107results about How to "Choose accurately" patented technology

The invention relates to the technical field of big data analysis and artificial intelligence, in particular to a multi-source heterogeneous big data fusion and reasoning method and system based on a knowledge graph, and the method comprises the following steps: constructing mapping interfaces of various data sources, collecting and storing entities in the data sources, and constructing nodes in the knowledge graph; performing representation learning on a graph structure of the knowledge graph based on a graph neural network, extracting an embedded vector of a node, and extracting a fusion node through the embedded vector and a mapping interface; performing knowledge completion on the fusion nodes by using a neural network based on symbolic logic; generating candidate reasoning paths by adopting a depth-first search algorithm based on the knowledge graph; reasoning is carried out based on an attention mechanism and path similarity, candidate reasoning paths are screened and sorted to obtain an optimal reasoning path, a graph neural network is adopted to carry out representation learning of a knowledge graph, symbolic logic and the neural network are combined, the interpretability of the reasoning process is guaranteed, and the learning ability is high.

The invention relates to an automatic switching method of a terminal and a satellite based on multi-attribute decision, and belongs to the technical field of satellite communication. The objective of the invention is to solve the problems of frequent switching and network performance reduction caused by high-speed movement of satellites in a 5G NTN low-orbit satellite communication scene. The method comprises the following steps: firstly, a terminal collects information of satellites in a visible range, including longitude and latitude, signal strength, serial numbers, channel resources and the like, and reports the information to a service satellite; and the service satellite calculates the remaining coverage time of the current satellite on the terminal according to the reported information, and judges whether to trigger switching. And if switching is triggered, the service satellite screens a candidate satellite set, quantitative evaluation is carried out according to attributes such as a satellite elevation angle, residual service time, the number of channels and switching time delay, and the satellite with the maximum comprehensive quantitative value is selected as a target satellite. And finally, the service satellite initiates a switching request to the target satellite and starts a switching process to complete switching between the terminal and the satellite.

The invention proposes a hard negative sampling comparison recommendation method based on a graph generation diffusion model, and the method comprises the following steps: coding a user-article interaction graph through a graph convolutional neural network, obtaining a node embedded vector, and carrying out the inner product of the node embedded vector, and obtaining a recommendation probability; the model parameters of the graph convolutional neural network are obtained through the following training steps: S1, generating a comparison view 1 and a comparison view 2 from the node embedding vector through an inline diffusion model view generator and a parameter topological noise reduction view generator; the inline diffusion model view generator comprises an inline diffusion model, and the parameter topological noise reduction view generator comprises a parameter topological noise reduction network; and S2, carrying out comparative learning on the comparative view 1 and the comparative view 2. According to the method, a signal alignment strategy is combined, a double-view generator comparison learning framework is built, and the problems that the model generation capacity is insufficient and the influence of data noise is large are solved, so that the performance of a recommendation system is improved.

The invention provides a remote sensing multispectral imaging system for agricultural monitoring, which relates to the technical field of space remote sensing, and realizes accurate selection by indicating an economic crop determination module, analyzing historical soil salinity and water shortage data and selecting sensitive crops; the correlation evaluation module calculates a water shortage index change rate, analyzes correlation between historical data and current data, and improves prediction accuracy; the value range determination module divides soil salinity intervals and standardizes management data in combination with evaluation coefficients; the indication area determination module realizes accurate monitoring by using a multispectral image and enhances representativeness; the humidity evaluation module analyzes local spectral data to quantify the soil humidity; the prediction model construction module predicts the salinity content by using machine learning, and can accurately predict dynamic changes of soil salinity and moisture; the correlation analysis module verifies real-time monitoring data in combination with the prediction model; and the division module optimizes a soil management strategy according to the water shortage index and the humidity coefficient.

The invention discloses a digital audio watermark processing method and device, equipment and a storage medium, and belongs to the technical field of artificial intelligence. The method comprises the following steps: receiving a watermark adding instruction, and carrying out Fourier transform on original host voice to obtain an audio signal; identifying a watermark information adding bit in the audio signal through the audio watermark model, and embedding a watermark message in the watermark information adding bit; the interfering audio signal and the audio signal carrying the watermark message are combined. Receiving the message extraction signal, and separating an interference audio signal in the combined audio signal; a watermark information adding bit in the audio signal is identified, and a watermark message is extracted from the watermark information adding bit. The invention also relates to the technical field of block chains, and the host voice is stored on a block chain network. According to the method, the watermark information adding position is automatically identified through the audio watermark model, so that the watermark information is added and extracted, the dependence on expert knowledge is remarkably reduced, and the watermark processing speed and efficiency are improved.

