43 results about "Butterworth filter" patented technology

The invention provides a dance video generation method based on multi-mode music driving and frequency domain-space double-flow decomposition, and the method comprises the steps: extracting multi-granularity music features through a composite encoder (Librosa + Jukebox), employing a beat gating attention mechanism, enabling the key actions such as dancing hand raising, kicking and the like to be strictly aligned with a music re-beat point, and enabling the synchronization error to be reduced to 118 ms through the verification of a test data set; for the problem of visual detail loss, a frequency domain-space double-flow decomposition architecture is provided, a Butterworth filter bank is used to decouple a reference image into a low-frequency energy diagram and a high-frequency residual error, and a double-flow diffusion mechanism is used to optimize a global attitude and local details respectively; a joint confidence prediction module is introduced for the generation stability in a shielding scene, and the motion trail of an abnormal joint point is dynamically corrected through a time domain sliding window weighted fusion strategy, so that a reasonable action conforming to ergonomics can still be generated under a 50% limb shielding rate.

The invention relates to the technical field of medical signal processing, in particular to a BCG signal processing method combining TimeGAN and a Butterworth filter, which comprises the following steps: carrying out preliminary noise reduction on an input BCG signal by adopting a Butterworth band-pass filter to obtain a filtered signal; performing phase compensation on the filtered signal by adopting a bidirectional filtering mode to obtain a signal without phase distortion; the trained TimeGAN model is adopted to carry out signal optimization on the signal without phase distortion, and an optimized signal is obtained; the TimeGAN model comprises a convolutional neural network which is used for inputting data in the TimeGAN model to carry out feature extraction so as to obtain corresponding local features; the long short-term memory network is used for carrying out time sequence modeling on the local features to obtain corresponding time sequence features; the multi-head attention mechanism is used for calculating weighted features based on the time sequence features; and the generator is used for generating an optimized signal according to the weighted characteristics.

The invention discloses a heart tension index dynamic calculation and risk assessment method based on HRV, and the method comprises the steps: carrying out the filtering and denoising of an electrocardiosignal through a Butterworth filter, and obtaining a filtered electrocardiosignal; calculating interval data of adjacent wave crests; processing the interval data according to fast Fourier transform to obtain frequency domain features, and combining the interval data to obtain time domain features; performing normalization processing on the time domain and frequency domain features based on a maximum and minimum normalization method; calculating a cardiac tension index according to the normalized features and a pre-trained multiple linear regression model; matching a pre-constructed reference database according to the individual age, gender and body mass index to obtain a reference cardiac tension index; calculating a cardiovascular risk score in combination with the normalized frequency domain features and a reference cardiac tension index; and repeating the above steps regularly, smoothing through a moving average method based on continuous multiple cardiovascular risk scores, and analyzing the trend to send out cardiovascular risk early warning information.

The invention relates to the technical field of laser radar detection, in particular to a frequency-modulated continuous wave laser radar high-precision distance measurement method, which comprises the following steps of: 1, firstly, performing noise suppression by using an improved FTRLS (Fast Transverse Recursive Least Squares) adaptive filtering algorithm aiming at a low signal-to-noise ratio signal; 2, adjusting the added limitation and filter parameters, iterating the optimal parameters, and applying the optimal effect to subsequent spectrum correction; 3, windowing the denoised signal, and carrying out FFT (Fast Fourier Transform) to obtain a frequency spectrum of the signal; and 4, smoothing the signal by using a Butterworth filter, suppressing a side lobe of a frequency spectrum due to the characteristics of the filter, and highlighting the energy of a main lobe of the signal and the position of a spectrum peak. According to the method, a redundancy feedback-dynamic restart mechanism is applied to an adaptive filtering algorithm RLS, the FTRLS algorithm which is added with a transverse filter and leads to reduction of denoising precision is greatly improved, numerical divergence of the FTRLS algorithm is inhibited by using a regularization factor, and the signal-to-noise ratio is improved.

The invention discloses a knee joint acoustic signal analysis and inflammation detection method based on deep learning, which is based on a knee joint acoustic signal acquisition system, acquires acoustic signals generated in the motion process of a knee joint in real time, and inputs the acoustic signals into a deep learning model for analysis. Acoustic signals generated when the knee joint moves are collected; the collected knee joint acoustic signals are preprocessed through a Butterworth filter, short-time energy segmentation and Mel spectrum feature extraction method; a deep learning model based on Transformer is used for training the extracted features, a complex mode in the knee joint acoustic signal is learned, whether inflammation exists or not is judged through automatic classification, and a diagnosis result is output; real-time monitoring data is transmitted to a cloud end through wireless communication, and the health state of the knee joint is checked; according to the scheme, features are automatically extracted through deep learning, and high-accuracy recognition of the inflammation state is achieved.

