32results about How to "Well represented" patented technology

The invention relates to the technical field of federated learning, and discloses a discriminator knowledge distillation personalized federated learning method based on a conditional generative adversarial network. The problems of appearance of massive participants and influence of non-independent identical distribution of data on federal learning data cost and personalized learning efficiency are solved; a discriminator knowledge distillation model fused with the conditional generative adversarial network is put forward to be embedded into a personalized federated learning framework; the differentiated feature representation of the client model is enhanced through the discriminator knowledge distillation model, so that the personalized model effectively recognizes and captures specific modes and features of individuals, the representation capability and the personalized effect of the model are enhanced, the higher model compression rate is achieved, and meanwhile, the user experience is improved. The conditional generative adversarial network fused with the discriminator generates synthetic data specific to each client, and the synthetic data improves the generalization ability of personalized federal learning and the adaptability of the model on the premise of fully ensuring the data privacy.

The invention discloses a user abnormal load identification method based on a stereo portrait library and dual-channel weighting, and the method comprises the steps: collecting the load data and industry labels of a user, cleaning the data to obtain a load data set, and dividing a training data set and a test data set; a user load characteristic analysis module is constructed, the unbalance degree between different users and different industry types in the training data set is counted, and a user multi-time-scale typical power consumption curve and a multi-dimensional load characteristic index are extracted in a clustering mode; combining a user load characteristic analysis result with an industry label, performing weighting on each industry user, performing secondary clustering to extract an industry multi-time scale typical power consumption curve and a multi-dimensional load characteristic index, and constructing a three-dimensional portrait library; and after load curve-feature dual-channel weighting is carried out on each industry in the stereo portrait library, load type and abnormal power consumption behavior identification is carried out on unknown users in the test data set. According to the invention, high-accuracy identification of the user load type and the abnormal power consumption behavior is realized.

The invention provides a cancer aided decision-making method based on knowledge-enhanced hypergraph comparative learning, and relates to the technical field of intelligent medical treatment, and the method comprises the steps: carrying out the feature extraction of original drug and cell line data, and constructing a training hypergraph with a drug and a cell line as vertexes and a cooperative relation as a hyperedge; a hypergraph neural network is used for learning, and embedded representation of the combined medicine and cell line is obtained; enhancing the hypergraph so as to carry out comparative learning on the model; performing feature extraction on a to-be-detected data set based on the learned model and updating the hypergraph to obtain an embedded representation of a to-be-detected vertex; and based on the embedding representation of the to-be-detected vertex, obtaining the cooperative relationship of the drugs in the to-be-detected data set. According to the method, the high-order dependency relationship between the medicine and the cell line is considered, good characterization is obtained and is better used for predicting the cooperative relationship, and a good basis is provided for clinical auxiliary decision making.

The invention provides an Iwan model modeling method and device of a bolted connection structure and electronic equipment, and the method comprises the steps: constructing a dynamic pressure distribution function of the bolted connection structure based on a contact boundary of a contact surface in the bolted connection structure by applying a linear pressure distribution function; solving a density function based on the dynamic pressure distribution function and a friction theory to obtain a density function of yield displacement; and introducing the density function of the yield displacement into a general Iwan model to construct an Iwan model. According to the method provided by the invention, the dynamic pressure distribution function of the bolt connection structure is constructed by selecting the linear pressure distribution function and combining the contact boundary of the contact surface. And density function solving is carried out based on the dynamic pressure distribution function and the friction theory, and a density function of yield displacement is obtained. The density function of the yield displacement is introduced into the universal Iwan model, the Iwan model of the bolt connection structure is obtained, Iwan model construction with few parameters and good representation effect is achieved, and the calculation complexity of the Iwan model is reduced.

The invention belongs to the technical field of ecological restoration, and particularly relates to a method for comprehensively quantifying a shrub-herb-biological crust three-dimensional structure, which comprises the following steps of: collecting sample data, selecting a plurality of shrub, herb and biological crust coexisting sample plots, arranging a plurality of quadrats, respectively obtaining horizontal and vertical indexes of vegetation communities in the quadrats, and calculating the three-dimensional structure of the shrub-herb-biological crust. Representative description indexes of horizontal indexes and vertical indexes of different levels of vegetation are screened out and preprocessed; constructing a model, and determining the weight of each vegetation level and the path weight of the representative description index; and calculating a shrub-herb-biological crust three-dimensional structure index, performing deviation normalization processing on the standard representative description index, and calculating the shrub-herb-biological crust three-dimensional structure index in combination with the path weight of the standard representative description index. According to the method, the three-dimensional structure characteristics of the shrub-herb-biological crust multi-layer vegetation community can be comprehensively and quantitatively represented, and the morphological characteristics of the vegetation community of a forest stand space structure can be better represented.