The invention discloses a visual odometer method, system and device based on point-line features and a medium. Visual odometers are crucial to navigation and planning of autonomous robots, but weak light conditions, violent illumination changes and low texture scenes make a significant challenge to odometer estimation. Therefore, according to the visual odometer method based on the point-line features, the point-line features and deep learning are integrated, the point features and the line features are obtained through a self-supervised learning model for interest point detection and a line detection mode combining line light flow tracking and cross constraint, and the influence of complex lighting conditions is overcome. Meanwhile, the key frame is selected based on point feature counting and line feature overlapping angle constraint, and sparse matching is performed by integrating the convolutional neural network and the graph neural network, so that the accuracy and the calculation efficiency are improved.

The invention discloses a transaction contract dispute processing method, a transaction contract dispute processing device and a computer program product. Relates to the financial science and technology field or other related fields, and the method comprises the steps: obtaining a transaction contract dispute request, determining M evaluation objects according to the transaction contract dispute request, obtaining the dimension data of the M evaluation objects, and obtaining M groups of dimension data, each group of dimension data comprising N pieces of dimension data; for one evaluation object, determining a weight coefficient of each dimension data associated with the evaluation object, and determining a time decay parameter of the evaluation object; inputting the N pieces of dimension data associated with the weight coefficient of the evaluation object and the time decay parameter into a score prediction model to obtain an evaluation score; and selecting Y target evaluation objects based on the M evaluation scores. Through application of the method and the device, the problem of low efficiency during transaction contract event processing caused by low matching accuracy during selection of the evaluation object for evaluating the transaction contract event in related technologies is solved.

The invention discloses a torsional vibration suppression strategy for an electric drive system based on a double-layer model predictive control, and relates to the technical field of electric vehicle motors. The invention establishes a mathematical model of a permanent magnet synchronous motor taking into account the dynamic characteristics of the motor based on a stator voltage equation in a d-q coordinate system; based on the dynamics of a gear system, a 19-degree-of-freedom gear transmission system model is established taking into account nonlinear factors such as time-varying meshing stiffness, meshing error, and tooth side clearance; based on the model of the electric drive system, a torsional vibration suppression strategy based on a double-layer model predictive control is proposed, which suppresses the torsional vibration of an integrated electric drive system by controlling the motor target torque and reducing current harmonics and torque fluctuations, thereby enhancing the stability of the electric drive system and improving the comfort of the whole vehicle.

The utility model discloses an integrated sampler for farmland soil quality detection, and relates to the technical field of soil sampling machinery. The device comprises a mounting box, a control screen is arranged at the top of the mounting box, a first electric telescopic rod is arranged at the top of an inner cavity of the mounting box, a mounting block is arranged at the output end of the first electric telescopic rod, a rotating motor is arranged at the bottom of the mounting block, and a spiral drill rod is arranged at the output end of the rotating motor. A plurality of mounting grooves are formed in the side face of the spiral drill rod, second electric telescopic rods are arranged on the side faces of the mounting grooves, and sampling assemblies are arranged at the output ends of the second electric telescopic rods. According to the utility model, separated sampling is carried out through the arranged sampling assembly, so that soil samples can be accurately selected according to the measured depth required actually, layered grouping sampling at different heights can be carried out on the same sampling point through one-time operation, the operation is convenient and labor-saving, and the sampling efficiency is improved. The soil samples are not influenced and mixed in the sampling process, so that the independence of each soil sample is ensured.

The invention belongs to the technical field of airdrop and special vehicles, and particularly discloses a wheel type airdrop vehicle parachute opening impact strength simulation evaluation and optimization method and system. Comprising the following steps: performing parachute opening dynamics simulation to obtain an accurate boundary of parachute opening strength impact simulation; performing whole vehicle parachute opening strength dynamic simulation analysis to obtain an analysis result of vehicle body strength and deformation, evaluating the analysis result, outputting design parameters if the analysis result meets requirements, and otherwise, performing failure reason analysis; air-drop system modal analysis is added, and failure reasons are judged; according to failure reasons, partial grids are intercepted or static simulation is adopted to replace dynamic simulation, and a reduction optimization model is obtained; and based on HyperStudy, carrying out multi-objective optimization according to the structural characteristics of the parts of the failure parts and failure reasons. According to the method, high-precision calculation of the re-loading air-drop process of multi-stage parachute opening of multiple parachutes is achieved, and scientific optimization of the parachute opening strength structure of the air-drop vehicle is achieved by optimizing the simulation means.