The invention relates to the technical field of power transmission line load regulation and control, in particular to a power transmission line dynamic load regulation and control method suitable for a high-voltage power transmission system. Determining the fluctuation coefficient of the current load of the power transmission line at each moment; obtaining the frequency of the oil field high-voltage power transmission system at each moment, and taking the frequency as the center frequency of a Butterworth filter at each moment; determining the order of a Butterworth filter at each moment by using the numerical value of the fluctuation coefficient, and optimizing the Butterworth filter in combination with the center frequency and the order; and carrying out harmonic extraction on the current signal at each moment by using the optimized Butterworth filter, and carrying out current compensation based on a harmonic component. The precision of power transmission line load regulation and control is improved.

The present invention discloses a simplified multi-joint manipulator friction parameter identification method based on the Stribeck model. The method process is as follows: the multi-joint manipulator system is split into multiple single-joint subsystems; a friction force model of each subsystem is constructed; for each subsystem, the driver is used to drive the subsystem to sample its angle, angular velocity, and torque; the sampled angular velocity is filtered, and the angular acceleration is obtained using a Butterworth filter; the angle, angular velocity, torque, and angular acceleration are substituted into the model of the single-joint subsystem to calculate the friction torque, and the friction parameters are identified based on the least squares method; the obtained friction parameters are used as the friction model parameters of the multi-joint manipulator system. The simplified multi-joint manipulator friction parameter identification method based on the Stribeck model proposed in the present invention can reduce the computational complexity of the multi-joint manipulator system friction parameter identification and can provide accurate modeling of the manipulator friction.

The invention relates to a time-frequency fault diagnosis method based on frequency band filtering. The method comprises the following steps: (1) acquiring a time-domain sound signal of electromechanical equipment through a sensor; (2) carrying out fast spectral kurtosis analysis on the time domain signal, determining the center frequency and bandwidth of an optimal band-pass filter, and carrying out frequency band filtering by using a Butterworth filter; (3) dividing the filtered signal into equal-length fragments, and generating a time-frequency graph through short-time Fourier transform (STFT); (4) constructing a two-path parallel convolutional neural network (CNN), processing the filtered one-dimensional time sequence signal in the first path, processing the two-dimensional time-frequency graph in the second path, extracting features respectively, and performing feature fusion through flattening and splicing; and (5) inputting the fused features into a Softmax classifier, and outputting a fault classification result. The accuracy rate of the method in the measured data set reaches 97.92%, and the method is obviously superior to a traditional method and is suitable for intelligent operation and maintenance of industrial equipment.

This invention provides a method for suppressing cogging torque in a permanent magnet synchronous motor based on FxLMS, ​​comprising the following steps: The unfiltered torque demand waveform output from the PI speed loop is obtained; the torque demand is processed by a static low-pass second-order Butterworth filter to obtain a filtered torque demand waveform; the torque demand is simultaneously processed by a static high-pass second-order Butterworth filter to obtain a waveform containing only the cogging torque component, forming a noise reference signal; this signal is input into a modified FxLMS filter to obtain an anti-noise signal; this signal is input to a summing module and superimposed with the filtered torque demand waveform to generate a combined torque demand for cogging torque cancellation; the combined torque demand is input to a gain module to generate a q-axis current reference value; this reference value is input to the PI current loop; the combined torque demand is simultaneously fed back to the PI speed loop and combined with the actual speed error of the motor to generate a new torque demand value; the system then returns to the beginning, forming a closed-loop operation.

A faceplate for a tap housing is disclosed, designed to enhance the operating frequency of a hardline tap by modifying the equivalent circuit configuration. The faceplate includes a plurality of taps electrically coupled to the input and output ports of the housing, along with a first finger and a second finger that respectively contact a first KS pin in the input port and a second KS pin in the output port when the faceplate is installed. These fingers introduce capacitive elements to the KS pins, converting the tap housing's equivalent circuit from a 2-pole Butterworth filter to a higher-order Butterworth filter, such as a 4-pole configuration, thereby extending frequency performance.. The invention also includes a hardline tap incorporating the faceplate and a method for upgrading existing taps by replacing the faceplate.

