281results about How to "Reduce computing cost" patented technology

The invention belongs to the technical field of knowledge graph construction, and particularly relates to a multi-source heterogeneous data fusion and knowledge graph automatic construction system, which comprises a semantic extraction module for extracting minimum semantic fragments from multi-source data and performing cross-source alignment to generate a symbolized semantic framework; the data purification module converts multi-source data into symbolic predicates according to the framework, filters low-evidence data and outputs a purification set; the ternary generation module inputs the initial triad candidate set into a minimum rule grammar, and extracts entities, relationships and attributes to generate an initial triad candidate set; the disambiguation calibration module generates entity two-hop topological fingerprints based on the candidate set, completes disambiguation alignment and corrects conflicts, and outputs unambiguous structured knowledge; the mapping fusion module fuses the mapping information with the minimum connected ontology, and constructs and verifies an initial mapping knowledge domain; and the incremental updating module processes newly added data, performs incremental merging and maintains consistency, and forms a complete knowledge graph. According to the method, through full-link automatic construction, efficient fusion and high-quality atlas are realized.

The invention relates to the technical field of industrial vision and artificial intelligence, and discloses an aluminum coating formula prediction method and system based on industrial vision and double-model fusion. The method comprises the following steps: acquiring visible light and near-infrared band images of the surface of an aluminum material coating through multispectral image acquisition, identifying a defect area to generate a binary mask, extracting three types of features of color, texture and spectral reflection, and respectively inputting feature vectors into a random forest regression model and a convolutional neural network model to obtain a target image; and dynamically calculating a fusion weight according to the verification set error, and carrying out weighted fusion on the prediction results of the two models to generate a formula component content prediction value. According to the method, the technical problems of low precision, low efficiency and insufficient generalization ability of a traditional aluminum coating formula determination method are solved, and rapid and accurate prediction of the formula is realized.

The invention discloses a digital electroencephalograph, and belongs to the technical field of brain-computer interface signal processing. An electroencephalogram signal processor of the digital electroencephalograph acquires digital neural signals of a plurality of channels for a target object, wherein the digital neural signals comprise LFP signals and pulse signals; performing feature extraction on the two signals based on a unified time window and a sliding step length, and integrating a time-aligned feature matrix into a first fusion feature tensor; performing dynamic normalization processing on the first fusion feature tensor, wherein the statistical magnitude is updated along with the change of the input data; and n parallel LSTM networks are used to decode the normalized second fusion feature tensor to obtain n decoding results of different dimensions, the n LSTM networks are mutually independent, and each LSTM network has an independent optimizer and an independent parameter. According to the method, multi-modal features can be aligned in time, signal drift is compensated, decoding robustness is improved, and inter-task optimization decoupling and personalized training are achieved through independent parameters and an optimizer.

The invention provides a target identification method and system based on multi-source information fusion, and relates to the technical field of low-altitude target detection. The method comprises the following steps: acquiring target radar track data, interception equipment track data and position area information data; based on target radar track data, extracting a first feature in an RCS form dimension, and extracting a second feature in a motion dimension; based on the target radar track data and the track data of the monitoring equipment, determining a frequency spectrum monitoring correlation factor and regional position information features; performing feature fusion on the first feature, the second feature, the spectrum interception correlation factor and the regional position information feature to obtain a target feature, and identifying a target type; and identifying a target threat level based on the target type and the target radar track data. The method and the device are used in a target identification process based on multi-source information fusion, and the technical problem that the target threat degree cannot be accurately identified in a complex environment in the prior art is solved.

The application discloses a mixed expert network-based optical flow estimation method and system, belonging to the field of computer vision and artificial intelligence. First, the input two frames of images are preprocessed, low-resolution features are extracted through a mixed expert feature extractor (MoEE) containing a sparse activation mechanism to reduce redundant calculation; then, dot product operation is performed on the low-resolution feature maps to construct a 4D correlation volume to capture pixel motion matching relationship; subsequently, a mixed expert updater (MoEU) is used to iteratively update the hidden state and regress the residual flow increment through a dynamic expert selection mechanism to optimize the optical flow accuracy; finally, a multi-scale upsampling module is used to reconstruct the high-resolution optical flow field to output the high-precision optical flow result. The algorithm realizes dynamic resource allocation through the MoE architecture, significantly reduces the calculation cost while ensuring the accuracy of optical flow estimation, and can adapt to resource-constrained scenarios such as automatic driving and unmanned aerial vehicles, and can balance efficient inference and flexible deployment.

