112results about How to "Guaranteed processing accuracy" patented technology

The utility model discloses an updating method for a moving object based on the road transportation network and GPS, aiming at enabling a central server to trace the former and current position of the moving objects. The utility model is characterized in that the moving object uses GPS to measure the latest running data continuously, through matching with the transportation network, the coordinate of the latitude and the longitude and the direction data are converted into the data format based on the road network, then the current running vector is compared with the running data submitted by the updating on the last position according to the principle of the inertia, and the position updating condition is determined, the current running vector of the moving object is reported to the central server only when the moving object does not keep running as the principle of the inertia. Additionally, the utility model optimizes the updating of the position in the road crossing, only one time position updating of the moving object is performed at most in the road crossing. The utility model has advantages of reducing the communication frequency and improving the data processing efficiency under the premises of securing the processing precision compared with the prior art.

The invention discloses precision hole processing equipment for a thin-wall casing type part. An equipment body is installed on the outer side of the casing type part, a base is arranged on the bottomof the equipment body, a vertical rod is installed on the middle of the base, internal fixators are installed on the circumference of the top end of the vertical rod, a screw rod is installed on thetop of the vertical rod, and a convex edge is arranged at the top end of the screw rod. A sliding sleeve is arranged on the outer side of the screw rod in a sleeving mode, top fixators are installed on the circumference of the sliding sleeve, and bottom fixators are installed on the outer side of the base. According to the precision hole processing equipment for the thin-wall casing type part, thescrew rod and the sliding sleeve are cooperatively used, the sliding sleeve can drive the top fixators to move up and down on the screw rod, the top end of the casing part is fixed and prevented frommoving in the vertical direction, meanwhile, flexible adjustment can be carried out according to parts with different heights, simplification and practicality are achieved, and the application rangeof the equipment body is expanded.

The invention discloses a method and a device for rapidly implementing pre-coding. The device comprises a pre-coding module, a pre-coding code book table storage unit, a pre-coding code book table construction module and a pre-coding mode setting module. The method comprises the following steps: (A) the pre-coding code book table construction module establishes and stores a pre-coding code book table; (B) the pre-coding mode setting module sets different terminal velocity grades and relevant pre-coding modes; (C) when a terminal accesses to a base station, the pre-coding module determines a standby pre-coding mode according to the terminal velocity grade of the terminal, and searching pre-coding code books from the pre-coding code book table according to the number of the current base station-side antennae and the number of the current base station-side mapping layers; and (D) the pre-coding module performs the pre-coding operation by use of the standby pre-coding mode and the searched pre-coding code books. The method ensures the data processing accuracy while reducing the complexity of pre-coding operation, reduces the processing time delay, saves the system resources, and ensures the pre-coding accuracy.

The invention provides a prediction and correction method of nodes of a star point track image, and the method comprises the following steps of: (1) the prediction and the imaging of a star point dynamic track with nodes; (2) the judgment of the nodes and the selection of a template; (3) the correction. The prediction and correction method of nodes of a star point track image of the invention performs a selective judgment and a position correction with pronounced features to the nodes on a star point trajectory, ensures the prominence and the position precision of the template at the node, consequently ensures the processing precision of a deep-space star point track image, and facilitates the improvement of the matching precision of the next step.

The invention discloses a follow-up clamping device for clamping and positioning pipe pieces. The device comprises a first bearing body, a second bearing body arranged on the first bearing body in a sliding manner, a clamping mechanism fixed on the first bearing body, and a bearing mechanism arranged on the second bearing body in a lifting manner, wherein the clamping mechanism comprises two groups of telescopic butting heads forming angles; the two groups of butting heads are symmetrically arranged left and right; the bearing mechanism comprises two rows of rollers symmetrically arranged left and right; rotating axle lines of the rollers in one group are parallel to one another, and are positioned in a same plane; two planes, where the rotating axle lines of the two rows of rollers are positioned, are intersected with each other; the pipe pieces are clamped between the two groups of butting heads and the two rows of rollers; and each of the rollers is arranged along the tangential direction of the pipe pieces. The pipe pieces are stably clamped and positioned through the two rows of rollers and the two groups of butting heads so as to provide guarantee to the technique precision of following operations for the pipe pieces; and the device has a reasonable structure, is easily and reliably adjusted, and is suitable for the pipe pieces with different pipe diameters.

