of underwater noise from ship propellers, obtaining the noise components caused by blade vibration and the flow noise components separately.

This document describes various techniques for unsupervised thickness measurement and corrosion estimation in materials using ultrasonic inspection systems that utilize non-destructive testing (NDT) methods. The system uses cepstrum analysis to analyze the spectral power of the acoustic data signal acquired after the ultrasonic probe assembly emits ultrasound, identifying different frequency contents associated with multiple echoes. Cepstrum analysis allows the system to distinguish between different types of echoes, such as echoes from the front and back walls of the material, as well as echoes from any defects present within the material.

The application is based on a laser shock peening quality online monitoring method of acoustic emission rebalanced Mel cepstrum, and realizes surface hardness quality monitoring of LSP process. Aiming at the problems of data-driven monitoring method based on acoustic emission technology, such as wide AE high sampling rate frequency band, weak LSP subsurface plastic deformation characteristics, and difficult LSP process monitoring, an LSP plasma acoustic emission signal acquisition system is built. After signal acquisition, the signal is indirectly represented by material surface hardness, the signal key frequency band boundary frequency is obtained by power spectral density analysis, and based on the rebalanced Mel cepstrum method, the segmented function of feature enhancement and weakening of the signal key and non-key frequency bands is adaptively generated. Then the feature mapping is obtained to improve the Mel time-frequency diagram. Finally, the local and global features of the signal image are fused in parallel through the model, so as to realize the acoustic emission monitoring of the surface hardness quality of the LSP process. The method proposed in the application is simple and efficient, has high feature discrimination, and has strong engineering applicability.

The present application relates to underwater acoustic engineering, sonar signal processing and harmonic detection technical field, especially a kind of jammer equipment identification method based on retransmission harmonic feature, first the active sonar echo signal obtained is carried out matching filtering and sampling, then using cepstrum algorithm to extract its corresponding cepstrum, finally the obtained cepstrum is detected without direct current component, whether the target is jammer equipment is judged by the ratio of non-direct current component and noise energy.The present application utilizes the retransmission mechanism of underwater decoy target, exposes its harmonic signal characteristics through matching filtering, and on this basis, relies on harmonic signal identification means to identify and eliminate it.The present application effectively improves the identification ability of jammer equipment, and the algorithm process is simple, the calculation complexity is low, the portability is strong, provides an important component module for constructing the detection and identification system of underwater target.

The present application belongs to the technical field of oil and gas exploration seismic data interpretation. The present application discloses a method and device for identifying faults based on differential cepstrum. The method performs differential cepstrum processing on seismic data of a target layer section, extracts first-order differential cepstrum coefficients of the differential cepstrum, sets a processing factor to calculate eigenvalues of a covariance matrix of the first-order coefficient data body of the differential cepstrum, and generates a high-resolution fault information data body. The differential cepstrum operation adopted by the method strengthens the weak information singularity feature, can identify small faults, and is beneficial to improving the accuracy and precision of fault identification of seismic data. The device for identifying faults based on differential cepstrum provided by the present application comprises a Fourier transform spectrum generator, a logarithmic spectrum generator, a Fourier inverse transform processor, a differential cepstrum first-order coefficient data body generator, and a fault data generator. The device realizes the method for identifying faults based on differential cepstrum.

The invention relates to the technical field of rotary mechanical vibration signal processing and fault diagnosis, and discloses a coal mine underground mechanical excitation identification method based on self-adaptive cepstrum exponential window filtering, which comprises the following steps: S1, preprocessing an original vibration signal; s2, constructing a band-pass filter to filter the preprocessed vibration signal; s3, performing equal-length segmentation on the vibration signal after band-pass filtering, and calculating a linear kurtosis value of each sub-segment; s4, obtaining an angle domain stable vibration signal; s5, adaptively determining a characteristic cepstrum boundary according to the cepstrum energy distribution; and S6, performing inversion on the cepstrum signal after cepstrum filtering to obtain a time domain forced excitation signal. The index window damping factor is automatically set by referring to the analytic relationship between the cepstrum delay and the suppression ratio, the limitation that a traditional cepstrum index window depends on empirical parameter estimation is overcome, cepstrum editing can adapt to different structures and working conditions, and the stability and effectiveness of system transfer characteristic suppression are guaranteed.

