648 results about "Initial sample" patented technology

The invention provides a training data synthesis method and device based on error extrapolation and inference chain analysis, a medium and a program product. The method comprises the steps of obtaining an initial sample set; performing multiple sampling reasoning on the problem of each task sample by using a small language model to generate a plurality of reasoning chains; calculating an overall error score of each reasoning chain based on a preset error evaluation rule, and determining a to-be-corrected reasoning chain; the inference chain to be corrected and the corresponding question are input into the large language model together, and a corrected answer is generated; forming a new task sample by the question and the corrected answer, and finely adjusting partial parameters of the small language model; repeatedly executing the process until the performance index change rate of the model on the task evaluation set is lower than a preset threshold value, and outputting a final task sample; and forming a training sample set by a plurality of final task samples, and performing all-parameter fine tuning on the small language model. According to the method, the training data self-optimization path is constructed by taking the model error as guidance, so that the semantic consistency and the data validity are improved.

The invention belongs to the technical field of speech enhancement, and discloses a Schrodinger bridge-based diffusion model speech enhancement method and system, and the method comprises the steps: converting a diffusion process of a diffusion model into a solving process of a stochastic differential equation, and determining the stochastic differential equation according to the theoretical principle of a Schrodinger bridge, and the complex spectrum is directly used as the input of the diffusion model, so that the tedious work of extracting a phase spectrum and an amplitude spectrum from the complex spectrum and the memory overhead caused by inverse transformation are omitted, and meanwhile, the alignment problem between the phase spectrum and the amplitude spectrum is also avoided. According to the speech enhancement method, unique features of time sequence signals are captured through a Transformer module, multi-scale information is fused through a U-Net module, a loss function covering a time domain, a frequency domain and a time-frequency domain is matched, the difference between a prediction sample and a clean sample is gradually reduced, the nonlinear diffusion process from a noisy sample to the clean sample can be directly learned, and the speech enhancement accuracy is improved. Therefore, the structure information of more initial samples is reserved.

The embodiment of the present invention discloses a method for predicting corneal reshaping parameters based on an artificial neural network, and relates to the technical field of orthokeratology lenses. The method comprises: obtaining an initial sample data set, deleting target samples that do not contain necessary parameters, and obtaining a basic sample data set; preprocessing the basic sample data set to obtain a final sample data set; training the initial artificial neural network model to obtain a preliminary model; using the preliminary model to forward propagate the input layer data of the basic sample data set to obtain a calculation result and compare it with the result of the basic sample to obtain a final sample data set, and using the final sample data set to train the preliminary model to obtain a prediction model; inputting actual eye detection data into the prediction model to obtain corresponding corneal reshaping parameters. The present invention splits the data to be predicted into multiple dimensions, constructs different prediction models according to each data to be predicted, and then performs distributed training and uses them in combination.

The invention discloses an optimization design method for a control surface of a wide-speed-range large-attack-angle reusable carrier, and relates to optimization design of aerodynamic configuration of an aircraft. The method comprises the following steps: S1, setting a control surface aerodynamic configuration design variable range, and generating an initial sample library by adopting Latin hypercube sampling; s2, aerodynamic parameters are obtained through CFD simulation, and a Kriging proxy model is trained; s3, evaluating the precision of the proxy model by taking a U learning function as a criterion, and stopping adding points when the minimum U function value is smaller than a threshold value; s4, constructing an optimization model which takes maximization of the lift-drag ratio and the static stability margin under the hypersonic speed as a target and takes the condition that the aerodynamic parameters of the supersonic speed / subsonic speed are not lower than a reference value and the hinge moment as constraints; and S5, performing iterative optimization by adopting an improved multi-target particle swarm algorithm based on genetic algorithm crossover mutation operation, updating the proxy model after the optimal solution of each generation is subjected to CFD verification, and outputting an optimal solution set. The problem of wide-speed-range aerodynamic configuration contradictions is solved, and the comprehensive performance of the carrier is remarkably improved.

The invention discloses a small sample self-adaptive mixed gas sensor dynamic response modeling method, which belongs to the technical field of gas sensors, and specifically comprises the following steps: testing mixed gas through a gas sensor array, and generating a response signal; pulse signals representing gas concentration mutation characteristics are generated, the pulse signals and response signals form an initial sample set, short-time-sequence fragments are obtained through truncation separation, initial state signals are generated at a time step before a response rising point, and a short-time-sequence fragment sample set is obtained; forming an enhanced data set through non-playback extraction and carrying out normalization processing on the enhanced data set; and constructing and training to obtain gas sensor dynamic response models of different gas sensors. By simulating the complex dynamic response of the gas sensor, the data sample enhancement of the mixed gas is completed, the difficulty of training a mixed gas component identification model by using a small sample is reduced, and the model can quickly adapt to environmental change and performance difference of different sensors; and reliable and effective data enhancement support is provided for mixed gas component identification.