The invention discloses a training method and device for a federal learning decision tree model, and relates to the technical field of big data. A specific embodiment of the method comprises the following steps: acquiring first credit characteristic data and a credit label of each user in a plurality of users; determining a first dissimilarity matrix according to the first credit feature data; receiving a second dissimilarity matrix sent by the data provider; determining global perception sample selection probability distribution according to the first dissimilarity matrix and the second dissimilarity matrix; and then cyclically executing the following processes: training to obtain a decision tree model: determining a user core set of the current node according to global perception sample selection probability distribution, then determining each sub-node after the current node is split according to the user core set, and taking each sub-node as the current node. According to the embodiment, the accuracy of model prediction can be improved, the performance of the model is improved, the optimization efficiency of the model and the training speed of the model are improved, and the training cost of the model is reduced.

The invention discloses a lightweight class attention mechanism fused improved YOLOv7 flame detection method and system. The method comprises the following steps: acquiring flame and non-flame images, preprocessing the flame and non-flame images, constructing a flame detection data set based on the preprocessed flame and non-flame images, and labeling and dividing the flame detection data set; the original YOLOv7 network model is optimized, and an improved YOLOv7 network model is obtained; training the improved YOLOv7 network model by using the training set, and adjusting model hyper-parameters by using the verification set in the training process to obtain a flame detection model; evaluating the performance of the flame detection model by using the test set; and deploying the flame detection model in a monitoring camera of the power distribution station, and carrying out real-time monitoring and alarming on flames in the power distribution station by using the model. The technical problem that in the prior art, light weight and detection precision of a model are mutually balanced, and actual application requirements are not easy to meet can be solved.

The invention discloses a static model of a SiC MOSFET device based on physical characteristics, including the drain current I. d Model, channel voltage drop V ds Model, gate-source effective voltage V gs The model and the model of the inversion layer channel carrier mobility μ. The drain current I d The model is an improvement on the standard long-channel device model, characterized by the representation of physical processes such as channel modulation effect and depletion layer bulk charge effect; the gate-source effective voltage model introduces the gate-source effective voltage V. gs The interface state trap is characterized by an asymptotic function as a function of the external gate-source voltage V. GS The process by which charge carriers gradually fill the space between the MOSFET and the charge carriers. This model is highly versatile and has good characterization effects on different types of SiC MOSFET devices. The model has high accuracy, and the model parameters all have certain physical meanings, which is conducive to further research based on this model. At the same time, the model is compact, the simulation speed is fast, and it is suitable for circuit simulation.

The invention discloses a three-dimensional terrain recognition method, device and equipment and a computer readable storage medium, and belongs to the technical field of computers. The method comprises the following steps: acquiring a first depth image of a target area acquired by a first image acquisition module at a first moment, a first view image of the target area acquired by a second image acquisition module at the first moment, and first inertial data acquired by an inertial sensor; performing dynamic compensation on the first depth image and the first view image based on the first inertial data to obtain a compensated second depth image and a compensated second view image; generating point cloud data of the target area according to the second depth image and the second view image; and according to the point cloud data, topographic features of the target area are extracted, and the topographic features are used for representing the topography of the target area. The accuracy of the topographic features of the target area extracted by the method is higher.

The invention discloses a method and a system for quantitatively analyzing the vacuum degree of a vacuum switch. The method comprises the following steps: acquiring laser plasma spectral data of a preset target material in a to-be-detected vacuum environment; extracting spectral intensity data of a characteristic spectral line according to the spectral data; laser plasma spectral characteristic parameters are obtained according to the spectral intensity data; and according to the laser plasma spectrum characteristic parameters, obtaining the vacuum degree to be measured by using a preset relation between the laser plasma spectrum characteristic parameters and the vacuum degree. The target material preset in the to-be-detected vacuum environment is induced by laser to generate the laser plasma, the spectrum and morphology characteristics of the laser plasma are detected to obtain the vacuum degree of the to-be-detected vacuum switch, non-contact vacuum degree detection is achieved, the stability of the spectral characteristic parameters of the laser plasma representing the vacuum degree is higher, and the detection accuracy is higher. The vacuum degree can be better represented, the anti-interference capability is higher, and the method is more practical.