The invention relates to the technical field of construction progress regulation and control, and discloses a construction progress regulation and control system and method in engineering construction based on an unmanned aerial vehicle. The method comprises the following steps: firstly, acquiring an inclined image of an engineering construction site, extracting feature points by using a light beam method and an SURF algorithm, matching the feature points, and establishing an engineering construction site live-action model; secondly, clustering site clouds of the engineering construction site in combination with an engineering construction site real scene model, extracting a construction site building point cloud set, performing deviation analysis, and adjusting the engineering construction progress; a BP neural network is trained, and the engineering construction completion rate is predicted; and finally, according to the engineering construction completion rate, the engineering construction progress plan is adjusted and optimized based on a litsea rotundifolia optimization algorithm, and engineering construction progress control is realized. The purpose of regulating and controlling the construction progress is achieved by adjusting and controlling the engineering construction progress, and the method is accurate and objective.

The license plate manufacturing production line comprises a rack, a license plate feeding mechanism, a license plate conveying mechanism, a character die storing and taking mechanism, a character pressing mechanism, a license plate transferring mechanism and a character erasing mechanism. The automatic word pressing and erasing device has the advantages that the functions of automatic word pressing and automatic word erasing can be achieved, trouble and labor are saved, the production efficiency can be improved, and the problem that the product percent of pass is reduced due to manual errors can be solved. And meanwhile, each type matrix is independently selected, so that a plurality of type matrixes can be selected at the same time, waiting time is avoided, the situation that the type matrixes are not enough can be avoided, and the type matrixes can be accurately and quickly selected even if a plurality of number plates with continuous numbers appear.

The invention discloses a wide-angle-domain low-scattering antenna design method based on a characteristic mode theory, and belongs to the technical field of antenna stealth design. The antenna is designed according to radiation performance requirements, then radiation mode analysis is carried out, the important radiation mode is determined according to the mode amplitude, and the area to be reserved of the antenna is determined according to the mode current distribution condition of the important radiation mode of the antenna; then, calculating a wide-angle-domain RCS and a scattering mode of the antenna according to a stealth requirement, selecting a common important scattering mode in the wide-angle domain according to the scattering mode amplitude, and selecting scattering current which can be removed or cut according to the current distribution condition of the common important scattering mode of the antenna and the current distribution characteristics of the important scattering mode; and the mode current of an important radiation mode is not influenced, so that the wide-angle-domain low-scattering antenna design is realized. The antenna designed by the invention has the advantages of integration of radiation and scattering, wide angular domain and low scattering, and can meet single-station / double-station RCS stealth requirements.

The invention discloses a multi-modal ultrasonic image recognition method and device, which effectively combines an ultrasonic image and a corresponding task text, and improves the recognition precision of the ultrasonic image by utilizing the interaction of visual information and language information. By preprocessing the ultrasonic image and the task text and generating a visual Token sequence and a language Token sequence, data of different modalities can be effectively fused to generate a multi-modal sequence. The sequences are subjected to context interactive fusion of a multi-head attention model to obtain a multi-modal embedded representation containing visual and language information, so that the model can accurately analyze a task instruction while understanding the image content, and the recognition accuracy is improved. And furthermore, through selection and weighted fusion of dynamic experts and in combination with background knowledge provided by static experts, the processing capability of the model is improved. And finally, the output result is adjusted and optimized by using the fusion control parameters, so that the efficiency and the accuracy of ultrasonic image recognition are improved.

The utility model discloses a sampling device for indoor environmental pollution abatement, which comprises a bottom plate, universal wheels fixedly arranged at four corners of the bottom of the bottom plate, a handrail fixedly arranged on one side of the upper end face of the bottom plate, an air storage tank fixedly arranged on the upper end face of the bottom plate, an exhaust valve arranged at the upper end of the air storage tank, and an air pump fixedly arranged on the upper end face of the bottom plate. The air outlet end of the air pump is communicated with the air storage tank, the air inlet end of the air pump is communicated with a sampling mechanism, the sampling mechanism comprises a telescopic rod mechanism, a hollow ball is fixedly arranged at the upper end of the telescopic rod mechanism, and an inner cavity of the hollow ball is connected with the air inlet end of the air pump through a hose; compared with the prior art, the sampling device has the advantages that the sampling device adopts the design of the telescopic rod mechanism and the hollow ball, so that the position of the sampling head can be flexibly adjusted, the sampling position can be more accurately selected, a plurality of uniformly distributed air inlet holes are formed in the side surface of the hollow ball, air samples in multiple directions can be simultaneously collected, and the sampling efficiency is greatly improved; and the operation time is reduced.