The invention relates to a mobile measurement method fusing an SLAM technology in a complex environment, and belongs to the technical field of point cloud data processing. The measurement method comprises the following steps: acquiring laser point cloud data and the pose of a carrier; eliminating motion distortion of the laser point cloud data by utilizing the pose of the carrier; carrying out theregistration of the sub-maps through the SLAM technology, and constructing a local factor graph, wherein the local factor graph comprises a speedometer factor, an IMU pre-integration factor, a firstregistration factor, a small closed-loop factor and a sub-map; optimizing the local factor graph to obtain a local motion trail of the carrier and an optimized sub-map; constructing a global factor graph by taking the optimized sub-map as an optimization unit, wherein the global factor graph comprises a second registration factor, a continuous registration factor, a large closed-loop factor and aGNSS factor; and optimizing the global factor graph to obtain a global motion trail and a global map of the carrier. According to the method, the SLAM technology and the GNSS / IMU positioning technology are fused, so that the smoothness of the positioning track is effectively improved, and the map construction precision is improved.

The invention discloses lost foam cutting equipment, which is provided with a sliding clamp with a function of fine adjustment. The sliding clamp can be used for achieving fine height adjustment for an electrothermal wire, convenience in operation, high precision, effective improvement of productivity, and economization of production cost. A standard table guide is reasonably designed and used with a sliding bracket in sliding, so that guidance precision is high. The bracket made of wood is light in weight. A cutting foam flat-plate is made of foam, so that processing and manufacturing precision of the cutting foam flat-plate can be guaranteed effectively.

The invention discloses an open flow flow table refreshing method. Table items in a table item lookup chip are divided into a plurality of position intervals according to priorities to enable the priorities of the plurality of continuous table items in the same position interval to be equal to each other, the table items of which the priorities are lower than the priority of a new table item in an open flow flow table are determined when whether the new table item needs to be inserted or not is determined and a first table item is placed behind an end table item in every position interval corresponding to the table items in turn, and accordingly the new table item can be inserted into the unoccupied position behind the table items of which the priority is equal to that of the new table item. Accordingly, the updating efficiency of the open flow flow table in the table item lookup chip is effectively improved under the condition that the processing accuracy is ensured.

The invention discloses a high density surface wave exploration data acquisition and processing method comprising two parts of data acquisition and data processing. The data processing part is dividedinto preprocessing, waveform processing, frequency dispersion analysis, inversion analysis and visualization processing. The high density surface wave exploration data acquisition and processing method comprises the following concrete steps that S1: data acquisition uses the exploration mode of measuring point encrypting and overlapping, and 24 detectors are arranged on the measuring line according to a certain interval to form a sensor array in the step one. The geotechnical engineering surface wave exploration data acquisition method and processing method and flow are innovated and standardized, including preprocessing, waveform processing, frequency dispersion analysis, inversion analysis and visual display. Meanwhile, the high-precision and multimodal frequency dispersion analysis andinversion analysis method is adopted so that the computing accuracy is high and the effect is great. Besides, the analysis result of each stage is visually outputted so that result interpretation isconvenient and visual.

The invention discloses a high-precision temperature measurement circuit with a compensation function and belongs to the technical field of power electronics. The high-precision temperature measurement circuit comprises a thermistor signal processing circuit group for measurement, a thermistor signal processing circuit group for compensation and a temperature compensation circuit group. The thermistor signal processing circuit group for measurement comprises a thermistor for measurement, a constant current circuit, a follower circuit and a subtraction circuit; the thermistor signal processingcircuit group for compensation comprises a thermistor for compensation, a constant current circuit, a follower circuit and a subtraction circuit. Therefore, the influence of thermal resistance from atemperature measurement point to the thermistor body for measurement on processing errors can be weakened and the temperature value of the temperature measuring point can be obtained more accurately;meanwhile, an adjustable voltage division circuit is adopted, and adjustment can be conducted according to the actual application working condition, so that the application range is widened; the constant-current circuit is applied to provide power for the thermistor, and subsequent linear electric signal processing is facilitated.