The invention discloses a carbon dioxide oil displacement burying multi-objective optimization method based on a self-adaptive agent model, and relates to the technical field of oil reservoir injection and production optimization. The method comprises the following steps: firstly, constructing a carbon dioxide flooding embedding injection-production optimization model, then acquiring a plurality of initial samples by adopting Latin hypercube sampling to construct a database, preferably selecting each target agent model based on the database, and then generating a Pareto leading edge by utilizing a dominating class search strategy, a decomposing class search strategy and an index class search strategy; and in the optimization stage, a preferred potential solution is searched according to a hypervolume improvement maximum strategy, numerical simulation is carried out, the database is updated until a preset number of times is reached, each iteration optimization scheme, an oil reservoir net present value and a carbon dioxide burying amount are output, and multi-target optimization of carbon dioxide flooding burying is completed. According to the method, the carbon dioxide flooding multi-objective optimization efficiency is improved, meanwhile, the search direction is dynamically adjusted through the hyper-volume evaluation index, the Pareto frontier is accelerated, and accurate prediction of the carbon dioxide flooding development scheme is achieved.

The invention discloses a wind field rapid prediction method and system based on optimized Latin hypercube sampling and POD-BPNN, and the method comprises the steps: carrying out Latin hypercube sampling to generate an initial sample point set, introducing a sensitivity weight, and carrying out the iterative optimization of sample distribution through a greedy strategy; assembling the CFD numerical simulation flow field data of all sample points in the sample point set into a flow field snapshot matrix, and determining a dynamic multi-target truncation order and a corresponding POD mode and coefficient; building and training a BPNN model, packaging the trained BPNN model, the POD modal matrix, the mean field and the flow field reconstruction logic into an FMU module, obtaining the FMU module which can be called in a cross-platform manner, and achieving the real-time prediction of a wind field. The invention relates to the technical field of wind field prediction, significantly improves the precision and calculation efficiency of wind field prediction, and provides an efficient and accurate solution for the rapid prediction of a wind field.

The embodiment of the invention discloses a model training method and device applied to direct current arc discharge detection. The method comprises the steps of performing at least two times of data processing on first time sequence data and second time sequence data and forming a sample set; selecting an initial neural network model obtained by training the initial sample set as a current detection model, taking the initial sample set as a current test set, and testing the current detection model by using the current test set to obtain a current test output result; and determining a next sample set based on the current test output result, taking a next neural network model obtained by training the next sample set as a current detection model, and taking the next sample set as a current test set for model test until the current test output result meets a preset requirement. According to the technical scheme, the precision and the accuracy of the output model can be improved, so that the accurate detection of the direct current arc discharge of the photovoltaic system is realized, and the safe and stable operation of the photovoltaic power generation system is ensured.

The invention discloses a parallel evolutionary solution method for search space segmentation, comprising: S1, randomly generating solution schemes via a search space sampling and segmentation module, calculating fitness values by combining initial sample schemes with an optimization objective function, statistically screening sample schemes, determining the dimension direction K of the segmented search space, and dividing the search space into subspaces; S2, executing a global search algorithm in the segmented subspaces to obtain a global initial solution scheme; S3, using the global initial solution scheme as a starting point, obtaining a precise solution scheme via a local search algorithm. This method employs evolutionary sampling of the search space, deriving fitness terrain analysis results, and guiding the division direction and step size of the search space. Multiple search subspaces are segmented to enable parallel search for subsequent evolutionary calculations.

The invention discloses a post-pumping air tank robustness collaborative design method considering hydraulic parameter time-varying characteristics, which comprises the following steps: defining a hydraulic parameter uncertainty fluctuation interval of a water delivery system in a full life cycle, and setting a body structure decision variable range of a post-pumping air tank; performing combined sampling in the water conservancy parameter uncertainty fluctuation interval and the body structure decision variable range by using a test design method to generate an initial sample set; performing steady-state and transient-state coupling simulation on the initial sample set, constructing a constant-flow operation condition of the water delivery system by using a hydraulic equation, updating a water pump working point and pipeline pressure distribution, performing transient simulation by using a characteristic line method to obtain a hydraulic response index, and training to obtain a water hammer response agent model; constructing a robustness optimization objective function based on failure probability constraint; and performing global optimization on the target function by using an intelligent optimization algorithm, calling the water hammer response agent model to perform random simulation, evaluating a failure probability, and outputting a target design scheme.