The present disclosure relates to a method and apparatus for automatically determining spinal deformity from an image. The invention describes a method for automatically determining spinal deformity from an image showing a plurality of vertebrae of a spine, the method comprising the steps of: a) detecting center points of the plurality of vertebrae shown in the image; b) constructing a center line based on the detected center points; c) calculating local tilt at points along the center line; d) determining a positive tilt maximum and a negative tilt maximum of the local tilt and selecting two reference vertebrae having center points closest to the determined tilt maximums; e) segmenting an upper endplate of the cranial reference vertebra; f) segmenting a lower endplate of the caudal reference vertebra; g) calculating an angle between the upper endplate of the reference vertebra and the lower endplate of the reference vertebra and outputting the angle. The invention also describes a related apparatus.

The invention belongs to the technical field of uranium extraction from seawater, and discloses a porous hydrogen bond material for extracting uranium from seawater and a preparation method of the porous hydrogen bond material. The preparation method comprises the following steps: in a solvent containing inorganic base, taking a cyano compound and hydroxylamine hydrochloride as raw materials, and carrying out heating and stirring reaction to obtain an amidoxime compound; the preparation method comprises the following steps: dissolving an amidoxime compound in an organic solvent, and volatilizing at room temperature to obtain the porous hydrogen bond material for extracting uranium from seawater. The preparation method is simple and easy for industrial production. The prepared porous hydrogen bond material has a three-dimensional structure with uniformly distributed pore channels, uranium in seawater can be selectively adsorbed through a pore channel screening effect, the adsorption capacity of U is about 50 times that of V, the optimal pH range of the porous hydrogen bond material is 7-8, and the condition that the pH of real seawater is 8 is better met.

The invention discloses a key data extraction method driven by user demand, which includes sending the text of user demand into a preset text encoder model to extract text feature vectors; sending each image in the query image data set into an image encoder to generate image feature vectors, combining the obtained image feature vectors with image statistical features to obtain image quality scores; calculating the relevance score sim of the image and the text; multiplying the image quality score by the relevance score to obtain a quality-weighted relevance score; discarding image data with a relevance score less than a threshold; clustering the remaining image data, and taking the image closest to the cluster center as key data. The key data output by the invention is not only closely related to user demand, but also has higher quality itself, can better meet user demand, and reduces the adverse effects of low-quality data on user decision-making.

The invention discloses a wind power cabin fire early warning method for fusing a multi-modal large model with a multispectral image, and relates to the field of fire early warning, and the method comprises the steps: carrying out the image fusion of a visible light image and an infrared image, and obtaining a sample fusion image of a wind power cabin in different working states; the sample fusion image fuses the image features in different spectral modes for advantage complementation, so that the image features of the wind power cabin in the current working state can be better represented, and then the sample fusion image is subjected to image labeling and text description to construct a fire common sense image-text data set; the pre-trained VILA multi-mode large model is subjected to parameter adjustment, so that the established wind power cabin fire early warning model can understand the basic common knowledge of wind power cabin fire early warning, search can be dynamically adjusted according to the target, the recognition and positioning accuracy of the abnormal target of the wind power cabin in the fire early warning state can be improved, and the accuracy of the abnormal target recognition and positioning of the wind power cabin in the fire early warning state can be improved. The accuracy of fire early warning can be improved.

The invention provides a transmission electron microscope sample preparation device and method for high-entropy alloy nanoparticles, and relates to the field of high-entropy alloys and microscopic characterization thereof. The preparation device comprises a substrate and a round cover with holes, grooves of the substrate are in one-to-one correspondence with the holes in the round cover, and the device is made of a metal material with high temperature resistance. According to the preparation method provided by the invention, transmission electron microscope characterization results of the nanoscale high-entropy alloy particles and the high-quality high-entropy alloy nanoparticles can be obtained. The obtained high-entropy alloy particles are uniform in distribution, controllable in size and high in experiment success rate, and a plurality of samples can be prepared at the same time. Through the pulse laser deposition technology, the high-entropy alloy nano-particles which are evenly distributed and controllable in size can be obtained, the experiment success rate can be increased, pollution in the experiment process can be avoided, the capacity of preparing a plurality of transmission electron microscope samples at the same time is achieved, and a high-quality microscopic characterization result can be obtained; and the quality and the preparation efficiency of the high-entropy alloy transmission electron microscope nanoparticle sample are remarkably improved.