The invention provides a voice voiceprint frequency division acquisition and detail extraction method and system. The method comprises the following steps: dividing a voice signal into a plurality of frequency bands for acquisition according to a barker scale by using a voice signal frequency band division mode based on the barker scale so as to realize frequency division processing on the voice signal; for each divided frequency band, multi-frequency-band voiceprint detail feature values are extracted, and the multi-frequency-band voiceprint detail feature values comprise the time domain feature, the frequency domain feature and the cepstrum feature of the voice signal of each frequency band; collecting voice samples corresponding to different human physiological states, and constructing a physiological voiceprint reference data group according to different physiological states; the method comprises the following steps: comparing a feature value obtained by performing frequency division acquisition and feature extraction on a to-be-detected voice signal with a physiological voiceprint reference data group, calculating a difference feature, and clustering the difference feature by using a preset clustering algorithm to complete physiological group division; according to the physiological group division result, a physiological voiceprint block map is drawn, and voice-based human physiological state or disease auxiliary diagnosis is achieved.

The present application relates to a method for detecting lung disease by cough sound, which specifically comprises: removing non-cough sound segments from original cough audio to obtain cough audio, and obtaining the start and end time of each single cough segment in the cough audio; generating a logarithmic mel-cepstrum matrix corresponding to each single cough segment according to the start and end time of each single cough segment, and calculating a position encoding matrix corresponding to each single cough segment; obtaining a feature matrix of the cough audio according to the logarithmic mel-cepstrum matrix and the position encoding matrix corresponding to each single cough segment; multiplying the normalized feature matrix of the cough audio by a proportion factor, adding the obtained product to the feature matrix of the cough audio, and inputting the result into a classification network for classification. The present application also relates to a device comprising a cough sound detection unit, an audio signal processing unit, a splicing unit, a normalization unit and a classification network unit. The method and device of the present application can improve the accuracy of the detection result.

The application provides a rolling bearing compound fault diagnosis method based on enhanced harmonic vector analysis, vibration signals of rolling bearings in different fault and normal states are acquired in advance as source signals, and are convolved and mixed; the mixed signals are denoised by a wavelet threshold value method, and the denoised mixed signals are subjected to harmonic structure enhancement by using a cepstrum threshold value method; a two-dimensional time-frequency mask function similar to Wiener is constructed, the mixed signals subjected to the harmonic structure enhancement are subjected to blind source separation in a time-frequency domain, and the frequency domain scale of the separated signals is aligned by using reverse projection; for each separated signal, a fast spectral kurtosis method based on a 1 / 3 binary tree structure is used to calculate the kurtosis diagram of the signal, the frequency band with the maximum kurtosis value is found, and band-pass filtering is performed on the frequency band; the filtered signals are subjected to Hilbert envelope spectrum analysis, fault characteristics are obtained, and compound fault diagnosis is completed. The application improves the diagnosis precision of rolling bearing compound faults under strong noise.

A collision moment detection method based on MFCC feature extraction and a CNN classification algorithm belongs to the technical field of signal processing, and comprises the following steps: an audio data acquisition and processing module acquires audio data of a target environment in real time; the audio data preprocessing module carries out preprocessing operation on the audio data to obtain standard audio data; an MFCC feature extraction module sequentially performs transformation, filtering and cepstrum analysis operation on the standard audio data to obtain 39-dimensional feature data; a CNN classification and recognition module performs classification and recognition operation on the 39-dimensional feature data to obtain an impact moment prediction value; and the impact moment detection output module compares the impact moment predicted value with a set impact threshold value, when the impact moment predicted value is larger than the set impact threshold value, it is indicated that the impact moment is detected, and otherwise, it is indicated that no impact event is detected. The method has the advantages that interference of noise on the impact moment recognition effect can be reduced, and the impact moment detection precision is improved.