539 results about "Prior information" patented technology

The present invention relates to the technical field of coal mine water inrush disaster prevention and control, and relates to an artificial intelligence-based method for identifying a water inrush point in a mine and parameters of a simulation model, comprising the following steps: step 1: establishing a numerical model on the basis of observation data, and determining prior information of parameters to be identified including position coordinates of a water inrush point; step 2: on the basis of the numerical model and the prior information of the parameters, generating a training sample dataset and a test sample dataset for an alternative model; step 3: constructing and training an alternative model neural network; step 4: testing the accuracy of the alternative model; and step 5: executing a simulated annealing algorithm to identify the position of the water inrush point and parameters of a simulation model. By using the method provided by the present invention, the problem in accurate positioning of water inrush points in underground coal mines can be solved, which is of great significance in carrying out scientific disaster management and rescue efforts in a timely manner.

The present disclosure provides an artificial intelligence-based method for enhancing mine safety by identifying and predicting locations of water inrush points in a mine, including the following steps: S1: constructing a numerical model to determine priori information of parameters to be recognized based on observation data, including coordinates of locations of water inrush points; S2: generating a training sample dataset and a test sample dataset of an alternative model based on the numerical model and the priori information of the parameters; S3: constructing and training a neural network of the alternative model; S4: testing an accuracy of the alternative model; and S5: performing a simulated annealing algorithm to identify the locations of water inrush points and simulation model parameters.

The invention discloses an electromechanical equipment abnormal behavior detection and fault prediction method based on a causal time-space Transform, and belongs to the technical field of electromechanical equipment abnormal detection, and the method comprises the steps: S1, constructing an initial causal graph according to the causal relationship between the physical structure and the functional part of electromechanical equipment; s2, correcting the initial causal graph based on historical operation data of the electromechanical equipment to generate a causal graph; s3, introducing the causal graph as prior information into a Transform model, predicting sensor time sequence data of the electromechanical equipment to be detected through the trained Transform model, and outputting corresponding high-dimensional feature representation; and S4, according to the high-dimensional feature representation, calculating the abnormal weight of each component, and carrying out component-level abnormal identification and fault prediction. The method breaks through the limitation that only the data correlation is fitted and the causal relationship is ignored in the electromechanical equipment anomaly detection of a traditional time sequence model, and explicit modeling of an equipment fault chain propagation mechanism is realized by fusing a causal reasoning mechanism and feature modeling.

The invention discloses a document table extraction method and device, equipment and a medium, and relates to the technical field of computer information processing. The extraction method comprises the following steps: performing OCR (Optical Character Recognition) on a to-be-processed document table image to obtain a text block; performing visual feature coding on the document table image to obtain deep visual features; performing semantic feature coding on the text sequence of the text block to obtain a semantic feature vector; performing spatial feature coding on the bounding box of the text block to obtain a spatial feature vector; performing feature fusion processing on the deep visual features, the semantic feature vectors and the spatial feature vectors to obtain multi-modal guide features; and performing structured decoding processing on the multi-modal guide features to obtain structured representation of the table. According to the method, the text and the position information pre-recognized by the OCR are fused with the visual features of the document table, so that the visual features are guided to be expressed again and are actively aligned to the logic structure defined by the prior information, and the extraction accuracy of the table logic structure is improved.

The invention discloses a large-model-driven cross-modal power insulator defect detection method and system, and the method comprises the steps: constructing an image-text dual-channel feature learning frame based on a pre-trained vision-language large model, introducing fine-grained semantic description as prior information, and carrying out the detection of the defects of a cross-modal power insulator. Feature enhancement and space focusing under semantic guidance are realized through a cross-modal attention fusion mechanism, and the robustness and discrimination ability of the model in a complex scene are further improved in combination with a multi-scale pyramid structure and a channel re-calibration mechanism. The method can effectively solve the practical problems of difficult recognition of tiny targets, strong environmental interference, limited sample data and the like.