The application discloses a WUPyV infection model and a construction method and application thereof, and the construction method comprises the following steps: (1) culturing proximal differentiated lung organoids; (2) removing collagen from the lung organoids to prepare lung organoid ball suspension, mixing the lung organoid ball suspension with a WUPyV sample, and co-incubating and infecting; and (3) re-embedding the lung organoids in basement glue after infection, and continuing to culture in a basic culture medium to obtain the WUPyV infection model. The WUPyV infection model is constructed by using three-dimensional human lung organoids, WUPyV can be separated and cultured, the infection process of WUPyV in human lung organs can be better simulated and characterized, and a new infection model is provided for WUPyV infection pathogenesis and body innate immune response mechanism and interaction between viruses and host cells.

The present invention relates to the field of transmission electron microscope sample preparation, and specifically to a method for preparing transmission electron microscope samples of nanomaterials by transferring single-walled carbon nanotube films. First, a single-walled carbon nanotube film with high purity, low catalyst residue and high transmittance prepared by floating catalyst chemical vapor deposition is pressed onto the nanomaterial on a substrate, and the filter membrane is removed after being moistened with alcohol; the substrate is etched in an etching solution, and the single-walled carbon nanotube film and the nanomaterial are combined with van der Waals forces to form a composite film; then the composite film is transferred to clean water for cleaning, and finally it is picked up using a carrier net or a support ring, and washed and dried by dripping alcohol to obtain a transmission electron microscope sample of the nanomaterial. The present invention solves the problem that it is difficult to cleanly and losslessly transfer nanoparticles or two-dimensional materials on a substrate to prepare transmission samples, and avoids the high-temperature heating process and organic polymer residues when traditionally transferring nanomaterials using organic polymers.

The application discloses a gait recognition method, system, device and medium based on point cloud, which comprises the following steps: performing point cloud projection on gait point cloud data to obtain a depth image sequence; encoding the depth image sequence to obtain initial features, inputting the initial features into a large-scale spatial attention module and a small-scale spatial attention module to extract multi-scale enhanced features, and performing feature fusion on the initial features and the multi-scale enhanced features to obtain first depth features; inputting the first depth features into a multi-scale space-time convolution network module to perform space dimension and time dimension feature extraction and fusion through a pseudo 3D convolution chain of several levels to obtain second depth features, and performing aggregation on the second depth features to obtain target depth features; and inputting real-time acquired gait data into a preset gait recognition model to determine a recognition object. The application can improve the accuracy and efficiency of gait recognition.

The application relates to a building three-dimensional reconstruction method based on multi-coplanar geometry and a graph neural network, which comprises the following steps: pre-processing aerial images and ground images and obtaining an initial point matching set; iteratively extracting multiple groups of local homography matrices between the image pairs; scaling the aerial images and the ground images and constructing a multi-scale pyramid representation; performing matching by using a deep matching network; after the matching results are converged, the inverse transformation of the homography matrix is used to project back to the original image coordinate system to form a candidate cross-platform corresponding point line set; and verified cross-platform control point set is used to replace the matching input of the reconstruction process to output a final building three-dimensional model. The application can obtain sufficient, uniform and geometrically consistent cross-platform control points in cross-platform image pairs with large angle differences, scale changes and local occlusions, thereby replacing or reinforcing the matching input of traditional three-dimensional reconstruction, and realizing high-precision, high-completeness three-dimensional reconstruction and detail recovery of a city building scene.

The application discloses a kind of knowledge distillation methods, devices and equipment of mask autoencoder and storage medium, the method is by respectively establishing the teacher model and student model of mask autoencoder, wherein the teacher model and the student model are all visual transformation model, and the size of the teacher model is greater than the student model;The teacher model is pre-trained;The student model is pre-trained based on the pre-trained teacher model, so that the student model learns data generalization ability from the pre-trained teacher model, and obtains better image features representing ability;The student model is fine-tuned based on the pre-trained downstream task training, and the student model can be deployed in the power edge side lacking of computing resources, while ensuring that the model accuracy does not decrease, the model parameter is reduced, and the real-time inference speed is accelerated.