The invention discloses an integrated leveling construction method for a translation type hatch cover system of a large ocean-going bulk cargo ship. The integrated leveling construction method comprises the following steps: 1, manufacturing a special jig frame on a foundation flat ground; 2, splicing a hatch coaming steel structure on the special jig frame, 3, pre-assembling a hydraulic power pipeline and a cable system on the foundation flat ground, 4, installing the hatch cover on the hatch cover steel structure; 5, installing a reinforcing plate on the hatch cover system, hoisting the hatchcover system to a deck to be assembled, assembled and welded, and fixing a pre-assembled hydraulic power pipeline and a cable system to the hatch cover system and the deck through clamping sleeves. The construction environment coefficient, the construction safety coefficient, the construction production efficiency and the productivity are all improved. A great effect is achieved on preassembly maximization of modern shipbuilding enterprises, the shipbuilding period is shortened, the process precision is guaranteed, and the requirements of ship inspection and shipowners are met.

The invention provides an image processing method based on a neural network system and the neural network system, and the method comprises the steps: enabling the neural network system to identify a current input image frame, and judging whether the current input image frame is a basic image frame or not; if the basic image frame is the basic image frame, performing calculation processing by the multi-layer convolutional layer of the neural network; if not, determining that the image is a basic image frame; if so, comparing the image data with the image data of the basic image frame to extractimage data different from the basic image frame; and performing calculation processing on the basis of the image data by the plurality of convolution layers, and finally, replacing the image data ofthe corresponding position of the intermediate complete data stored in the layer with the data calculated and output by the last convolution layer of the neural network by the last convolution layer of the neural network to obtain the image data of the new complete image. According to the method provided by the invention, when the neural network system processes data, unchanged information in front and back frames can be hidden, and only the changed information is processed, so that a large amount of computing resources can be saved, and the processing speed is increased.

The invention relates to the technical field of artificial intelligence, and discloses a self-adaptive window width and window level adjustment method and device, a computer system and a storage medium, and the method comprises the steps: extracting the gray value of each pixel in a to-be-adjusted image, and collecting the gray values to obtain an input feature vector; calculating a truncation adjustment coefficient of each gray value in the input feature vector through a derivable truncation model, summarizing the truncation adjustment coefficients to form a truncation adjustment vector, andadjusting the input feature vector according to the truncation adjustment vector to generate an output feature vector; and sending the output feature vector to a preset neural network, updating the weight of the derivable truncation model by the neural network according to the output feature vector to generate an output feature vector conforming to a neural network loss function, and generating awindow width and window level image according to the output feature vector. The window width and window level image obtained by the invention not only meets the requirement of a user for adjusting thewindow width and window level, but also meets the requirement of a neural network for processing or classifying the window width and window level.

The invention provides an automatic impedance matcher for a magnetostrictive transducer. The automatic impedance matcher comprises an impedance matching module, a signal collection module, a feedbackmodule, and a single-chip microcomputer control module; the impedance matching module comprises two sets of 8-channel 5V relay modules with optocoupler isolation functions, a 5V DC power supply, 16 CBB matching capacitors and 16 resistors with resistance of 1kohm; the two sets of relay modules have the same structure and respectively comprise 8 relays connected in parallel; DC + and DC- Ports of each set of relay modules are connected to a positive electrode c and a negative electrode d of the 5V DC power supply respectively; and an assorted H jumper cap is respectively inserted into each of high and low level trigger selection ends S1 to S16 of each relay. The automatic impedance matcher provided by the invention enables the transducer to continuously and stably work in a resonant state under different environments, working conditions and loads, thereby significantly improving the working efficiency of the transducer.

The invention discloses an interference fringe image splicing method of inclined entering type laser interferometry measurement. The interference fringe image splicing method is characterized in that a helical gear tooth flank is used as a measurement object, and a measured surface is equally divided into W parts in an X axis direction, and is equally divided into H parts in a Y axis direction, and therefore the measured surface is divided into (W+1)*(H+1) grid points; any one grid point is marked to be G (h,w), h=0,1,..,H, w=0,1,..,W, and the space coordinate is marked as D (h,w); the normal direction of any one grid point G (h,w) is marked as n (h,w), and space positions of measured curve surfaces are used as splicing references, and registration between various interference fringe images and the measured curve surfaces is carried out; determining is carried out based on the processing reliability of the widths of the fringes; a changing coordinate system model is established; the resampling of the shape errors of the measured curve surfaces is carried out to complete the splicing of the interference fringe images. The splicing method provided by the invention is capable of realizing the splicing and the reconstructing of the shape error information of the various interference fringe images having main optical axes on different positions.