The invention relates to a vehicle and unmanned aerial vehicle combined scheduling method for wide-range low-cost inspection. The method comprises the following steps: acquiring prior information; modeling the unmanned aerial vehicle inspection problem of each target area according to the prior information to obtain a mixed integer non-convex optimization problem with the goal of minimizing the weighted sum of the total execution time and the energy consumption of all the inspection unmanned aerial vehicles; performing linearization on a non-convex bilinear term in the mixed integer non-convex optimization problem, and performing discretization processing on a nonlinear function by adopting piecewise linear approximation; an approximate mixed integer linear programming problem is obtained and solved, and an unmanned aerial vehicle scheduling strategy is obtained; modeling according to the unmanned aerial vehicle scheduling strategy and the prior information to obtain an inspection vehicle path planning problem taking the comprehensive driving cost as a target; the routing inspection vehicle path planning problem is converted and modeled into a Markov decision process, a routing inspection vehicle is used as an intelligent agent, a state, an action and a reward function are defined, and a routing inspection vehicle scheduling strategy is obtained. Therefore, combined inspection of the inspection vehicle and the unmanned aerial vehicle is realized, and the inspection range is expanded.

The invention discloses an interpretable building energy consumption prediction method and system based on a large language model, and belongs to the technical field of building electric power energy consumption prediction. Comprising the following steps: acquiring and processing prior information of building energy consumption prediction, and generating prefix vector representation; energy consumption time series data of the building are obtained, reversible instance normalization processing and patch embedding operation are sequentially carried out on the energy consumption time series data, reediting is carried out in combination with the text prototype vector representation, and embedded vector representation is generated; splicing the prefix vector representation and the embedded vector representation, and performing feature processing on a splicing result through a large language model to obtain a multi-dimensional vector representation; and performing mode conversion on the multi-dimensional vector representation and mapping the multi-dimensional vector representation to an output space to obtain an energy consumption prediction result. The interpretability and the prediction efficiency of building energy consumption prediction are improved, and the problems that existing energy consumption prediction is limited by complex influence factors, the energy consumption prediction time is long, and the result lacks interpretability are solved.

The invention discloses a coordinate system transformation fusion filtering tracking method and system for a dual-base-station radar, and relates to the technical field of radar target tracking. The method comprises the following steps: firstly, establishing a state and observation equation, and determining an initial state of a target based on prior information under Cartesian coordinates; then, obtaining a target motion state through one-step prediction, generating a sigma point by utilizing U transformation, and calculating a mean value and a covariance for constructing state space prediction; thirdly, fusing the measurement data and the prediction data by using a Kalman filter to obtain optimal state estimation and a covariance matrix; and finally, converting the updated state estimation back to the Cartesian coordinate system based on U transformation, and circulating the process until the tracking is finished. According to the method, the nonlinear filtering problem in the updating process can be avoided, and the tracking robustness and precision are improved.

The invention discloses a discrete wavelet steel rail damage acoustic emission signal detection method based on filtering parameter adaptive search, which comprises the following steps of: firstly, preprocessing acoustic emission data acquired by a sensor, and removing a pure noise segment in the signal to relieve the subsequent operation pressure; then, an evaluation function combining a reduction degree index and a sparseness index is designed, an evaluation index of a filtering parameter represented by each sparrow is calculated in combination with discrete wavelets, and a filtering parameter enabling the evaluation index to be minimum is searched by adopting a self-adaptive sparrow search method, namely an optimal filtering parameter group; and finally, filtering the original signal based on the optimal filtering parameter group, extracting a high-frequency component, obtaining the time-frequency characteristic of the high-frequency component by adopting short-time Fourier transform, and proposing a steel rail damage judgment condition based on the dual-band time-frequency characteristic, which can judge whether the damage exists or not through the time-frequency characteristic and calculate the moment when the damage is captured by a sensor. The problem that a traditional method generally depends on system parameters and prior information is solved.

The invention discloses a generative intention recognition method based on prior information and related equipment, and relates to the technical field of intention recognition, the method comprises the following steps: obtaining entity data and user verbal skill, the entity data comprising entity data in a navigation multimedia domain and entity data in a cabin domain; performing feature extraction on the user verbal skill to obtain a first user verbal skill vector; performing feature extraction on the entity data in the navigation multimedia domain based on the first user verbal skill vector to obtain an entity vector in the navigation multimedia domain; performing feature enhancement on the entity data in the cockpit domain to obtain an entity vector in the cockpit domain; and performing intention recognition based on the first user verbal skill vector, the entity vector in the navigation multimedia domain and the entity vector in the cabin domain to obtain an intention recognition result.