The invention belongs to the technical field of high-temperature alloy strip foil preparation, and particularly relates to a nickel-based wrought high-temperature alloy strip foil preparation method and a high-strength plasticity control method thereof. The method is designed according to the rolling process deformation amount and the annealing process of the GH3536 alloy strip foil, so that the rolling process deformation amount is controlled reasonably, the internal organization structure is more uniform, intermediate high-temperature rapid annealing is achieved, coarse internal grains are inhibited, and the grain size and uniformity of the internal organization are adjusted; in the final cold rolling state annealing process step, based on the performance use requirements, vacuum annealing and online continuous annealing processes are designed, and the grain sizes in the foil direction of the GH3536 alloy strips with different thicknesses are controlled through recrystallization annealing, so that the ratio of the thickness to the average grain size is larger than 7. The high-strength and high-plasticity GH3536 alloy strip foil is good in surface quality, the room-temperature mechanical properties can meet the requirements that the yield strength (Rp0.2) is not lower than 310 MPa, the tensile strength (Rm) is not lower than 760 MPa, the room-temperature ductility (A) is not lower than 15%, the performance is uniform and stable, and the requirement for high quality is met.

The invention relates to the field of computer systems utilizing computational models, and particularly provides a method, computer equipment, a system, a computer readable storage medium and a computer program product for weather forecast. A method for weather forecast includes: obtaining weather data; deep learning features are extracted from the meteorological data; estimating the horizontal displacement corresponding to the movement of the reference unit in the first frame from the reference position to the target position in the second frame in the feature space; based on the horizontal displacement, determining the vertical displacement of the reference unit in the feature space according to a three-dimensional continuity equation; performing direction convolution operation on the deep learning features; and weighting the direction convolution result based on the vertical displacement. According to the method, the vertical displacement is determined in the feature space through the three-dimensional continuity equation, direction selection is carried out on the deep learning features based on the vertical displacement, the state of the atmosphere in the vertical direction in the real space is combined into calculation of the deep learning network, and the accuracy of weather forecast is improved.

The present invention discloses a characterization method for recording the puff intensity in a single-puff specific puffing behavior, and relates to the research field of dynamic changes in the puff resistance of consumers in the specific puffing behavior of cigarettes. The characterization method is implemented by the following steps: obtaining puff flow data, drawing a puff flow curve, determining the effective puff flow He, determining the puff rise time Tr, determining the effective puff time Te, and characterizing the puff intensity Ss. The present invention can objectively, accurately, quickly, and easily record and characterize the level of the puff intensity of consumers in a single-puff specific puffing behavior, providing a new feasible characterization method for consumers' single-puff specific puffing behavior, facilitating the study of dynamic changes in the puff resistance of consumers in the specific puffing behavior of cigarettes, and having the advantages of accuracy, efficiency, ease of operation, strong practicality, and a wide range of applications.

The invention provides a preparation method of a bimetallic atom catalyst, which comprises the following steps: S1, transferring a metal crystal into a vacuum cavity, cleaning the surface of the metal crystal through ion etching and annealing treatment, introducing oxygen, and preparing a single-layer metal oxide carrier on the surface of the metal crystal; s2, depositing M1 atoms on the metal oxide carrier; s3, depositing an M2 atom on the metal oxide carrier at low temperature, capturing the M2 atom by the M1 anchoring site, and performing first heating treatment to obtain an M1-Ol-M2-carrier structure bimetallic atom catalyst; or S4, after the step S2, oxygen is introduced, the oxygen is dissociated at the M1 site to form an M1-Oa anchoring site, M3 atoms are deposited on the metal oxide carrier at low temperature, the M3 atoms are captured by the M1-Oa anchoring site, and then the M1-Oa-M3-carrier structure bimetallic atom catalyst is obtained through second heating treatment. The bimetallic atom catalyst which is good in atom dispersion and regular in structure is synthesized on the metal oxide carrier; meanwhile, the synthesis method has good metal adaptability and is simple and efficient.