The invention relates to a limestone granularity detection method based on YOLO-ADM, which comprises the following steps: firstly, collecting a raw material crushing section limestone data set, then carrying out data enhancement on an initial sample limestone data set, and dividing the initial sample limestone data set into a training set, a verification set and a test set; a deformable attention mechanism DAttention, a multi-scale sequence fusion module SSFF and a triple feature decoding module TFE are integrated in the YOLOv8 framework, and a YOLO-ADM model is obtained; on this basis, a training loss function is improved, and a loss function MioU based on the auxiliary frame and the minimum point distance is designed; and training the YOLO-ADM model through the training set and testing the performance of the network model through the test set. According to the method, the deformable attention DAttention is introduced into the YOLOv8 network, the YOLOv8 neck network is redesigned based on the multi-scale sequence fusion module SSFF and the triple feature decoding module TFE, the designed loss function MioU is combined, the excellent performance is shown in limestone detection, and the generalization performance is good.

The invention relates to the technical field of component analysis, and discloses an X-ray fluorescence spectrum-based rock and ore silicate component analysis method, which comprises the following steps: S1, sample collection and pretreatment: collecting, crushing and drying an initial sample, S2, sample tabletting preparation, S3, standard sample preparation, S4, X-ray fluorescence spectrum analysis, and S5, data processing and result analysis. According to the method, a complex interference characteristic mode between elements can be automatically captured through an established deep learning model and learning of a large number of known components and fluorescence spectrum data, and in an example of correcting interference between multiple elements in rocks and mines, the model can utilize a network structure of the model and comprises an established convolutional layer, a pooling layer and a full-connection layer. Effective information is extracted from multiple groups of sample data, the interference relation between elements is accurately learned, and compared with a traditional method, the precision of analyzing the content of silicate components in rock ore is remarkably improved, so that a measurement result is closer to a real value.

The invention provides a structure reliability analysis method based on an adaptive variable fidelity model, and the method comprises the steps: generating an initial sample point set which is uniformly distributed and has representativeness through an improved random sampling method KMODMC, enabling the initial sample point set to comprise a low-fidelity sample set and a high-fidelity sample set, training a BP neural network through employing the low-fidelity sample set, and carrying out the training of the BP neural network through employing the high-fidelity sample set; a low-fidelity BP neural network model is obtained; meanwhile, based on an error training Kriging model of a high-fidelity sample set and a low-fidelity model predicted value, an error correction Kriging model is constructed, the low-fidelity BP neural network model and the error correction Kriging model are combined to form a multi-fidelity mixed agent model, and adaptive iterative optimization is performed through a double-model alternate point adding sampling strategy. And finally obtaining a high-precision multi-fidelity hybrid agent model for structural reliability evaluation. According to the method, the problems of high cost of high-fidelity simulation calculation and insufficient precision of a low-fidelity model are solved, and the adaptive capacity and prediction precision of the model in a complex nonlinear problem are effectively improved.

The invention discloses a power grid typical operation mode generation method and system based on UMAP and k-means + + clustering, and relates to the technical field of power systems, and the method comprises the steps: collecting power grid operation mode data and measurement data, carrying out the arrangement of the collected data, obtaining an initial sample set, carrying out the normalization processing of each equipment vector in the initial sample set, and obtaining a power grid typical operation mode; the method comprises the following steps: normalizing data of a power grid, performing dimension reduction processing on the normalized data, determining an optimal hyper-parameter combination, determining an optimal classification number by using an elbow rule, performing k-means + + clustering on a sample set, extracting a sample point closest to a central point in each category from a clustering result, and obtaining a typical operation mode of the power grid according to the sample point closest to the mahalanobis distance. According to the method, the data processing efficiency and accuracy are improved, the clustering effect is optimized, the optimal classification number is automatically determined, the safe and stable operation level of the power grid is improved, optimal dispatching of the power grid is supported, the coping capacity of the power grid to emergencies is enhanced, and intelligent development of the power grid is promoted.