The invention discloses a heat storage burning system. The system comprises a heat supply system, a combustion supporting solvent system, a heat storage oxidization system, a gas inlet system, an exhaust system and a smoke spraying system, wherein the heat supply system, the combustion supporting solvent system and the gas inlet system are connected to an inlet of the heat storage oxidization system in series, the exhaust system is connected to an outlet of the heat storage oxidization system in series and connected with the smoke spraying system, a diesel combined heat supply system preheats a heat storage chamber in the heat storage oxidization system, waste gas enters the heat storage chamber through the gas inlet system to be heated and then enters an oxidization chamber in the heat storage oxidization system to be subjected to oxygenolysis, and smoke passes through the exhaust system and is exhausted to the atmosphere after acid gas in the smoke is removed through the smoke spraying system. According to the heat storage burning system, the waste gas is subjected to heat storage oxidization through automatic alternate switching, time is saved, efficiency is improved, heat emission generated after waste gas treatment is reduced, high heat efficiency is achieved, VOCs in the waste gas is high in decomposition rate, and the system is safe and reliable, easy to operate and low in running cost in use.

The invention relates to a detachable U-shaped corrugated belt double-layer inclined-slit steel plate damper, and belongs to the technical field of civil engineering energy dissipation and shock absorption. The metal damper is composed of a plurality of U-shaped energy dissipation parts, double-layer inclined-slit steel plates, an upper cover plate and a lower cover plate. The metal damper is based on the design concepts that multidirectional energy dissipation is achieved, energy dissipation parts are detachable, the shear bearing capacity is improved, out-of-plane buckling is restrained, multiple shear energy dissipation mechanisms work cooperatively, and a double-insurance mechanism is adopted, and the limitation of a traditional metal damper is broken through. U-shaped energy dissipation units are connected in parallel, so that the multidirectional energy dissipation of the damper is effectively realized. Through a reasonable design, the damper can achieve multi-position yield, theenergy dissipation performance is stable, and the shear bearing capacity is greatly improved. The U-shaped energy dissipation parts and the double-layer inclined-slit steel plates dissipate energy together to form the double-insurance mechanism. After an earthquake, the normal use function can be recovered only by replacing the energy dissipation parts. The metal damper develops towards the assembling direction, is easy and convenient to replace and obvious in economic benefit, can be effectively combined with a support, and provides lateral force for the support.

The invention discloses a dark room broadband RCD measurement method based on compression sensing. The method comprises steps of (1) designing a sparse measurement matrix, and configuring compression sampling testing parameters; (2) placing a scaling body and a measurement object on a rotary table in sequence, and completing echo data sampling of the measurement object, wherein the rotary table rotates at the uniform speed in the sampling process; (3) completing scaling processing to the measurement object according to the obtained sampling data; (4) subjecting object data after scaling to sparseness by the sparse measurement matrix; and (5) using a two-dimensional data reconstruction algorithm to perform data reconstruction of other angles and other frequencies. The method utilizes the coupling of the two-dimensional signal, and improves the imaging resolution under a precondition of the same data size.

The invention discloses an image model training method, an image processing method, a chip, equipment and a medium. The method comprises the steps of obtaining a sample image; wherein one sample imageis associated with one original network output; inputting the sample image into a preset neural network model containing the initial parameters, and performing symmetric quantization on the sample image to obtain a first quantization network output; determining a first cosine similarity between the first quantization network output and the original network output; performing optimization on the preset neural network model according to the first cosine similarity, a preset similarity threshold and a preset mixing precision quantification method, and obtaining preset network output of the preset neural network model after optimization; and determining a loss value between a preset network output and the original network output, and recording a preset neural network model as an image processing model when a convergence condition is preset for the loss value. According to the invention, the parameters of the preset neural network model are reduced, and the calculation rate of the preset neural network model is improved.

The invention relates to an image median filtering method based on an incomplete rapid sorting algorithm. The image median filtering method includes the following specific steps that according to a row-major sequence or a line-major sequence, all pixels of an image are scanned and all pixels in a certain neighbourhood of a target pixel of the image are obtained; an element in the sequence is randomly selected as a pivot point, assuming that the position of the pivot point after one-time scanning processing is changed to be m, all the elements which are in an image pixel sequence and are smaller than the pivot point value are moved to be in front of the pivot point and all the elements which are in the image pixel sequence and are larger than the pivot point value are moved to be in rear of the pivot point, and an original sequence is divided into two parts; the position of the pivot point is judged after rapid sorting, if the position of the pivot point is just the middle position of the sequence, a k value is obtained, the K value is used for replacing the value of the image target pixel, and the process is terminated; otherwise, according to the size relation between m and k, a proper subsequence is chosen to repeat the process. According to the image median filtering method based on the incomplete rapid sorting algorithm, a large amount of data sorting and time-consuming data moving operation is reduced and the processing speed is greatly improved when the processing precision is ensured.