The present application relates to the field of metal casting technology, and specifically to a method for testing the fluidity of inorganic sand for green casting, and accordingly provides a sand core structure and a sand core mold that match the method. Aiming at the fluidity characteristics of inorganic sand, the present application designs a slender, curved sand core structure that can correspond to the actual working conditions, and designs a corresponding sand core mold and a method for testing the fluidity of inorganic sand to match the sand core structure. The method prepares a number of sand core samples with specific structures, and assesses the degree of filling of the mold cavity by the sand core samples to evaluate whether the sand core samples are qualified, calculates the pass rate based on this, and then characterizes the fluidity index of the inorganic sand based on the pass rate. The present application can accurately, stably and quickly detect and characterize the fluidity index of inorganic sand, solving the problem that the fluidity of inorganic sand cannot be characterized in the prior art.

This invention provides a precise characterization method for manufactured sand particle shape. The manufactured sand of all particle sizes is divided into two parts with different particle size ranges: one part with particle size and the other with mass and mass , respectively, and a particle size less than . The particle shape of the manufactured sand with particle size is characterized using the unit flow time method, while introducing flow rate to transform the particle shape result into a dimensionless value. The particle shape of the manufactured sand with particle size is accurately and three-dimensionally characterized using the sphericity method. Based on the mass ratio of the two parts, the particle shape results of the two parts are converted into a particle shape index, thus transforming the abstract quality of manufactured sand particle shape into a concrete numerical value that can be mathematically calculated. This invention combines the flow time method and the sphericity method to form a new detection method. The new method effectively overcomes the shortcomings of single detection methods while leveraging their advantages of precise detection for a specific particle size range.

This invention discloses a method for characterizing multi-stage carbonate reservoirs based on time-varying Gaussian window transient extraction transform, used for characterizing multi-stage carbonate reservoirs. The method includes: inputting the seismic signal to be analyzed, where is time; constructing a short-time Fourier transform containing the time-varying window width parameter to be optimized, where represents frequency; determining the optimal window width parameter at each time step based on the local Rényi entropy minimization criterion; substituting and calculating its time partial derivative; constructing a transient extraction operator based on the time partial derivative results, including a local maximum criterion, an energy derivative zero-point criterion, and a group delay migration criterion; and proposing a time-varying Gaussian window transient extraction transform under the action of this operator for characterizing multi-stage carbonate reservoirs.

The present invention relates to the field of ancient building protection and digitization technology, a method and device for 3D digitization of real scenes of ancient buildings by fusion of multi-source data of surveying and mapping geographic information, comprising: after confirming that accurate image acquisition of ancient buildings can be performed using an acquisition parameter set and a reference occlusion area set, fitting an acquisition path and coordinates for acquiring images of ancient buildings based on the acquisition parameter set and a reference coordinate system, photographing a target ancient building using an image collector to obtain an acquired image set, confirming that the acquired image set is a preset target image set, generating a first 3D point cloud map for characterizing outdoor features of the ancient building using the target image set, obtaining a second 3D point cloud map for characterizing indoor features of the ancient building, fusing the first 3D point cloud map and the second 3D point cloud map to obtain an optimized 3D point cloud map, and constructing a target ancient building model based on the optimized 3D point cloud map. The present invention can improve the accuracy and intelligence of 3D digitization of real scenes of ancient buildings.

The invention discloses a user interest prediction method, device and system, and a storage medium. The method comprises the following steps: S1, acquiring historical interaction data of a user; s2, according to historical interaction data of the user, mapping the representation in the Euclidean space to different geometric spaces; s3, performing diffusion depolarization on the object embedding in the different spaces to obtain depolarized user representations of the different spaces; and S4, mapping the user representations in different spaces back to the Euclidean space for reconstruction and matching to obtain the actual interests of the users. By adopting the technical scheme provided by the invention, the interest and preference of the user can be predicted more accurately.

A small sample SAR target recognition method based on a two-stage contrastive learning framework involves image recognition. Aiming at the practical problems of difficulty in obtaining SAR data sets and time-consuming and labor-intensive data annotation, a contrastive learning framework based on a two-stage training strategy is proposed to solve the SAR target recognition under small sample data conditions. The training process is decoupled into a representation learning stage and a classification learning stage, including: 1) In the representation learning stage, supervised contrastive learning is used to train the representation module of the network on the original SAR data set; 2) In the classification learning stage, the encoding module is fixed and the classification module of the network is trained using a rebalancing strategy. Based on contrastive learning in machine learning and combined with the idea of ​​decoupled learning, the method has strong practicality and good stability, and can meet the requirements of high standards for SAR target recognition accuracy in cases of imbalanced categories and small sample data.