The present invention relates to the technical field of charging control, and in particular to a modular high-power charging pile control system, method, device and medium. The control system includes: a plurality of control modules, and each control module includes: a mean low-pass filtering unit, a Butterworth filtering unit, a PI control unit, and a modulation unit. In the present invention, a mean low-pass filter and a Butterworth filter are connected in series. The series combination of the two forms wide-band noise suppression through complementary frequency domain characteristics. By connecting the mean low-pass filter in series, the anti-interference ability of the control system is improved, and at the same time, the dynamic performance of the control system is improved. Further, sliding mode control is used to perform voltage equalization control on the input side of the power unit, so as to achieve the goal of power balance of the output power of the power unit. Thereby, the problem that the influence of interference brought by the sampling signal on the control system in the prior art easily leads to ineffective balance control between internal modules of a modular high-power DC-DC charging pile is solved.

The invention discloses a transformer oil extraction robot positioning method and system based on multi-source data fusion, and the method comprises the steps: obtaining IMU, optical flow sensor and GPS data, carrying out the second-order Butterworth filtering preprocessing, and automatically selecting a fusion algorithm according to a scene; inclination interference is compensated through weight factor adjustment and an error feedback closed loop; gPS and IMU data are fused by adopting extended state Kalman filtering outdoors, and electromagnetic interference is suppressed by introducing accelerometer zero offset compensation and dynamic noise covariance adaptation. According to the invention, the problems of insufficient positioning precision, weak anti-interference capability and scene switching fault of a robot in a complex environment of a converter station are effectively solved, high-precision continuous positioning of indoor positioning errors is realized, and the operation requirement of accurate butt joint of the oil taking port of the transformer is met.

The utility model discloses a high-precision anti-interference rotary transformer decoding circuit. The decoding circuit is composed of a second-order Butterworth filter excitation circuit and a third-order Butterworth filter receiving circuit. The excitation circuit completes the function of a low-pass filter by selecting an adaptive gain and wide-product operational amplifier and calculating the resistance gain and the common-mode voltage of the amplifier, and meets the output requirement of an excitation signal at the same time. And the receiving circuit configures a corresponding input resistor and an operational amplifier by calculating a single-side on-state amplification factor and a cut-off frequency of the circuit so as to adapt to the use requirement of a rotary transformer decoding chip and complete low-noise receiving of a rotary transformer signal. The circuit is simple in structure, high in precision, strong in decoding capability, strong in inhibition capability for interference signals, and high in electromagnetic compatibility and reliability.

The application relates to the field of avionics, and discloses a lifting speed filtering method and system based on a second-order Butterworth filter, which comprises the following steps: collecting an atmospheric pressure signal of an airplane in real time, calculating an air pressure height according to the atmospheric pressure signal, calculating a lifting speed signal according to the air pressure height, performing amplitude-frequency characteristic analysis on the lifting speed signal to determine a cutoff frequency of the second-order Butterworth filter, dynamically adjusting the cutoff frequency of the second-order Butterworth filter according to the flight state of the airplane, inputting the lifting speed signal into the second-order Butterworth filter with the adjusted cutoff frequency, and obtaining a filtered lifting speed signal. The application realizes scientific design of filter parameters by performing amplitude-frequency characteristic analysis on the lifting speed signal and obtaining a digital filter structure through denormalization processing and bilinear transformation, determines the cutoff frequency based on the physical characteristics of the signal, and avoids the disadvantages of selecting parameters by experience in the traditional method.

The invention discloses a construction method of a Verilog HDL (Hardware Description Language) code model and a related device. Firstly, a basic Butterworth filter and a VCode template are created in GCKontrol, and an IP core equivalent to an addition and subtraction operation unit in the filter is designed in Vivado based on service requirements. Then configuring a VCode template port to be consistent with the addition and subtraction operation unit to generate a VCode module, after the addition and subtraction operation unit is replaced, enabling a GCKontrol code generation directory to point to a Vivado path, and finally instantiating an IP core to the VCode module to complete model generation. According to the method, GCKontrol and IP core multiplexing are combined, automatic standardized code generation is achieved, manual work and skill dependence are reduced, efficiency and interface consistency are improved, and errors and development cost risks are reduced.