The invention discloses a blue-green algae identification and quantification method and device based on unmanned aerial vehicle remote sensing images and deep learning and a medium, and relates to the technical field of information data processing. Combining meteorological data and water quality monitoring data to construct an adaptive dynamic environment algorithm to extract EXIF metadata including camera parameters and attitude angle information, and establishing a geometric projection model to perform coarse orthographic correction on an original aerial image; the method comprises the following steps: extracting a multi-scale cyanobacterial bloom image feature map by stages based on a ResNet architecture and in combination with a feature pyramid network FPN fused with an attention mechanism, and obtaining an instance mask of cyanobacterial bloom through an anchor-free region proposal network, ROI Align and a head network; based on an improved GIS space projection and deep learning algorithm, carrying out high-precision area measurement and calculation on a binary mask image converted from the instance mask; according to the method, the influence of irrelevant interference on blue-green algae identification can be reduced, and the accuracy and comparability of blue-green algae identification and area calculation are improved.

The invention discloses a speculative reasoning method for a multi-modal large language model based on mobile edge collaboration. The speculative reasoning method comprises the following steps: constructing an equipment-side double-clue probe model; task semantic distillation and feature alignment are carried out, joint optimization is carried out on double heads of a double-clue classifier, knowledge of an edge end MLLM is distilled to a double-clue probe model, and an equipment end minimum modal delay decision is executed; asynchronous mode screening and speculative reasoning are realized through a hierarchical gating network; and executing a rollback decision to obtain a corrected final reasoning result. According to the embodiment of the invention, a lightweight dual-clue probe model is constructed at a mobile client, task semantic features are extracted from the deep layer of an edge MLLM, joint estimation of reasoning confidence and modal sufficiency is realized, and a minimum modal delay decision of an equipment end is executed; a hierarchical gating mechanism is adopted at an edge end, a confusion-guided rollback mechanism is introduced, an early reasoning result with insufficient confidence is corrected, and early recognition and efficient reasoning of a minimum full subset in an asynchronous mode are achieved.

The invention provides a method and device for calculating rudder deviation required by unit normal overload of a statically unstable aircraft, and the method comprises the steps: selecting a plurality of calculation state points, and obtaining a small disturbance linearization equation at each calculation state point; a stability augmentation control law is constructed, the pitching moment characteristic of the aircraft is changed by feeding back the attack angle, the pitching damping characteristic of the aircraft is improved by feeding back the pitching angle rate, and therefore the aircraft is not divergent under the control of the stability augmentation control law and has the expected stability and damping characteristic; establishing a calculation model of rudder deflection required by unit normal overload based on a stability augmentation control law and a small disturbance linearization equation at a calculation state point, calculating normal overload by the calculation model through a pitch angle rate, an attack angle and a vacuum speed, and processing the calculation model to obtain a transfer function from a control channel elevator instruction to the normal overload; and calculating a normal overload value generated by unit elevator skewness based on the transfer function, wherein the reciprocal of the normal overload value is the rudder skewness required by unit normal overload.

A kind of time domain ground penetrating radar intelligent forward method based on physical parameter guidance, comprising the following steps: step one: build underground medium model and generate supervision data;Step two: the ground penetrating radar GPR forward simulation is expressed as the mapping problem of " physical parameter field to radar profile ";Step three: build PhyTANet network based on encoder decoder;Step four: the PhyTANet network obtained by step three is trained, fine-tuned;From the above steps, finally, after training fine-tuning ends, the dielectric constant and conductivity distribution of any given underground medium are input, and the corresponding prediction result is obtained.The purpose of the present application is to solve the technical problems of the existing ground penetrating radar forward technology, such as insufficient accuracy, insufficient modeling of key physical parameter coupling relationship and limited time domain feature expression in complex heterogeneous medium, and the present application is proposed.

The application relates to the technical field of batteries, and provides a battery negative electrode potential evaluation system, a battery negative electrode potential evaluation method and an electric vehicle. The system comprises a data acquisition device configured to acquire battery surface stress data; and a controller configured to: determine a current high stress concentration area and an area of the current high stress concentration area based on current battery surface stress data; determine a current change rate of the area of the high stress concentration area with respect to a battery SOC based on the area of the current high stress concentration area, a current battery SOC, a historical area of the high stress concentration area and a historical battery SOC; and in the case that the current change rate is greater than a preset change rate threshold, substitute the current battery SOC into a mapping relationship model of the SOC and the negative electrode potential to obtain a current battery negative electrode potential. The application does not need to depend on a complex electrochemical model or additional hardware, has low calculation cost, good real-time performance, can be deployed on a large scale at a vehicle end, and is beneficial to early warning of lithium precipitation risk before lithium precipitation occurs.

