63 results about "Multiscale modeling" patented technology

The invention discloses a muck treatment capacity monitoring method and a muck treatment capacity monitoring system, and relates to the technical field of shield construction and muck treatment capacity prediction treatment, and the method comprises the following steps: collecting operation parameters of a shield tunneling machine deslagging and conveying link, calculating a muck flow coupling index, evaluating a deslagging link coupling state, and carrying out optimization; the load coupling index of the shield tunneling machine is calculated, the energy distribution balance is judged, and dynamic adjustment is conducted according to the unbalance state; calculating the energy consumption deslagging propulsion efficiency, evaluating the matching degree, and optimizing the power and conveying parameters of each link; a CNN and LSTM combined multi-scale modeling method is adopted to predict the future muck handling capacity, and online correction is carried out through a sudden change adaptive network; and finally, fusing the standardized data set with the actual operation parameters to generate a unit time muck handling capacity sequence. By means of the method, the deslagging and propelling matching degree of the shield tunneling machine can be improved, the energy utilization efficiency is optimized, and high-precision prediction and real-time regulation and control of the muck handling capacity are achieved.

The invention relates to the technical field of ocean current data analysis, in particular to a multi-scale modeling and verification method and system based on ocean current spring layer structure analysis. And obtaining the thermocline and halocline parameters of the target sea area and carrying out abnormal value elimination and spatial interpolation to form a thermocline data set. A convolutional neural network is used to extract spring layer space structure features, and based on Gaussian curvature quantification boundary complexity, a thermocline and halocline equivalent geometric model is constructed. On the basis, taking the equivalent geometric model as a boundary condition, and performing multi-scale decomposition on the Navier-Stokes equation set to generate a dynamic approximate equation; and then inputting the approximate equation and the equivalent geometric model into a physical information neural network for constraint training to obtain a multi-scale flow field prediction model, and comparing the multi-scale flow field prediction model with actually measured flow field data to complete verification. According to the method, automation, refinement and multi-scale coupling of spring layer structure modeling are achieved, and scientificity and engineering applicability of ocean current modeling are improved.

The invention provides a pathology image report generation method combining hierarchical visual fusion and prototype alignment, and the method comprises the steps: collecting a pathology image and a diagnosis report as a data set, and carrying out the preprocessing of the pathology image in the data set; hierarchical visual context information fusion is carried out by constructing a hierarchical relation matrix; converting a lexical element sequence in a diagnosis report into a decoder target sequence and text embedding, inputting the text embedding and the fused visual features into a prototype mediated cross-modal semantic alignment module, taking the fused visual features and the text embedding as queries, and respectively taking a preset learnable prototype as a key and a value at the same time; training and cross-modal semantic alignment are carried out through a cross attention mechanism, and a pathological image report is generated. According to the method, the multi-scale modeling and semantic understanding capabilities of the model are remarkably improved through a dual mechanism of hierarchical visual fusion and prototype alignment, and a reliable path is provided for automatic pathological auxiliary diagnosis.

The invention discloses a multi-label interference identification method and device based on dual-domain asymmetric feature fusion, and the method comprises the steps: enabling a received interference signal to generate two types of time-frequency images through short-time Fourier transform and continuous wavelet transform, converting the two types of time-frequency images into RGB images, and inputting the RGB images into a dual-branch AsymResNet18FPN network to extract multi-scale features; cross-domain feature adaptive fusion is realized through channel splicing and attention weight generated by MLP; and finally, outputting an interference type combination by the multi-label classifier. According to the method, the complementarity of double-domain features is fully utilized, the direction perception and multi-scale modeling capability is enhanced, the problem of feature submerging under the low interference-to-signal ratio is effectively relieved, synchronous recognition of multiple composite interferences is supported, and the recognition precision and robustness in a complex electromagnetic environment are remarkably improved.

The invention discloses a method for generating a micromechanical property distribution diagram of a solid waste composite cement-based cementing material, and belongs to the technical field of performance prediction of solid waste composite cement-based cementing materials, and the method comprises the following steps: acquiring BSE, EDS and nanoindentation data of a sample; identifying indentation areas in the nanoindentation mark graph, and generating masks corresponding to the areas; performing statistics on the regional masks to obtain element content data of each indentation region; establishing a prediction model by adopting machine learning based on the regional element content and the elasticity data; performing super-pixel segmentation on the BSE image to match the scale of the indentation area; calculating element features of the super-pixel regions, and predicting the elastic modulus of each super-pixel region; and mapping the predicted elastic modulus to a corresponding region to generate an elastic modulus distribution diagram. According to the method, prediction of the two-dimensional space distribution of the internal elastic modulus of the solid waste composite cement-based cementing material can be realized, and data and visual support are provided for material heterogeneity research and multi-scale modeling.