This invention relates to a method for measuring the rotational speed of a fully enclosed compressor using a multi-signal processing technique, comprising the following steps: establishing a current signal speed measurement model; acquiring the current signal, which originates from the current signal of the fully enclosed compressor; preprocessing the current signal, which involves first removing the DC component of the signal, then passing it through a third-order low-pass Butterworth filter, followed by extracting the rotational speed signal using a HILBERT transform; setting a refined frequency band interval, then performing wavelet transform for spectral refinement, and then performing spectral correction using a ratio correction method; then using a Kalman filter to optimally estimate the rotational frequency of the fully enclosed compressor after correction, in order to smooth the rotational frequency of the fully enclosed compressor, and finally obtaining the rotational speed value and displaying it on a host computer; using the above method, the problems of low measurement accuracy and high computational resource consumption of fully enclosed compressor rotational speed can be solved, thereby improving measurement accuracy and computational speed.

The invention relates to the technical field of safety monitoring, and discloses a six-degree-of-freedom safety monitoring system, which comprises the following modules: a hardware acquisition module; the system comprises a high-precision displacement meter, an IMU (Inertial Measurement Unit) which is installed at the same point with the high-precision displacement meter, and an environment sensor for monitoring temperature and humidity, the edge calculation and data fusion module adopts an embedded data acquisition terminal and acquires and fuses the data acquired by the hardware acquisition module; and the cloud processing module uploads the monitoring data to a cloud server in real time through a 4G / 5G wireless network to form a real-time data stream. By combining a high-precision sensor and a real-time data acquisition technology, high-precision and real-time monitoring of structural health is realized. The system reduces data noise through Kalman filtering and Butterworth filtering, and can carry out temperature compensation according to the environment temperature to ensure the measurement precision. The multi-sensor data dynamic fusion algorithm automatically adjusts the sensor weight, and the data accuracy and the system stability are improved.

The invention provides a heavy-load drilling machine lifting control method for optimizing a sliding mode by combining a filtering self-adaptive chaos frost ice algorithm. The heavy-load drilling machine lifting control method comprises the steps that S1, an acceleration sensor is used for obtaining the drilling machine lifting motion acceleration adistob (t); s2, designing a Butterworth filter to filter the movement acceleration adistob (t) signal obtained in the step S1 to obtain a gas (t) signal, so as to filter out external unknown nonlinear disturbance and weaken the influence of buffeting of the sliding mode controller on the system; s3, a sliding mode controller about the acceleration error e and the sliding mode surface is designed to achieve accurate control over the lifting acceleration of the drilling machine; and S4, optimizing a sliding mode controller (SMC) by using a self-adaptive chaotic frost ice algorithm to improve the performance of the controller. According to the invention, self-adaptive stable control of lifting of the heavy-load drilling machine can be realized, so that the problem that the lifting acceleration of a drilling machine lifting system is unstable due to the influence of heavy load and external unknown nonlinear disturbance is solved.

The invention discloses a muscle fatigue intelligent detection method and system based on multi-signal acquisition, and relates to the technical field of muscle fatigue detection, and the method comprises the following steps: S1, synchronously acquiring a surface electromyogram signal and a multi-channel pressure signal; step S2, carrying out preprocessing including dynamic Butterworth filtering and wavelet soft threshold denoising on the acquired signals; s3, performing continuous wavelet transform on the preprocessed signal, and extracting frequency characteristics; s4, dynamically distributing fusion weight according to the real-time signal-to-noise ratio of each channel; s5, fusing the weighted feature information based on a Bayesian inference model, and outputting a muscle fatigue state judgment result; through efficient filtering, accurate feature extraction and effective multi-information fusion, the accuracy and reliability of muscle fatigue detection are improved by using a Bayesian reasoning method.

The present invention discloses a blood oxygen saturation estimation method and system based on PPG signals, which relate to the technical field of blood oxygen saturation estimation and include the following steps: S1: designing a bandpass filter based on a Butterworth filter, designing a comb filter based on the characteristics of motion artifacts in PPG signals, and removing motion artifacts in the PPG signals by using the bandpass filter and the comb filter; S2: performing preliminary screening of PGG signal features by a threshold filtering method based on a correlation coefficient; S3: performing wrapping method feature screening from three dimensions of time domain, frequency domain and statistics by a recursive feature elimination method, and extracting a PGG signal feature set; S4: estimating the PGG signal feature set by an integrated regressor based on a random forest algorithm; and being able to extract blood oxygen saturation information from human PPG signals and realize real-time blood oxygen saturation monitoring.