The application discloses a vehicle control method and device on a transverse slope road, equipment and medium, relates to the field of advanced driving assistance, and comprises the following steps: acquiring the vehicle speed, yaw angular velocity and first lateral acceleration of a vehicle, determining the second lateral acceleration based on the vehicle speed and yaw angular velocity; high-order filtering the first lateral acceleration and the second lateral acceleration respectively; determining the target acceleration based on the filtered first lateral acceleration and the filtered second lateral acceleration, and high-order filtering the target acceleration; determining the transverse slope compensation torque through the filtered target acceleration and a torque compensation coefficient, so as to control the vehicle. The vehicle speed, yaw angular velocity and first lateral acceleration and other information in the application are information that can be acquired in a low-cost manner, therefore, the application reduces the calculation cost of the transverse slope compensation torque, and through high-order filtering of the lateral acceleration and the target acceleration, the application improves the smoothness of the vehicle in a curve and reduces oscillation.

The present application belongs to the field of defect detection, in particular to a pipe network weld flaw detection defect recognition method and system, comprising: acquiring a pipe network weld initial radiographic image, extracting a global feature atlas through a first deep convolution network, and decoding to generate a defect prediction confidence map; extracting a weld key geometric structure, and constructing a geometric prior weight map; applying Bayesian variational inference to the network, statistically dispersing the results of multiple random forward propagation, representing cognitive uncertainty and generating an uncertainty map; pixel-level weighted fusion of the confidence map, the uncertainty map and the geometric prior weight map to obtain a probability heat map, based on which a composite sampling guide vector field is constructed, sampling points are arranged along the vector field, and a multi-angle scanning imaging system is controlled to collect high-resolution local projection data; three-dimensional reconstruction of the projection data to obtain local features, fusion of the local features and global features through a cross-attention module to generate an enhanced defect representation, and output of the class, three-dimensional spatial position and size of the weld defect.

The invention relates to the technical field of digital twinning simulation, in particular to a digital twinning-based temperature and humidity verification box multi-physics field real-time monitoring method, which comprises the following steps of: constructing a digital twinning grid index system, and solving a no-load flow field by using CFD (Computational Fluid Dynamics) simulation to obtain reference turbulent energy; obtaining geometric data of a to-be-detected device, mapping the to-be-detected device to the digital twin grid index system by using a voxelization algorithm, and calculating a thermal inertia correction coefficient; the flow field is iteratively solved based on the thermal inertia correction coefficient, a source item is generated by applying the virtual resistance density and the turbulent energy, and real-time corrected turbulent energy is obtained through calculation; calculating a reliability weight according to the ratio of the real-time corrected turbulent energy of the grid unit where the sensor is located to the reference turbulent energy, wherein the reliability weight is in negative correlation with the ratio; and weighting the monitoring data by using the reliability weight to realize closed-loop control. According to the method, the problem of monitoring misalignment caused by flow field topology change is effectively solved, and the monitoring accuracy is improved.

The invention provides an online multi-axis trajectory planning method and device, electronic equipment and a storage medium. The method comprises the following steps: dividing a parameterized n-axis continuous geometric path into a plurality of grid points on a parameter domain, and defining a state quantity set corresponding to each point; constructing an original optimization problem by taking minimization of the total motion time as a target, introducing an auxiliary variable, converting a nonlinear objective function of the auxiliary variable into a linear objective function about the auxiliary variable, and constructing a linear constraint set about the auxiliary variable and a first motion state quantity related to the motion speed at the same time; constructing and solving a target optimization problem based on the linear target function and the linear constraint set to obtain a state quantity optimal solution corresponding to each grid point; and obtaining a motion track on the geometric path based on the optimal solution. According to the method, the time optimality and the complex constraint are strictly satisfied, the calculation cost is low enough, and the stable online planning capability which is theoretically guaranteed is ensured.