The invention belongs to the technical field of traditional Chinese medicine, and particularly relates to a personalized traditional Chinese medicine inquiry system based on a traditional Chinese medicine large model and multiple knowledge bases, which comprises the traditional Chinese medicine large model, a traditional Chinese medicine pre-training embedding model, a sorting model, a traditional Chinese medicine vector database and a traditional Chinese medicine graph database, the output end of the traditional Chinese medicine pre-training embedded model is connected with the traditional Chinese medicine vector database, and the output end of the traditional Chinese medicine pre-training embedded model is connected with the traditional Chinese medicine graph database. And the construction difficulty of the preference database of the traditional Chinese medicine large model in RLHF training is reduced. The adaptation of the pre-trained semantic embedding model to the vectorization of the traditional Chinese medicine text is improved, and the accuracy of vector matching is improved. And a traditional Chinese medicine knowledge graph data source is increased. A traditional Chinese medicine large model is trained, and a sorting model is trained and introduced, so that context selection is more accurate. Vector data and a graph database are constructed for each famous traditional Chinese medicine, and personalized question and answer logic for each famous traditional Chinese medicine is realized.

The invention discloses a wharf process flow target path selection method and system based on big data, and relates to the technical field of wharf process flow target path selection management. The wharf process flow target path selection method based on the big data comprises the following steps of dynamic evaluation of wharf operation, static path evaluation and split and combined path evaluation. According to the method, the wharf process flow target path is selected through the obtained wharf operation dynamic variability index, if the static path is selected, whether the static path is switched to the split and combined path or not is judged based on the obtained static path efficiency evaluation index, and if the split and combined path is selected, the static path is switched to the split and combined path. If yes, whether the standby combined path is adopted or not is judged through the obtained initial combined path efficiency index, the effect of improving the wharf technological process target path selection accuracy is achieved, and the problem that in the prior art, the wharf operation dynamic change situation is not fully considered when the wharf technological process target path is selected is solved.

The present invention relates to the field of software development, and in particular to an update loading method based on dynamic isolation, comprising: obtaining a dependency library to be selected, determining a library state according to a dependency library complex comparison value and a dependency library complex reference value; determining a dependency library selection mode according to the library state of the dependency library to be selected; selecting a number of first prepared dependency libraries in response to different dependency library selection modes; performing secondary screening of the first prepared dependency libraries when the number of the first prepared dependency libraries is less than or equal to a preset number or when there is a dependency library complex reference value greater than or equal to a preset dependency library complex reference value; determining a secondary screening benchmark according to an API usage bias in the secondary screening, and selecting a target dependency library according to the secondary screening benchmark; determining a final dependency library for the target dependency library, and selecting direct loading or environment isolation loading according to whether the final dependency library has a dependency conflict; reducing the waste of data processing resources.

The present application relates to the field of satellite communication technology, and discloses a link transmission method, device, equipment and storage medium, including: obtaining interface data in a service data interface, and encapsulating the interface data based on a preset protocol frame format to obtain first service data; determining a first control message based on the physical channel information of the current node receiving end; parsing the first control message to obtain first control channel data and receiving status, and generating a link sending strategy based on the first control channel data and receiving status; according to the link sending strategy, allocating the first service data to different link caches through pre-packaging for transmission and forwarding. Packaging and transmission link selection are realized between the physical layer and the link layer through the on-chip system interface, and the upper layer protocol does not need to obtain limited channel information through self-testing, so that the link selection is more accurate and the service transmission rate is more stable.

The invention relates to the field of mechanical control, in particular to a method, system and equipment for controlling a mechanical arm through voice and a medium. The method comprises the following steps: collecting a voice control instruction signal of an electric power production field; noise reduction is carried out on the voice control instruction signal by adopting a spectrogram wavelet adaptive threshold noise reduction technology; converting the de-noised voice signal into a text or an instruction by using a pre-trained improved recurrent neural network model; and the mechanical arm control system executes corresponding mechanical arm operation according to the recognition result. The system comprises a voice module, a visual module and a mechanical arm control module. According to the invention, the quality of the voice signal is effectively improved through the spectrogram wavelet transform and adaptive threshold noise reduction technology, so that the voice instruction can be accurately recognized in a noisy working environment. And the accuracy and the intelligent level of voice recognition are improved.