The invention relates to the technical field of intelligent control and information security crossing, in particular to an anti-error data injection Kalman filtering method based on neural network enhancement, which comprises the following steps: firstly, constructing a neural network auxiliary correction module for identifying and correcting an extended Kalman filtering intermediate variable affected by an error data injection attack; secondly, designing a hybrid filtering architecture, and combining the learning ability of a neural network with the theoretical advantages of traditional Kalman filtering; and finally, realizing adaptive optimization of model parameters through a mixed training mechanism. Accurate state estimation can be achieved only through limited prior information, and meanwhile the calculation efficiency and interpretability of a traditional Kalman filter are kept. And the anti-interference capability of the system in a non-Gaussian noise environment is effectively improved. The application of the method on an induction motor model is obviously better than that of a traditional anti-error data injection scheme, and breakthrough progress is achieved in the aspects of estimation precision and robustness.

The embodiment of the invention provides an underground tunnel geometric parameter remote sensing inversion method, and belongs to the technical field of data processing, and the method specifically comprises the steps: building a sensitivity enhancement model corresponding to ground surface displacement caused by underground tunnel excavation in combination with an empirical model and a physical model; an InSAR geometric projection equation is introduced, and a function relation between underground tunnel geometric parameters and InSAR line-of-sight deformation is constructed; the method comprises the following steps: collecting time sequence I nSAR data of an underground tunnel in an excavation period, and measuring accumulative sight deformation of a research area according to the time sequence I nSAR data; based on the accumulated line-of-sight deformation, extracting tunnel deformation center positions in a plurality of cross sections and surface line-of-sight deformation of the cross section positions by using an image segmentation and morphological algorithm; and optimizing and constraining a parameter space based on known prior information, substituting the earth surface sight line direction deformation into the function relationship, and performing optimal solution by using a genetic algorithm to obtain other geometric parameters of the underground tunnel. Through the scheme of the invention, the reliability and adaptability of inversion are improved.

The invention discloses a two-stage checking method for the node water demand and the pipe section friction resistance coefficient of a water supply pipe network, and is applied to the technical field of real-time online correction of a water supply pipe network model. Comprising the following steps: grouping node water demand and a pipe section friction resistance coefficient based on prior information, calculating a covariance matrix of pressure and flow measuring points, calculating estimated values and covariances of the node water demand and the pipe section friction resistance coefficient through a first-stage sequential data assimilation algorithm according to monitoring data at the current moment, and calculating the node water demand and the pipe section friction resistance coefficient through a second-stage sequential data assimilation algorithm; and then a calculation result serves as an empirical value to be added into a target function of second-stage iterative calculation, an estimated value serves as an initial value of an iterative algorithm to be subjected to fine tuning optimization, an optimal estimated value of parameters at the current moment and a covariance matrix of the optimal estimated value are obtained, the calculation result is brought into the next moment, and the two-stage algorithm is repeated to check model parameters at the next moment. The advantages of the two-stage algorithm complement each other, the checking precision of the parameters is improved, and a new thought is provided for real-time online correction of the water supply pipe network model parameters.

The invention discloses a unified representation and dynamic knowledge reasoning method and device for an intelligent cluster system, and the method comprises the steps: determining a decyclization task sub-graph based on a composite task state space model of the intelligent cluster system through the priori information; constructing a real-time data sub-graph by using a real-time data feedback mechanism, and constructing a traceability mode sub-graph based on the decyclization task sub-graph; candidate traceability sub-graphs with the highest matching similarity with the traceability mode sub-graphs are determined from the real-time data sub-graphs to serve as traceability sub-graphs; constructing a reasoning path set based on the decyclization task subgraph, and calculating a probability value of each reasoning path in the reasoning path set by using a path probability model constructed based on a Bayesian conditional probability chain rule; and screening out the reasoning paths of which the probability values are smaller than a preset probability value in the reasoning path set by utilizing a dual path pruning strategy, and adding the reasoning paths containing abnormal nodes in the reasoning path set into an alarm path set so as to provide support for state prediction and risk management of the intelligent cluster system.