The application provides a three-dimensional anisotropic composite material heat conductivity coefficient prediction method and system, and relates to the technical field of heat analysis, and comprises the following steps: establishing a heat transfer unit cell model of a to-be-detected three-dimensional composite material in different directions based on structure parameters and material parameters of the to-be-detected three-dimensional composite material; the to-be-detected three-dimensional composite material is a three-dimensional orthogonal structure; the material parameters comprise first heat conductivity coefficients of each material; taking solid heat conduction as a current heat transfer form, determining equivalent thermal resistances corresponding to the heat transfer unit cell model of different regions by using an equivalent thermal resistance method and the first heat conductivity coefficients, establishing a heat transfer connection relationship between each equivalent thermal resistance by combining a Fourier equation and a thermal resistance network method, and determining first total thermal resistances along each heat transfer direction of the to-be-detected three-dimensional composite material; the heat transfer connection relationship comprises series connection and parallel connection; converting the first total thermal resistances based on overall size parameters of the heat transfer unit cell model, and determining an equivalent heat conductivity coefficient of a target direction.

This invention provides a target detection method, apparatus, electronic device, and storage medium. The method includes: acquiring echo data received by a vehicle-mounted radar; performing a Fast Fourier Transform (FFT) on the echo data to obtain range-Doppler two-dimensional data; and performing detection processing on the range-Doppler two-dimensional data according to a preset pre-detection tracking model to obtain a target detection result. The preset pre-detection tracking model includes at least a particle filter-based pre-detection tracking model. The preset pre-detection tracking model employs a first preset likelihood ratio function, which includes a background noise amplitude term and a sum term of the target's amplitude and the background noise amplitude. Using this invention can reduce the computational resources consumed by particle filter-based pre-detection tracking algorithms for target detection, thereby reducing its computational cost.

This invention discloses a UAV inspection method and system for bridge bearings. The method includes: planning a strip-shaped flight path and constructing a coarse 3D point cloud model of the bridge; then planning a scanning flight path around the bridge's outer envelope to collect fine point clouds of bridge components; firstly, using a small amount of labeled data combined with pseudo-label generation and knowledge distillation to complete lightweight training of the PointNet model; then, strengthening structural association features through a contextual attention mechanism and generating virtual points using an occlusion region feature completion method to complete the occluded region and obtain a complete bearing point set; finally, extracting bearing geometric parameters by weighted noise point removal, iteratively fitting bounding boxes, and smoothing; planning a flight path based on the bearing's location to collect the bearing's surface image; then performing image cleaning and enhancement processing, and finally identifying surface defects using image recognition methods to generate an inspection report. Comprehensive technical support is provided.

The invention discloses a virtual multi-view fusion millimeter wave point cloud imaging method, which belongs to the field of millimeter wave radars, and realizes three-dimensional point cloud imaging of a target by constructing a virtual observation view angle and fusing perception data of different view angles: establishing a multi-view synthetic aperture radar imaging model; a single-bit compressed sensing framework is introduced to construct a sparse imaging optimization problem, an accurate relay angle is solved by minimizing a model error, and a relay plane model error is corrected; and high-precision 3D point cloud imaging of the region of interest is realized based on the accurate angle. According to the method, a three-dimensional point cloud of a target is constructed by fusing virtual visual angle detection shielding characteristics formed by reflection of a plurality of relay surfaces, joint estimation of relay surface angle prior and 3D point cloud is realized by a nested iteration method, and artifacts and offset caused by prior errors are eliminated; the sparsity and the target continuity of the three-dimensional point cloud can be improved through the additional composite norm constraint, and the accuracy of virtual multi-view 3D point cloud imaging is effectively improved.

The invention discloses a three-dimensional FDTD metal conformal stabilization simulation method. The method comprises the steps of grid construction and boundary identification, effective electromagnetic parameter calculation, stability condition pre-verification, field quantity iteration updating and boundary condition convergence judgment. According to the method, high-precision and high-stability electromagnetic simulation of a complex medium boundary and a metal boundary is realized by combining an integral form of the effective dielectric constant and the magnetic conductivity, a posterior weighted stabilization strategy and a strict integral updating rule of the metal boundary, meanwhile, the robustness of an algorithm to a time step length is improved, and the practical application scene of the method is expanded.