The invention discloses an ultra-wideband weak signal detection method and system based on all-optical oscillation and phase-amplitude conversion. The method comprises the steps that spontaneous radiation light generated by an erbium-doped optical fiber amplifier is used as an optical carrier; modulating a weak radio frequency signal to be measured received by an antenna to the optical carrier through an optical modulator to form a modulated optical signal; the modulated optical signal is injected into an all-optical oscillation circuit composed of double loops, the all-optical oscillation circuit generates multimode optical oscillation by adjusting an optical attenuator, and the all-optical oscillation circuit comprises a double-loop interference structure; performing phase-amplitude conversion by adjusting the delay difference of the two light paths in the double loops to form an equivalent microwave photon filter so as to select and amplify the modulated light signal corresponding to the radio frequency signal of the set frequency band; and converting the selected and amplified modulated optical signal into an electric signal, and performing spectral analysis to detect the weak radio frequency signal to be detected.

The invention provides a replacement cable selection method, and relates to the technical field of cable replacement, and the method comprises the steps: calculating the conductor shielding field intensity of a to-be-replaced cable and the insulation shielding field intensity of the to-be-replaced cable; performing preliminary screening from existing selectable cables to obtain a sample cable set; obtaining all types of performance evaluation values of each sample cable in each monitoring period; constructing a performance comprehensive evaluation matrix corresponding to each monitoring period of each sample cable; a comprehensive performance evaluation value of each sample cable is calculated, and a replacement cable is selected based on the comprehensive performance evaluation value of each sample cable. The replacement cable is selected through systematic steps, the scientificity of the selection process of the replacement cable is ensured, the electrical performance parameters of the to-be-replaced cable are fully considered, the influence of the actual parameters and limiting conditions of the construction environment on the replacement cable is emphasized, the possibility of cable failure is reduced, and the construction efficiency is improved. And the total cost of laying and later maintenance is fully considered in the cable selection process, so that the operation cost is reduced.

The embodiment of the present invention provides a method, device, electronic device and storage medium for recommending game characters. A graphical user interface is provided through a first terminal device. The method includes: in response to a game account corresponding to the first terminal device entering a game character selection stage, displaying a game character to be selected on the graphical user interface; receiving recommendation information of a game character sent by at least one second terminal device, wherein the game account corresponding to the second terminal device and the game account corresponding to the first terminal device belong to the same camp in the current game; and determining a target game character from the game characters to be selected according to the recommendation information. Even if the player of the first terminal device of the embodiment of the present invention is not familiar with the game character, or does not pay attention to the game character selection and communication prompts of the player of the second terminal device, he can still correctly select the game character, avoiding adverse effects in the game character selection stage, and improving the player's gaming experience.

This invention discloses a method for measuring the safe distance between a construction crane and energized equipment in a substation based on energized interval scale calibration, comprising the following steps: S1: Constructing a crane construction scene dataset. S2: Inputting the target frame into a relative depth estimation network and outputting the relative depth value of the target frame. S3: Obtaining the relative pose from the target frame to the source frame using an inter-frame motion estimation network. S4: Constructing a crane motion confidence map and multiple loss functions. S5: Constructing a multi-loss joint optimization function and jointly training the relative depth estimation network and the inter-frame motion estimation network. S6: Converting the relative depth values ​​of the entire image into absolute scale depth values. S7: Calculating the safe distance and outputting the measurement result or warning result. This invention uses a monocular camera to record multiple consecutive frames of images and combines multiple loss functions for cross-frame temporal constraints and contour smoothing correction, ensuring accurate selection of the final measurement point, and can directly serve the safety monitoring scenario of crane operation near energized equipment.

The invention discloses a three-dimensional digital image method (3D-DIC) based on adaptive calculation of an optimal image sub-region size, and the method comprises the steps: estimating the size of an initial image sub-region and the value range of the size of the image sub-region based on a speckle evaluation method of an SSSIG (Sum of Quality of Square) threshold value of the image sub-region; performing attempt DIC calculation of time sequence matching and stereo matching based on the size of the initial image subarea to obtain an initial displacement field, and estimating a high-order displacement derivative based on a local least square fitting method; calculating an optimal image sub-region size based on a high-order displacement derivative and speckle related parameters; the initial displacement field serves as an initial value, and a final image displacement result is obtained through optimization calculation by combining the optimal image sub-area size optimization of time sequence matching and stereo matching; based on simulation image data generated by a simulation stereoscopic vision system, it is verified that the self-adaptive algorithm used in the method has higher measurement precision than a conventional three-dimensional digital image correlation method.