The invention discloses a multi-satellite direct positioning method based on CUDA (Compute Unified Device Architecture) parallel calculation, relates to the field of space-based electromagnetic signal positioning, and aims to be suitable for large-range and high-precision positioning of low-signal-to-noise-ratio signals. According to the technical key points, a positioning area is determined in a satellite coverage range according to prior information, and grid points are divided on the surface of the earth; two or more satellites with time-frequency synchronization obtain observed quantities of signals emitted by the same ground radiation source, a first signal model of the multiple observed quantities is constructed, and down-conversion is carried out on the first signal to obtain a second signal; constructing a cost function by using signal characteristics (such as time difference of arrival, frequency difference of arrival and the like) in a radiation source signal received by a satellite; a cost function corresponding to each grid point is subjected to multi-core parallel computing through the GPU; and summarizing results, and searching the maximum value of the cost function to determine the estimated position of the ground radiation source. According to the multi-satellite direct positioning method based on the CUDA, the characteristic that parallel calculation can be carried out when cost function grid searching optimization is carried out in direct positioning is utilized, calculation is directly carried out on satellites, satellite-ground data transmission is omitted, the positioning time is shortened by utilizing the independence of searching calculation among grids, and the real-time performance of a positioning system is guaranteed.

The invention discloses a physical information neural network hyperspectral band selection method for target identification, which uses physical information SDI as prior information to be combined with a channel attention mechanism added in a deep neural network to guide band selection, and evaluates the advantages and disadvantages of band weights through a reconstruction module after the weights are generated. Wave band selection is carried out accordingly; according to the hyperspectral image target recognition waveband selection method based on the physical information neural network, compared with a traditional method, the process is simplified and the precision is improved through end-to-end learning, the utilization of spatial context information can be enhanced by adding an attention mechanism, and the model has interpretability by adding prior SDI information.

The invention relates to a circuit board defect identification method and system based on multi-dimensional image data, and belongs to the technical field of data identification processing, and the method comprises the following steps: obtaining synchronous image data of a circuit board to be detected in a plurality of imaging modes; performing space-spectrum joint registration on each modal image to generate a multi-dimensional image cube with a unified coordinate system and pixel alignment; inputting the multi-dimensional image cube into a pre-trained multi-branch heterogeneous fusion neural network; generating a pixel-level defect probability graph by utilizing a defect sensing context decoder, and performing geometric constraint optimization on the probability graph by combining prior information of a circuit board design layout; outputting defect types, positions and confidence coefficients, and establishing an interpretable defect fingerprint database according to the multi-dimensional response characteristics of the defects; the method has the beneficial effects that false defect signals generated by image noise and circuit board surface texture interference can be effectively inhibited, the omission ratio and the false detection ratio are greatly reduced, and pixel-level accurate defect positioning is realized.

The invention discloses a ground penetrating radar pipeline intelligent inversion method based on multi-scale deep learning, and belongs to the technical field of ground penetrating radar pipeline data inversion. The method aims at solving the problems that a traditional inversion method is low in resolution, high in calculation complexity, depends on prior information and the like. The method comprises the specific steps that underground simulation model data sets are constructed in batches; forward modeling calculation is carried out based on time domain finite difference, and a matching data pair of the radar B-scan image and dielectric constant distribution is generated; constructing a multi-scale deep neural network comprising a residual attention module, a dual-path space attention module, a multi-scale feature fusion module and a reconstruction module; through end-to-end training and model test optimization, an optimal model is finally stored to realize inversion output from original radar data to dielectric constant distribution. According to the method, through a multi-scale feature extraction and fusion mechanism, the recognition precision of pipelines of different sizes is effectively improved, the inversion efficiency is greatly improved, and an efficient and accurate technical scheme is provided for underground pipeline detection.

The invention discloses a target positioning method and device, a storage medium and a program product, relates to the technical field of communication perception integration, and discloses a target positioning method, the target positioning method is applied to a communication base station, and comprises the following steps: sending a perception signal to at least one target object; receiving an echo signal of the sensing signal reflected by the at least one target object; obtaining prior information of the at least one target object, and processing the echo signal based on the prior information to obtain an angle measurement value and a distance measurement value of the at least one target object; and determining positioning information of the at least one target object according to the angle measurement value and the distance measurement value. According to the technical scheme, it can be guaranteed that multiple targets are accurately positioned, meanwhile, the calculation complexity of a positioning information measurement scheme is reduced, and better balance between positioning performance and complexity is achieved.