This invention belongs to the field of industrial manufacturing technology, specifically providing a method and system for visual detection of industrial defects by decoupling defect features from imaging conditions. The method includes: collecting defect samples of industrial products under different operating conditions; wherein the defect samples include defect images and corresponding defect annotations; based on the defect samples, constructing a region-guided causal decoupling network model by combining a feature decoupling mechanism guided by defect regions and a causal invariance learning mechanism based on imaging simulation; performing multi-supervised loss joint optimization training on the causal decoupling network model to obtain a converged causal decoupling network model; and detecting defects in industrial products under target operating conditions based on the converged causal decoupling network model to obtain defect detection results. This invention can significantly improve the generalization and robustness of defect detection algorithms in complex industrial environments, enabling rapid and low-cost model transfer between different operating conditions.

The invention belongs to the technical field of fracture mechanics calculation, and particularly relates to a hybrid unit beam structure dynamic fracture phase field calculation method. The method comprises the following steps: determining material parameters required by the dynamic fracture phase field calculation of the hybrid unit beam structure; establishing a hybrid unit beam structure dynamic fracture phase field mathematical model according to the Lagrange action quantity principle, wherein the mathematical model comprises a displacement field control equation and a phase field evolution control equation; numerical discretization is carried out on the mathematical model, and a hybrid unit beam structure discrete model is established; and adopting a Newmark method in combination with an interlacing algorithm to carry out numerical solution on the mathematical model after numerical discretization until the displacement field and the phase field both meet convergence conditions, and obtaining a dynamic fracture phase field evolution result of the hybrid unit beam structure. The problem that an existing fracture phase field model cannot accurately simulate crack evolution of the beam structure on the thickness under the dynamic load is solved, and meanwhile the calculation cost for simulating the dynamic fracture of the beam structure is remarkably reduced.

The embodiment of the present disclosure provides a file security acquisition method, device, medium and electronic equipment, and relates to the technical field of data processing. The method comprises the following steps: obtaining a model training request about a preset task initiated by a requestor for a to-be-trained model, the model training request carrying a training configuration parameter, verifying the model training request, confirming a container group and a resource access key corresponding to the training configuration parameter after the verification is passed, and allocating computing resource and data resource required by the container group; and accessing training corpus in the data resource according to the resource access key in the container group, so as to train the to-be-trained model according to the computing resource and the training corpus. The file security acquisition method adopted by the present disclosure controls the access authority of data through the resource access key, so that the model training is carried out under the condition of ensuring data security, the risk of data and privacy leakage is reduced, and the effect of model training is improved.

The application provides a six-dimensional force sensor optimization method based on CM-PINN, which depends on a six-dimensional force sensor optimization model based on CM-PINN. The application realizes the structured hard constraint of the flexibility matrix by embedding the B matrix projection into the CM-PINN network, replaces the traditional loss function soft constraint, ensures that the mechanical properties relied on in the optimization process always strictly meet the geometric symmetry and the inherent physical law of the flexibility matrix, fundamentally eliminates the physical violation phenomenon from the architecture level, makes the optimization result more consistent with the engineering actual mechanical properties, and improves the optimization reliability; through the design of the load special branch and the symmetric joint training mechanism, the strong high-dimensional nonlinear coupling problem between the elastomer mechanical properties and the external load, the geometric structure parameters is effectively solved. The method can greatly reduce the calculation cost, shorten the design cycle, realize the efficient global optimization of the sensor geometric parameters and the load range, and adapt to the design requirements of the engineering end rapid iteration and rapid verification.

The application relates to a river closure date prediction method based on historical similar year performance driving, which comprises the following steps: using a plurality of feature selection algorithms and leave-one-out cross-validation to screen target predictor sets respectively matched with selected machine learning models and statistical models; based on parameter sensitivity analysis and a Bayesian optimization algorithm, sensitive hyperparameters of the machine learning models are optimized to obtain optimized machine learning models; the statistical models and the optimized machine learning models are configured as candidate river closure date prediction models; a K-neighbor algorithm is used to search a set of similar historical years in a historical observation data set according to current observation data, and a target river closure date prediction model is dynamically optimized according to the comprehensive prediction error of the candidate river closure date prediction models on the set of similar historical years, and a prediction result is output, so that the advantages of multiple models are effectively fused, the generalization limitation of a single model in a complex non-stationary environment is avoided, and the accuracy and robustness of the prediction result are significantly improved.