The invention discloses a meter image extraction method and device, and the method comprises the steps: extracting a plurality of to-be-processed images from an image list corresponding to a target meter according to a preset time interval, obtaining a geometric figure of a dial plate through employing each to-be-processed image for shape analysis, and further selecting a time period through employing a plurality of dial plate shape analysis results, and obtaining a geometric figure of the dial plate according to the selected time period. And performing image extraction on the image list again according to the selected at least one target time period, calculating a definition score for the dial reading area in each extracted candidate image, selecting the candidate image corresponding to the highest definition score as a target image, and displaying the selected target image on the dial reading area. The method realizes staged image selection under the constraint of the geometrical shape and definition of the part with the reading in the meter corresponding to the image, helps to improve the extraction probability of the meter image with the highest reading definition, and helps to better support accurate identification of the reading of the meter.

The application provides a national encryption and international algorithm compatible method for eUICC card and eIM platform public key interaction and an eUICC card, relates to the technical field of Internet of Things communication, and comprises the following steps: configuring configuration data information corresponding to multiple eIM platforms in the eUICC card, wherein the configuration data information contains the platform unique identification and service public key data of each eIM platform, the service public key data contains service information verification public key and an added algorithm type field, and the algorithm type field is used for uniquely identifying national encryption algorithms or international algorithms; in a business interaction process, the eUICC card matches corresponding configuration data information through the platform unique identification in a service data packet message sent by a target eIM platform, and then performs signature verification on the service data packet information based on the national encryption algorithms or international algorithms selected through the algorithm type field and the service information verification public key, so as to realize the security verification of the business interaction with the target eIM platform. The beneficial effect is that the accurate compatibility of national encryption and international algorithms in the public key mode is realized.

The invention relates to the technical field of online sales service, and discloses an online sales service method based on clothing pattern design, which comprises the following steps: a customer inputs body data; obtaining a target garment purchased by the customer and the type thereof; analyzing all historical evaluation records of the customer, and judging whether a good evaluation record the same as the target garment type is included or not; if so, calculating an average value of various data in the favorable comment record according to a first strategy, and recommending the most suitable size; if not, judging whether the customer opens the historical browsing record or not. If not, screening out a target population similar to the customer body data, analyzing a good reputation record of the target population, and recommending a size; and if open, selecting a target size according to a data change trend by analyzing a historical browsing record of the customer. According to the method, accurate size recommendation is realized by comprehensively analyzing customer body data, historical evaluation records and browsing behaviors.

The invention discloses a data processing method and device for scene visualization, and the method comprises the steps: determining a scheduling scene type when a scene visualization demand is received, determining a corresponding scheduling mechanism according to the scheduling scene type, scheduling a corresponding node set according to the scheduling mechanism, and carrying out the loading display, so as to complete the visualization process of a scene, the scheduling mechanism is an internal and external memory scheduling mechanism based on a leap-over type detail level model, and the scheduling mechanism at least comprises a direct scheduling mechanism, a scheduling mechanism of a father node substitution method or a scheduling mechanism of a knapsack problem solving method.

The application provides a rotating machinery impact fault feature extraction method, device and storage medium, and belongs to the technical field of equipment fault diagnosis. The method comprises the following steps: acquiring an impact response vibration signal; calculating the envelope spectrum kurtosis of each sub-band signal, wherein the sub-band signal is a sub-band signal at different levels decomposed from the impact response vibration signal; marking the non-cyclostationary frequency band according to the envelope spectrum kurtosis; calculating the square envelope autocorrelation kurtosis of the unmarked sub-band signal; determining the frequency band of the sub-band signal corresponding to the maximum square envelope autocorrelation kurtosis as the optimal demodulation frequency band; and performing envelope demodulation on the impact response vibration signal in the optimal demodulation frequency band to obtain the impact fault feature of the rotating machinery. The method disclosed in the application sequentially evaluates the impact characteristics and cyclostationarity characteristics of the impact response vibration signal, selects the optimal fault frequency band, and realizes accurate extraction of the impact fault feature.