The invention discloses a multi-parallel acceleration method for CNN full-connection layer operation, and the method greatly reduces the reading operation of full-connection layer parameters, effectively saves the energy consumption, and improves the energy efficiency ratio of a system through employing the sparsity characteristics of an operation result of a convolutional neural network and interpreting the numerical value of a convolution layer processing result of the convolutional neural network in advance. By utilizing the characteristics of hardware resource reuse, strong expansion capability and the like of programmable logic devices such as an FPGA and constructing a parallel pipeline multiply-accumulate architecture, the processing time is effectively saved, and the processing efficiency is improved; by comprehensively analyzing factors such as target type data, target characteristic difference and error tolerance in an application process, a processing data format is reasonably set, so that the access efficiency of data and parameters is effectively improved while the processing precision is ensured, and the purpose of multi-parallel acceleration of a full connection layer is achieved.

The invention discloses a 3D printing pretreatment hierarchical algorithm based on the intensive feature. The method is implemented according to the following steps that firstly, a datum tangent planeis determined through minimum height triangular patch traverse searching and determining on a 3D model in the STL file format; secondly, upward and downward layering is conducted with the datum tangent plane as the datum, a minimum height triangular patch in an intersection triangular patch set is determined, the height of the triangular patch serves as the reference to determine the layer heightof the layer, self-adaption layering is conducted, and the appearance precision of the 3D printing model is guaranteed; and thirdly, by parity of reasoning, all layering work of the 3D model is finished. According to the algorithm, the position where the minimum height triangular patch of the 3D model is located serves as the layering datum, downward and upward layering treatment is conducted, layering is conducted according to the self-adaptation layering rule, the layering height is automatically set according to the patch height and size, the printing efficiency is improved, and the precision of the printed 3D model can be guaranteed as well.

The embodiment of the invention discloses an image defogging method, a terminal and a computer storage medium, and the image defogging method comprises the steps: determining the size parameter of a to-be-processed image after the to-be-processed image is judged to be a fogged image; determining a preprocessing strategy corresponding to the to-be-processed image according to the size parameter, wherein the preprocessing strategy is used for limiting the image size; if the preprocessing strategy is to divide the to-be-processed image, dividing the to-be-processed image to obtain sub-images corresponding to the to-be-processed image; defogging the sub-images according to the image defogging model to obtain defogged sub-images corresponding to the sub-images; and splicing the defogged sub-images according to the image splicing model to obtain a defogged image corresponding to the to-be-processed image.

The invention relates to a TEC-based thermal infrared imager automatic correction method and device. The method comprises the steps of: in a calibration process of the thermal infrared imager, measuring a suspected pixel of the thermal infrared imager in each temperature range, adopting the suspected pixel existing in each temperature range as a dead point, and enabling the rest suspected pixels to serve as unstable pixels of each temperature point; in the using process, acquiring an image of the thermal infrared imager by a processing circuit, completing non-uniformity correction, acquiring the dead point position and acquiring a current temperature; searching an unstable pixel according to the current temperature; directly compensating the dead point position in the image; and judging whether the unstable pixel is a blind pixel or not, if so, performing compensation, and otherwise, not performing processing. According to the method, dead point and unstable pixel calibration is carried out on all temperature points, meanwhile, non-uniformity correction is completed, in the using process, blind pixel judgment does not need to be carried out on each image, corresponding compensationmethods are directly adopted for dead points and unstable pixels for compensation, processing is easy, and the real-time performance is good.

The application provides an information processing method, device and system, which are applied to at least one physical examination device. The information processing method comprises the steps of receiving physical examination information which are transmitted by each physical examination device and corresponds to identity information, carried in an information processing command, of people subjected to physical examination by responding to the received information processing command transmitted by an identity card reader, and storing the received identity information of people subjected to physical examination and the physical examination information so as to realize the automatic measurement and storage of the identity information of people subjected to physical examination and the physical examination information corresponding to the identity information of people subjected to physical examination, so that the processing flow is simplified, and the effective storage of the information is realized on the basis of guaranteeing the processing precision in the physical examination information processing flow.