The invention relates to a computer implemented method for generating an aerial image of a photolithography mask in an image space, the method comprising: obtaining a representation of a design of the photolithography mask; applying a trained conditional diffusion model that is configured to sequentially revert a stochastic process to an initial sample in order to generate an aerial image of the photolithography mask, wherein the trained conditional diffusion model is conditioned on the representation of the design of the photolithography mask. The invention also relates to computer implemented methods for defect localization, alignment, repair shape generation, training data generation and corresponding systems.

The invention provides a multi-objective optimization method based on an adaptive point adding criterion and an agent model, and relates to the technical field of multi-objective optimizing.The method comprises the steps that a Latin hypercube sampling method is used for sampling an optimization algorithm, and initial sample points are obtained; constructing an initial agent model by using the initial sample points; based on an error standard and a self-adaptive point adding criterion, updating the initial agent model to obtain an updated agent model; and calculating the optimization algorithm by using the updated agent model to obtain a multi-target optimization result, and completing multi-target optimization. According to the method, the problem that multi-objective optimization is difficult to balance precision and convergence speed is solved.

The invention relates to a centrifugal pump cavitation performance optimization method and system based on feature selection, and the method comprises the steps: taking the cavitation entropy production and lift of a centrifugal pump as optimization targets, taking the efficiency as a constraint condition, carrying out the modeling simulation of the centrifugal pump, and determining an initial optimization variable and an optimization space range; extracting a plurality of groups of uniformly distributed variable combination schemes by adopting a test design method based on the optimized space range; carrying out modeling simulation on the variable combination scheme to obtain cavitation entropy production and a lift value, and constructing an initial sample library; training the first approximation model by utilizing the initial sample library, screening out a final optimization variable which has obvious influence on the optimization target, and obtaining a final sample library; and training the second approximation model by utilizing the final sample library, optimizing and verifying by utilizing an optimization algorithm, and outputting an optimal parameter combination. According to the method, parameters which have obvious influence on the optimization target can be better selected as optimization variables, and meanwhile, the cavitation entropy production of the optimization target is introduced, so that the comprehensive optimization effect of cavitation and lift is improved.

The embodiment of the invention provides an image processing model training method and device and an image generation method and device.The method comprises the steps that in response to selection operation for a sample generation assembly and a model training assembly, the sample generation assembly and the model training assembly are connected, and a model training task flow is constructed; obtaining an initial sample image, inputting the initial sample image into a sample generation component, generating an image prompt text set corresponding to the initial sample image through the sample generation component, and generating a target sample image set based on the image prompt text set; and inputting the target sample image set into a model training assembly, training the initial image processing model by using the target sample image set through the model training assembly, and obtaining a trained target image processing model. Through the modular design of the components, the user can use the model training function more conveniently, and the model training complexity is reduced.

The invention provides a photovoltaic power prediction method and system based on a double-current network and multiple attention, and the method comprises the steps: obtaining historical photovoltaic generating capacity and meteorological information, and obtaining initial sample data; performing variable factor influence analysis on the initial sample data to obtain a simplified initial sample data set; performing similar daily clustering processing on the simplified initial sample data set to obtain each scene data set after clustering; providing a photovoltaic power prediction initial model based on a double-current network and multiple attention, and training the photovoltaic power prediction initial model by using each clustered scene data set to obtain a photovoltaic power prediction model; and taking the historical generating capacity and meteorological data of the target photovoltaic power station in the set time period as input data of the photovoltaic power prediction model, and outputting a photovoltaic power prediction result. According to the invention, through the clustering integration method, the clustering robustness and accuracy are improved; and through multi-dimensional feature extraction and feature fusion, the prediction accuracy is improved.

Provided is a model processing method, a voice interaction method, an electronic device and a storage medium, relating to fields of artificial intelligence, big data and voice technologies. The model processing method includes: obtaining a candidate question set of each initial sample data in M initial sample data, wherein the initial sample data includes m rounds of question-and-answer between an object and an agent, the candidate question set includes a next round of question set corresponding to the mth round of question in the m rounds of question-and-answer; obtaining M training sample data based on the M initial sample data, the candidate question set and label data of each initial sample data, wherein the label data includes a target question to be generated by the agent in the (m+1)th round; and training a model to be trained by using the M training sample data to obtain a target model.