The invention relates to the technical field of computer vision, and discloses a precise medicine delivery method and system, and the method effectively strips illumination artifacts through the construction of a local gradient structure tensor field and an anisotropic screening mechanism, remarkably improves the robustness of a visual front end in a light and dark alternating environment, and improves the accuracy of medicine delivery. Rigid body motion constraint and Lie algebra continuous modeling are utilized, high-precision space-time alignment between heterogeneous sensors is realized on the premise that scene depth does not need to be recovered, hardware delay errors are eliminated, a physical-driven probability weight dynamic allocation mechanism is constructed by introducing a multipath scattering index and a structural information entropy flux, and a dynamic space-time alignment algorithm is established. Environment degradation is sensed in real time, the observation weight is adjusted in a self-adaptive mode, the influence of the non-line-of-sight multipath effect and visual texture missing is effectively restrained, high-precision inertial dead reckoning can still be maintained in a blind area where a sensor is in full failure in combination with momentum prior information generated through Schel complement marginalization operation, and the method has the advantages of being high in precision and high in precision. And the navigation continuity and reliability of the distribution robot in a complex scene are ensured.

The invention relates to the technical field of artificial intelligence, and discloses an unmanned aerial vehicle flight control-oriented intelligent obstacle recognition method, which comprises the following steps of constructing an adaptive wind disturbance fuzzy kernel based on unmanned aerial vehicle attitude data, and synthesizing a degraded image for training; a specific obstacle recognition model integrating a frequency domain and space domain joint feature decomposition module, a turbulence invariant feature enhancement module guided by physical prior information and a double-branch anti-fuzzy feature extraction network is adopted, and a fuzzy robustness contrast loss function is combined to carry out progressive training so as to learn feature representation insensitive to fuzziness; after the trained model is deployed, the wind disturbance fuzzy intensity of a real-time input image is estimated on line, the internal workflow of the model is dynamically adjusted according to the wind disturbance fuzzy intensity, the attention weight is adaptively adjusted, and the output of a trunk or an anti-fuzzy branch is selected. According to the invention, through coupling of the physical model and deep learning, the recognition robustness, accuracy and stability of the model in a dynamic wind disturbance environment are improved.

The invention discloses a reactor core neutron field ex-situ explicit inversion method based on a graph structure, and belongs to the technical field of reactor core safety monitoring and control. According to the method, reactor core space coordinates, energy spectrum grouping, material physical properties and prior information are fused through an explicit graph structure, a lightweight graph convolutional network architecture is adopted, and accurate inversion of a high-dimensional reactor core neutron field is achieved. And meanwhile, the inversion precision under the high-dimensional resolution can be remarkably improved, and reliable technical support can be provided for safe operation and intelligent monitoring of an advanced reactor type. Compared with an existing deep learning black box model, the method has the advantages that the occurrence probability of local high deviation can be remarkably reduced at different reactor core change positions, and more uniform and stable prediction performance can be realized in the whole space. And under the condition that the reactor core has multi-region change, better inversion performance is still shown. When the middle and low complexity changes, the fine multi-region features of the reactor core can be effectively and stably captured, and the method has better overall error suppression capability.

The invention discloses a deep extension VAE soft measurement method based on instant learning, and belongs to the technical field of industrial process control. In the modeling process, DE-VAE is formed by stacking multiple layers of E-VAE, the output of the E-VAE comprises reconstructed original input data, expected output data and hidden variables of the previous E-VAE, correlation between extracted features and the original input and output is enhanced by multiplexing prior information, meanwhile, quality variables are introduced in the pre-training process, and the quality of the original input and output is improved. And the extraction capability of the model on target related features is enhanced. In the online prediction stage, the model is updated by using instant learning, a novel modeling method different from the existing soft measurement modeling method is designed, and the prediction precision of the soft measurement model is well improved.

The invention relates to a radar target recognition method and device under main lobe jamming based on prior information assistance, and the method comprises the steps: constructing a point distribution function under jamming capable of depicting jamming characteristics and an HRRP distortion rule, effectively extracting jamming prior information, and combining a multi-scale feature extraction network layer and a cross attention fusion mechanism to recognize a radar target under main lobe jamming. Deep fusion of HRRP features and interference prior information is realized, so that the target recognition accuracy and robustness of the model under typical interference conditions such as intermittent sampling and forwarding are remarkably improved. Therefore, the problems that in the prior art, under the typical interference condition, the radar target high-resolution range profile recognition robustness is poor, and the recognition accuracy is reduced are solved.