The invention discloses a method for preparing Pu'er raw tea through 39 procedures in the technical field of preparation methods of Pu'er raw tea, the quality of ancient tree tea is comprehensively screened, fresh leaves which have certain and uniform bud-leaf-stem proportion, complete buds and leaves, no sandwiched leaves or single leaves, basically consistent size and length and no impurities are adopted for picking fresh leaves, the high quality of the raw materials is ensured, then the fresh leaves are placed in a green leaf storage frame, the freshness of the fresh leaves is kept, the thickness of the stacked fresh leaves during spreading and airing is based on the fact that the bottom layers of the fresh leaves do not generate heat, green grass gas is volatilized, conversion of components contained in the fresh leaves is promoted, full-manual fixation is conducted, spreading and airing are conducted slightly after fixation, so that the temperature of the leaves is reduced, then sunlight drying and sunning treatment is conducted, the whole drying area must reach a pollution-free standard, so that the quality of raw tea raw material selection can be guaranteed, the process precision of raw tea preparation can be effectively guaranteed, the quality of raw tea preparation is improved, the yield of raw tea is effectively improved, and the quality of the Pu'er raw tea can be improved.

The invention relates to the technical field of stamping part machining, in particular to a refrigerator stamping part machining treatment method. The refrigerator stamping part machining treatment method comprises the following steps that firstly, discharging is conducted, metal plate raw materials are checked, after qualification is reached, a metal plate is cut according to the size of a refrigerator stamping part, and a to-be-stamped material is obtained; secondly, first-time edge trimming is conducted; thirdly, forming is conducted; and fourthly, second-time edge trimming is conducted. Inthe preparing process, the two-time edge trimming treatment is sequentially conducted on the stamping part; during first-time edge trimming treatment, the situation that fixing and stamping of the product are influenced due to burrs and other defects on the cut edge is avoided, and the precision of the subsequent machining technology is guaranteed; and during second-time edge trimming, the appearance and the product precision of the stamping part can be guaranteed. Before stamping, air blowing and dust removal are conducted on a stamping cavity and the to-be-stamped plate, the technology precision is further guaranteed, heating softening treatment is conducted before stamping, stress in the material can be eliminated, the phenomena of cracking and the like after stamping part forming areavoided, and the forming quality is reliable.

The invention relates to the technical field of computer application and discloses an information fixed-point processing device. The information fixed-point processing device comprises an output controller, a plurality of groups of fixed-point processing cores which are connected in series, and corresponding interconnection interfaces, wherein the fixed-point processing cores control functional modules, which process data information, are connected with the interconnection interfaces and transmit fixed-point processing information at the same time; the interconnection interfaces are connectedwith the fixed-point processing cores and different functional modules and transmit data and the fixed-point processing information among modules; and the output controller is connected with the interconnection interfaces and receives the output data and the fixed-point processing information for detecting precision and determining to output the results or continue the processing after a processing formula is corrected. The information fixed-point processing device has higher processing efficiency and processing precision.

The invention discloses a micro-expression recognition method and device, a computer device and a storage medium, and the method comprises the steps: firstly, judging whether a face image of a testedperson exists in an acquired current image to be recognized or not; if the face image of the tested person exists in the current image to be recognized, performing micro expression recognition on thecurrent image to be recognized based on a micro expression recognition model, and outputting an emotion recognition result of the current image to be recognized; if the face image of the tested persondoes not exist in the current image to be recognized, updating a dynamic threshold value; and obtaining the next image to be identified according to the updated dynamic threshold, taking the next image to be identified as the current image to be identified, and returning to execute the step of judging whether the face image of the tested person exists in the current image to be identified. And the calculation amount of the server side can be reduced while the processing precision is ensured, and the overall processing efficiency is improved.

The invention provides an audio processing method and device. The audio processing method comprises the following steps: acquiring a to-be-processed audio; preprocessing the to-be-processed audio to obtain a global frequency domain frame sequence corresponding to the to-be-processed audio; constructing an audio feature set corresponding to the to-be-processed audio according to the global frequency domain frame sequence, and inputting the audio feature set into an audio processing model for processing to obtain a local frequency domain frame sequence; and generating a target audio for replacing the to-be-processed audio based on the local frequency domain frame sequence and the global frequency domain frame sequence, thereby realizing accurate compensation of the to-be-processed audio, and effectively improving compensation efficiency and compensation accuracy.