The invention relates to a signal-to-noise ratio improvement method for long-distance optical fiber vibration detection, and relates to the technical field of long-distance optical fiber vibration signal-to-noise ratio improvement, and the method comprises the steps: obtaining differential data corresponding to each position and each time point of an optical fiber through distributed sound wave sensing equipment; performing integral conversion on the differential data of each position according to a time sequence to obtain vibration data, removing a direct current component through a Butterworth filter, and finally obtaining the vibration data of all positions of the optical fiber; forming a two-dimensional vibration waterfall image pixel matrix; utilizing two-dimensional variational mode decomposition to obtain mode components of different center frequencies; performing preliminary screening on all the obtained modal components according to the center frequency, and removing high-frequency modal components to obtain low-frequency modal components; and performing fast non-local mean denoising processing on each low-frequency mode component based on a logarithm-Cauchy adaptive weight kernel function, and reconstructing the processed mode components to finally obtain a long-distance optical fiber vibration signal with a remarkably improved signal-to-noise ratio.

The application discloses a high-order configurable current mode low-pass filter circuit, and belongs to the technical field of low-pass filters. The high-order configurable current mode low-pass filter circuit comprises a first-stage low-pass filter, a second-stage low-pass filter and a third-stage low-pass filter which are connected in sequence, the output current of the low-pass filter at a higher stage is poured into the input end of the low-pass filter at a lower stage through a current mirror between two adjacent low-pass filters; and a second-order current mode gm-C structure is adopted for each low-pass filter; a super source follower is arranged on one side of the first-stage low-pass filter, the output end of the super source follower is connected to the input end of the first-stage low-pass filter through two resistors, and a low-resistance connection to the ground is established. The gm-C filter of the three-stage current mode is utilized to realize a 6-order Butterworth filter, the bandwidth can be flexibly adjusted by controlling the sizes of the adjustable capacitors and gms, and thus the wireless communication system of multiple standards can be applied.

The application discloses a kind of Butterworth filter joint grey correlation degree weight's ocean remote sensing image denoising method, comprising: the ocean remote sensing image containing noise is carried out smoothing filtering, obtains initial noise image;The ocean remote sensing image containing noise and the initial noise image are carried out grey absolute correlation degree analysis, obtain target noise pixel image and target pixel area image;The target noise pixel image and the target pixel area image are carried out grey correlation degree weight denoising, obtain final denoising image.The application effectively distinguishes salt and pepper noise, accurately attacks noise, so as to realize the efficient and meticulous noise suppression of remote sensing image.

The present invention discloses a method for estimating the runway pose of a monocular vision aircraft based on line features. The method includes: using the least squares method and the angle bisector method to correct the left, right, and center line equations of the aircraft runway in the on-board monocular vision image; calculating the position and attitude information of the aircraft relative to the aircraft runway based on the corrected left, right, and center line equations of the aircraft runway and the information of the starting line of the aircraft runway; using a Butterworth filter to optimize the position and attitude information to obtain a reliable and stable estimation result of the aircraft runway pose. The present invention only uses the line features extracted from the image to estimate the aircraft attitude, avoiding the large number of feature point matching problems in the PnP (Perspective-n-Point) algorithm, and has good real-time performance and accuracy, so as to provide support for subsequent real-time autonomous landing.

The invention discloses a bank security abnormal behavior detection method and device, equipment and a medium, and relates to the field of machine learning, and the method comprises the steps: carrying out the preprocessing of a current original image in a monitoring video of a bank security scene, and obtaining a current target image; processing the current target image by using a feature extractor of the target image classification model to obtain an initial activation value; if the initial activation value is not smaller than the first preset threshold value, correcting the initial activation value to a preset activation value interval by using a Butterworth filter to obtain a target activation value; if the initial activation value is smaller than the first preset threshold value, determining the initial activation value as a target activation value; inputting the target activation value into a classifier of the target image classification model to output an abnormal behavior detection result; the abnormal behavior detection result represents that the operation behavior corresponding to the current target image belongs to or does not belong to the abnormal behavior in the bank security scene. And the accuracy of abnormal behavior detection of bank security and protection is improved.

The invention discloses a marine mobile platform unmanned aerial vehicle autonomous take-off and landing device adaptive to the ship speed of 16 knots, and belongs to the technical field of marine unmanned aerial vehicle take-off and landing. According to the invention, a double-layer prediction-compensation control architecture is adopted: a first layer is fused with an extended Kalman filter and an LSTM neural network, and 0.8-1.2-second prediction is realized in cooperation with 100 motion mode fingerprint databases; and the second layer adjusts feed-forward compensation in real time based on the Lyapunov adaptive law. Frequency-division decoupling is achieved through a third-order Butterworth filter bank, model prediction control is adopted in a low frequency band, adaptive boundary layer sliding mode control is adopted in a middle frequency band, and a disturbance observer is adopted in a high frequency band. The executing mechanism is of a hybrid drive structure, four magnetorheological dampers and eight piezoelectric actuators are matched to serve as main drives, and six electric push rods serve as backups. According to the method, the attitude prediction error is reduced within + / -0.3 degrees, the rolling / pitching control precision reaches + / -0.5 degrees, the heaving control precision reaches + / -50mm, and the take-off and landing success rate of the unmanned aerial vehicle under the 4-level sea condition reaches 98.6%.