The invention discloses an electromagnetic inversion uncertainty evaluation method based on ensemble Kalman filtering, and belongs to the technical field of geophysical electromagnetic inversion, and the method comprises the steps: carrying out the time recursion of an initial sample set through the prediction step of ensemble Kalman filtering, and obtaining a prediction state variable; an observation variable of a prediction state variable is calculated through an updating step of ensemble Kalman filtering, a cross-covariance matrix is constructed based on the prediction state variable and the observation variable, and localization operation is performed on the cross-covariance matrix by using a localization weight matrix so as to correct estimation deviation of the cross-covariance matrix. A Kalman gain matrix is constructed based on the corrected cross covariance matrix, and prediction state variables are updated in combination with actual observation data; and performing electromagnetic inversion uncertainty evaluation based on the updated state variable. According to the method, the electromagnetic inversion precision and stability can be effectively improved, and uncertainty evaluation is provided for the magnetotelluric two-dimensional inversion result.

The invention discloses a variant reconstruction AUV (Autonomous Underwater Vehicle) shape optimization method based on proxy-assisted multi-starting-point space reduction. The variant reconstruction AUV shape optimization method comprises the following steps: establishing a parameterized model of a variant reconstruction AUV; constructing a variant reconstruction AUV shape modeling-grid division-performance simulation automation framework, realizing input of design variables, and automatically performing variant reconstruction AUV shape modeling and watershed grid division; a variant reconstruction AUV shape hydrodynamic force simulation calculation framework is built, and assessment of variant reconstruction AUV hydrodynamic force performance is achieved; generating a plurality of groups of initial samples for a design space of design variables, calculating target function values corresponding to the samples by utilizing the simulation automation framework and the simulation calculation framework, and importing the samples and corresponding sailing resistance into a sample library; and performing optimization in the design space of the design variables by utilizing an agent-assisted multi-starting-point space reduction method to obtain an optimal solution of the variant reconstruction AUV shape design problem. According to the method, the reliability of an optimization result and the design efficiency can be improved, and the optimization time is shortened.

The invention relates to the technical field of zone-melting polycrystalline silicon rod stress uniformity detection, in particular to a zone-melting polycrystalline silicon stress detection method which comprises the following steps: S1, cutting along the radial direction of a silicon rod to obtain a slice; s2, performing multi-point sampling on the slices at the same interval to obtain an initial sample; s3, an oxide layer on the surface of the initial sample is removed, then cleaning and drying are conducted in sequence, and a final sample is obtained; s4, testing the final sample by using an X-ray diffractometer to obtain a data file; s5, processing the data file by using XRD data analysis software, and extracting position information of different final sample diffraction peaks; s6, judging whether the stress distribution is uniform or not according to the extracted position information of the diffraction peaks; the stress magnitude and uniformity of the polycrystalline silicon rod on the same level or the vertical level can be measured, and theoretical support is provided for preparation of zone-melting polycrystalline silicon.

The invention provides a method and a device for performing large-model self-distillation absorption depth reasoning by an intelligent computing center cloud platform through computing power, and relates to the technical field of intelligent computing centers, intelligent computing centers, computing power infrastructures and intelligent computing clouds, and the method comprises the following steps: S1, obtaining multiple groups of initial sample data; s2, constructing multiple groups of first sample data and multiple groups of second sample data based on the multiple groups of initial sample data; and S3, training the initial large model based on the multiple groups of first sample data and the multiple groups of second sample data to obtain a first large model. According to the method, the displayed natural language deep reasoning step is distilled to a continuous hidden space, and the first model can still ensure complete and excellent reasoning ability under the condition of not displaying and generating the middle deep reasoning step, so that resources occupied by a large model can be reduced; therefore, the large model service provided in the intelligent computing center cloud platform is greatly increased.

The invention discloses a proxy model auxiliary evolution method based on two-stage adaptive switching and Voronoi niche. The method comprises the steps of generating an initial sample by adopting optimal Latin hypercube sampling, obtaining a real fitness value through simulation, constructing an initial training data set, and setting an initial iteration counter k and an evaluation budget threshold value; training a proxy model based on the current data set, and executing evolutionary algorithm optimization by using the proxy model to obtain a current optimal individual; judging two stages according to an iteration counter k and a distance threshold value; the development execution stage comprises the following steps: evaluating the real fitness and updating a data set; the exploration execution stage comprises the steps of multi-modal approximate detection, Voronoi niche division, multi-modal collaborative search, real fitness evaluation and data set updating; judging a termination condition and outputting a result; according to the method, three core technologies of an adaptive stage switching mechanism, multi-modal approximate detection and Voronoi niche division and multi-extremum collaborative filling are creatively integrated, so that the defects of insufficient global exploration, low local development precision and poor convergence efficiency of a traditional method in a high-dimensional multi-modal expensive optimization problem are effectively overcome; and the solving efficiency and reliability of a complex and expensive optimization problem are remarkably improved.