To accurately estimate, without prior information, communications related to changes in values of transmitted data of a target type from communication logs between multiple devices.SOLUTION: An analysis system is configured to retain a communication log that shows a time series of combinations of data types and values transmitted between multiple devices, specifically at the points where the values change, extract multiple logs, from the communication logs, in a period between the time corresponding to a first value of the target type of data and the time corresponding to a second value, extract communication sequences indicating one or more combinations of data types and values, and the order of occurrence of change points corresponding to each of the one or more combinations, from the multiple logs, and determine candidate sequences related to the change from the first value to the second value of the target type of data on the basis of the occurrence status of the extracted communication sequences in the multiple logs.SELECTED DRAWING: Figure 1

The invention discloses a long-tail incremental image recognition method and system based on language guidance and a medium. The method comprises the following steps: constructing a long-tail incremental image recognition model; obtaining multi-batch long-tail distribution training data; a fixed language template is designed, and corresponding category text description is generated through a large language model according to category labels of new category images in each batch of long-tail distribution training data; inputting multiple batches of long-tail distribution training data and corresponding category text descriptions into a long-tail incremental image recognition model in batches, and training by adopting a two-stage training mode to obtain a trained long-tail incremental image recognition model; and inputting the to-be-recognized long-tail distribution data into the trained long-tail incremental image recognition model to obtain a recognition result. According to the method, by introducing the language prior information, the problems of new data few sample category overfitting and old task knowledge disastrous forgetting in a long-tail category incremental learning scene can be solved at the same time, the performance of the model is improved, and the accuracy of long-tail incremental image recognition is enhanced.

An online system uses a pose prior model and a pose objective function to estimate the pose of a client device. A pose prior model is a model for prior information known about client devices and their poses without reference to a particular client device and its pose data. The online system receives pose data from a client device and computes an estimated pose for the client device based on the received pose data, the pose prior model, and a generated initial candidate pose for the client device. The online system uses these as inputs to a pose objective function and optimizes the pose objective function to estimate a pose for the client device. The online system transmits this estimated pose to the client device, and may use the estimated pose as the pose for the client device for the purposes of delivering content to the user.

The invention discloses a frequency offset estimation method and device, electronic equipment and a storage medium, and relates to the technical field of communication. The method comprises the following steps: receiving a baseband spread spectrum signal from a low earth orbit satellite to obtain a received signal; performing segmented sweep frequency compensation on the received signal to obtain a compensated measured signal; despreading and descrambling the compensated measured signal by using prior information determined from the received signal, and completing coarse frequency offset estimation on the measured signal despread and descrambled for the first time; and de-spreading and descrambling the signal data after the coarse frequency offset estimation again, and finishing fine frequency offset estimation on the signal data after the second time of de-spreading and descrambling. According to the invention, prior information is introduced to directly lock and track a specified target beam signal, so that the directionality and efficiency of measurement are remarkably improved; by designing a low-complexity signal processing flow, the algorithm burden can be remarkably reduced, it is ensured that adjacent satellite beam frequency offset estimation is rapidly completed in service proceeding, and the real-time requirement of seamless switching is met.

The invention discloses a road intelligent diagnosis method based on an AI large model, and relates to the technical field of road engineering detection, and the method comprises the following steps: building a unified time base line, constructing a phase reference field, injecting a calibration pulse, separating dynamic shadow time sequence texture and crack initial features in a road image sequence, and obtaining a road image sequence; generating prior information of dynamic shadows and cracks; under the constraint of prior information, polarization spectrum three-dimensional coding imaging is carried out, polarization angle distribution and spectrum gradient are extracted, a feature dictionary of dynamic shadows and cracks is constructed, and phase drift in the prior information is calibrated. According to the method, a unified time base line and a phase reference field are constructed, and accurate distinguishing of dynamic shadows and cracks is realized by combining polarization spectrum imaging and anti-fact distillation; the point cloud and acceleration data are fused for geometric reconstruction, and through phase conjugate extinguishing and closed-loop feedback, the continuity and accuracy of diagnosis are significantly improved, and maintenance error decisions caused by misjudgment are avoided.

The invention relates to the technical field of image processing, and discloses a speckle structured light binocular stereo matching method and system based on multi-cost fusion, a computer readable storage medium and a computer program product. The method comprises the following steps: respectively carrying out epipolar correction on an obtained left camera image and an obtained right camera image; performing boundary segmentation on the corrected left camera image to obtain a corresponding boundary image; calculating a Census cost and a BT cost by using the corrected left camera image and right camera image, and carrying out weighted fusion to obtain a matched cost body; performing cost aggregation on the obtained cost body by using the boundary image as prior information; and calculating a parallax image corresponding to the left camera image through the aggregated cost body. According to the invention, abnormal matching points in a speckle image stereo matching process are reduced, and the real-time performance and robustness of speckle image stereo matching are effectively improved.