The invention relates to the technical field of mental detection, and discloses a mutual information-based regularization autism classification algorithm, which comprises the following steps of: firstly, extracting different frequency bands of theta (4-8Hz), # imgabs0 # beta (13-30Hz) and gamma (above-30Hz) from autism electroencephalogram data through a Butterworth filter, and secondly, calculating the power spectral density of a preprocessed signal through a Welch period method; according to the method, the important features with robustness can be adaptively selected by calculating the symmetric uncertainty between the features and the category labels, the important features with robustness can be adaptively selected according to the correlation between the features and the target category labels, noise or irrelevant features are eliminated, and the method not only can improve the classification precision, but also can improve the classification efficiency. And the feature dimension can be reduced, and the calculation complexity is reduced.

The present invention discloses a method for adaptively extracting digital holographic object image spectrum that eliminates stray spectrum. The holographic interference pattern of the object to be measured is collected and converted into a spectrum diagram, and the center coordinates of the object image spectrum are determined by phase information; iterative threshold segmentation processing is performed to obtain a spectrum binary segmentation diagram; the object image spectrum binary segmentation diagram is extracted and generated in the foreground area of the spectrum binary segmentation diagram according to the center coordinates of the object image spectrum; the object image spectrum binary segmentation diagram is subtracted from the spectrum binary segmentation diagram to obtain a stray spectrum area binary segmentation diagram, and a spectrum diagram after eliminating stray spectrum is obtained on the spectrum diagram according to the stray spectrum area binary segmentation diagram; the object image spectrum is filtered by an adaptively generated Butterworth filter, and a three-dimensional morphology diagram is reconstructed by unwrapping and distortion compensation. The present invention solves the contradiction between the amount of object detail information extracted and the amount of stray spectrum introduced, minimizes the amount of stray spectrum introduced while increasing the amount of object detail information extracted, thereby improving the quality of object three-dimensional morphology reconstruction.

The invention discloses a self-adaptive filtering method for frequency division multiplexing ultrafast imaging reconstruction, which belongs to the technical field of computational optical imaging and comprises the following steps of: simulating and generating a large number of multi-framing frequency division multiplexing ultrafast imaging original images by adopting a method of randomly generating a matrix and interpolating and combining a grating modulation principle; correspondingly removing an unmodulated random complex amplitude image from each original image so as to obtain a zero-order item-free image; the pix2pixGAN generative adversarial network is improved and trained; inputting a shot actual multi-framing frequency division multiplexing image into the trained network, outputting a multi-framing frequency division multiplexing image without a zero-order item, performing Fourier transform to obtain a spectrogram without the zero-order item, performing binarization processing on the spectrogram, and applying a K-Means + + clustering algorithm to obtain center coordinates of positive and negative first-order items of a frequency spectrum, so as to obtain a multi-framing frequency division multiplexing image without the zero-order item; single-frame image information corresponding to the center coordinates is extracted through a Butterworth filter, inverse Fourier transform is carried out, a frequency division multiplexing single-frame image is obtained, and self-adaptive filtering is achieved.

The invention provides a gearbox fault prediction and diagnosis method, which comprises the following steps of: acquiring sensor data of a preset point location of a gearbox, storing the sensor data according to a standardized format, and filtering interference noise such as strong electromagnetism and vibration of a wind power plant by customizing parameters of a Butterworth filter; inputting the standardized data set into a multi-branch parallel convolutional neural network model to extract feature sub-segments of each preset point location, and splicing global high-dimensional feature vectors; point feature weights are dynamically adjusted through point and segment training, fault high-incidence parts are focused, irrelevant data interference is reduced, fault types and severity scores are output in combination with an adaptive parameter random forest model, and the problems that a traditional model is poor in generalization ability and cannot adapt to gearboxes of different models / working conditions are solved. High-precision recognition and high-time-efficiency early warning across early faults of the gearbox are achieved, and the problem of gearbox fault monitoring and early warning lagging is solved.