The invention discloses a corn seed variety identification and contour segmentation method and device based on deep learning, and relates to the technical field of corn seed category identification. The problems that in the prior art, a trained corn variety recognition model is difficult to achieve integrated detection of variety recognition and seed segmentation, tiny differences between similar varieties and morphological changes between seeds of the same variety in a natural placement state are difficult to effectively process, the model parameter quantity is large, and the calculation cost is high are solved. The method comprises the steps that a training data set is constructed, each corn seed image comprises a plurality of corn seeds, an improved YOLOv11 model is constructed, on the basis of the YOLOv11 model, a backbone network is replaced with a ConvNeXt V2 network to optimize feature extraction efficiency and network weight reduction, an image-level variety classification head is newly added, a loss function adopts a double-task collaborative loss function, and the number of the corn seed images is smaller than the number of the corn seed images. Parallel output of image-level variety classification and seed contour segmentation is realized, and an improved YOLOv11 model is trained and applied.

The application provides a zero-shot song conversion method and device, equipment and storage medium, relates to the technical field of audio processing, and includes the following steps: obtaining a source song waveform and a reference song waveform; performing source condition extraction on the source song waveform to obtain a source song condition sequence; performing timbre coding on the reference song waveform to obtain a target timbre sequence; performing flow matching sampling on the source song condition sequence and the target timbre sequence by using a preset diffusion transformer to obtain a target continuous latent sequence; and decoding the target continuous latent sequence to obtain a target song waveform. The application can improve the similarity of song conversion in a zero-shot scenario.

The present application relates to the technical field of bearing life prediction, in particular to a kind of gradient composite coating sliding bearing life prediction method, comprising the following steps: step one, build the coating degradation physical simulation model of multi-field coupling;Step two, adopt active learning algorithm to build high-precision proxy model, design active learning query strategy;Step three, online monitoring and multi-domain feature extraction;Step four, real-time state mapping and life prediction based on transfer learning;Step five, prediction result output and model updating.The present application can improve the accuracy of gradient composite coating sliding bearing life prediction, greatly reduce the operation load, and improve the prediction efficiency of bearing life.

The application discloses a student programming answer prediction method combining semantic understanding and structured modeling, collects student programming homework data and pre-processes the data to obtain data samples, translates the data samples to obtain an English feature set; uses low-rank adaptation to fine-tune a pre-trained semantic understanding model, constructs input features, and predicts an answer correctness probability; performs graph neural network structured modeling based on the association relationship between problems and concepts, performs message propagation and feature updating to obtain an embedding set, inputs the embedding set and student embedding into a prediction layer to obtain a probability of answering a question correctly; and inputs the probability of answering a question correctly and the probability of answering a question correctly after fusion into a multilayer perceptron for nonlinear mapping to predict the probability of answering a question correctly. The application solves the problems of difficulty in effectively processing non-standardized codes submitted by students, insufficient semantic understanding of question texts and low prediction accuracy in programming knowledge tracking, and provides a scientific basis for personalized teaching and learning resource scheduling.

The present application relates to a kind of forest leaf instance segmentation method and system based on deep learning, method includes: obtaining vegetation image, vegetation image is input into leaf instance segmentation model, obtains leaf instance segmentation prediction result;Leaf instance segmentation model is trained using training set;Training set includes: vegetation original image;Feature extraction and enhancement are carried out using backbone module in leaf instance segmentation model, and adaptive spatial fusion mechanism in progressive feature pyramid network is integrated to dynamically adjust feature weight, generate dynamic fusion feature;Through the dynamic asymmetric spatial perception mechanism built-in in dynamic anomaly regression head module, the corresponding multi-source deformation feature layer of dynamic fusion feature is obtained, and the feature fusion strategy of top-down cascaded decoding module is used to optimize multi-scale feature, obtain multi-source fusion feature layer, further using multi-source fusion feature layer, generate leaf instance segmentation prediction result.The present application solves the problems of data scarcity, poor adaptability and low efficiency.

In order to solve the problems of long design time, low efficiency, high calculation cost and low prediction accuracy caused by information loss in the prior art, a resonance super surface design method and device based on spectrum perception improvement are provided.The method comprises the following steps: designing a GLSAT forward prediction network based on spectrum perception improvement;training and optimizing the GLSAT forward prediction network; designing a DNN reverse design network; cascading the DNN reverse design network and the trained GLSAT forward prediction network to obtain a cascaded reverse design network; inputting the ideal spectrum pretreated by GSSG into the cascaded reverse design network, training and optimizing the DNN; and completing the design of the resonance super surface by using the optimized DNN reverse design network.The method has the characteristics of short time consumption, high efficiency and low calculation cost while improving the design prediction accuracy.