Active machine learning systems for anomalous event detection and classification. Initial samples from an industrial environment may be received and labeled. Initially, a training pool of audio samples may be labeled. These labeled samples may be used to train an audio event classifier to detect and categorize sounds. Environment states may be calculated using outputs from the classifier. A batch of audio samples may then selected from an unlabeled pool for annotation, guided by a reinforcement learning agent. These selected samples may be annotated and added to the labeled training pool. The classifier may be retrained with this updated pool. Rewards may be calculated for each of the annotated samples based on their annotations. The environment states may be updated using the retrained classifier, and the exploration-exploitation parameter of the reinforcement learning agent may be adjusted. The reinforcement learning agent may be retrained using the updated environment states and rewards.

The invention discloses a sample screening method and system based on weight clustering and dynamic interval reduction. The method comprises the following steps: generating an initial sample pool through a Monte Carlo method; based on the initial sample pool, utilizing a sample screening process to carry out multiple rounds of iterative screening to obtain a screened sample pool; the sample screening process comprises the following steps: acquiring uniformly distributed candidate samples from a current sample pool by adopting a weight clustering method; predicting an upper threshold value and a lower threshold value of a limit state function in the sample pool through an agent model by utilizing an interval reduction method; and selecting the candidate samples of which the predicted values of the limit state functions are between the upper threshold value and the lower threshold value, and updating the sample pool. According to the method, candidate samples are dynamically screened in combination with a weight clustering method, interval reduction is carried out by utilizing upper and lower thresholds of a limit state function predicted by an agent model, the screening precision is ensured, meanwhile, the scale of the candidate samples is effectively compressed, the redundant calculation amount is reduced, and efficient iterative optimization is achieved.

The invention discloses a Kriging agent model-based engineering structure reliability analysis method, which comprises the following steps of: sampling from random variables and probability distribution of an object to be analyzed, and respectively obtaining a candidate sample set and an initial sample set; obtaining an initial experimental design set based on the initial sample set and the corresponding real response value, and constructing an initial Kriging agent model about the to-be-analyzed object; obtaining an optimal sample from the candidate sample set through a random weight learning function, and expanding the optimal sample and a corresponding real response value to an initial experimental design set; before the initial experimental design set enters the next round of data expansion, deleting an optimal sample in the current round and sample data in a preset redundancy distance range of the optimal sample from the candidate sample set; updating and iterating the initial Kriging agent model about the to-be-analyzed object through the experimental design set; and calculating the reliability index of the to-be-analyzed object according to the failure probability calculated by the current Kriging agent model.

The invention discloses an air conditioner voice assistant training method and device, a storage medium and a server, and relates to the technical field of air conditioners, and the method comprises the steps that air conditioner voice assistant training data are obtained, and the air conditioner voice assistant training data comprise multiple pieces of initial sample data of an initial protocol format; protocol format adjustment is conducted on initial sample data included in the air conditioner voice assistant training data, and multiple pieces of new sample data with the simplified protocol format are obtained; and training a to-be-trained base model by adopting the multiple pieces of new sample data in the simplified protocol format to obtain an air conditioner voice assistant model which is used as an air conditioner voice assistant. The method can improve the understanding accuracy of the air conditioner voice assistant for the intention expressed by the user voice in the air conditioner scene, reduces the consumed resources during model training, improves the efficiency, and improves the user interaction experience in the air conditioner scene.

The invention discloses a depth map super-resolution method based on blind degradation. The method comprises the following steps: forming an initial RGB-D sample pair by using a high-resolution RGB image and a corresponding high-resolution synthetic depth map; screening out a high-quality RGB-D sample pair from the initial RGB-D sample pair; inputting the high-quality RGB-D sample pair into a learnable degradation network, converting the high-quality RGB-D sample pair into a low-resolution training sample, and constructing an RGB-D training data set by using the low-resolution training sample and the high-resolution RGB image corresponding to the low-resolution training sample; using the RGB-D training data set to train the multi-modal super-resolution reconstruction network to obtain a depth map super-resolution model; inputting the low-resolution depth map and the corresponding high-resolution RGB image into a depth map super-resolution model to output a high-resolution depth map; the method solves the problems that an existing depth map super-resolution method mostly depends on a fixed degradation hypothesis, a real sensor and complex degradation distribution of compression transmission are ignored, and accordingly generalization ability is insufficient, edges are fuzzy and artifacts are increased in an actual scene.