The invention relates to a zero sample target counting device, and provides a technical scheme of fusing a fine-grained cross-modal enhancement module (FCE) and a multi-scale semantic adaptive module (MSA) in order to solve the problems of large counting error and poor generalization caused by insufficient cross-modal alignment precision and weak multi-scale modeling capability in a dense scene. A text-guided cross attention and self-adaptive gating mechanism is introduced through an FCE module, so that fine-grained semantic alignment of images and texts is realized, and the discrimination of visual features and the semantic perception capability are enhanced; multi-level visual features are fused through an MSA module, dynamic modulation is carried out in combination with text semantics, self-adaptive modeling of small targets and multi-scale targets is achieved, and high-precision target instance detection and number estimation are completed in the cross-modal decoding stage. The method is obviously superior to an existing model on multiple data sets without fine tuning, and the counting precision, robustness and open world generalization ability in a complex scene are improved.

The invention relates to the technical field of computer vision, in particular to a small target fine segmentation method fusing improved pyramid pooling and edge branch supervision, and the method comprises the steps: selecting a data set, carrying out the preprocessing of a data set image, carrying out the feature extraction of the preprocessed image through a backbone network MobilenetV3, and generating four sets of feature maps F1 to F4, the method comprises the steps of inputting F1 and F2 into an ABG module, generating feature maps F5 and F4 containing edge information, sending the feature maps F5 and F4 into an improved ASPP structure, obtaining a feature map F6 containing multi-scale context information, inputting F3 and F4 into an FCE module for feature fusion so as to supplement spatial details, finally splicing and fusing F5 and F6, enhancing through the FCE module, and gradually performing up-sampling back to the original size in combination with the supervision of the feature maps F3 and F4, so as to obtain the multi-scale contextual information. And a final segmented image is obtained. According to the invention, through integration of the segmentation network, full-link optimization of'feature extraction-edge enhancement-multi-scale modeling-feature calibration-precise decoding 'is realized.

The invention discloses a multi-scale modeling and simulation prediction method and device for a two-phase composite material, and the method comprises the following steps: S1, carrying out the molecular dynamics simulation of the penetration process of a solution in the composite material based on molecular dynamics, extracting the microscopic information in the simulation process, and obtaining a microstructure model of the composite material; s2, mapping the microstructure model to a finite element grid, and reserving structures between particles and interfaces to obtain a finite element model; s3, inserting a cohesive zone unit into an interface of the finite element model, and modeling an interface adhesion and separation phenomenon between a matrix and particles of the composite material based on the cohesive zone unit; and S4, carrying out analogue simulation on the whole damage evolution process based on the finite element model to realize mechanical simulation analysis and prediction. According to the method, composite material modeling is carried out based on a real microstructure, and more accurate mechanical simulation analysis is carried out based on the modeling.

The application discloses a bearing fault noise resistance diagnosis method based on IPDSCS and a Swin Transformer, and comprises the following steps: collecting vibration signals of a bearing under different working conditions to obtain multiple groups of original vibration signal data; performing wavelet transform on the original vibration signal data to convert the original vibration signal data into time-frequency images; constructing an improved inverted-pyramid deep separable convolution sequence feature extraction model to extract the time-frequency images to obtain multi-dimensional feature vectors; taking the feature vectors as input data, performing model training based on a Swin Transformer deep learning network, and establishing a bearing fault diagnosis model; and inputting bearing vibration signals to be diagnosed into the bearing fault diagnosis model to output corresponding fault categories. The method improves the noise resistance of signal processing through wavelet transform denoising and an IPDSCS feature extraction method; multi-scale modeling is performed on multi-dimensional features by using a Swin Transformer network, and the precision and robustness of fault diagnosis are significantly improved; and high-precision diagnosis of bearing faults is realized.

The present application belongs to the field of image data processing, and provides a multi-level concrete crack detection method based on unmanned aerial vehicles. In the acquisition stage, an edge computing platform based on unmanned aerial vehicles and a target detection model is used to perform real-time target detection on the surface of the building structure, so as to realize rapid identification of suspected crack areas and task-driven image acquisition. In the fine detection stage, a multi-scale modeling strategy is used, which integrates a multi-scale perception path of coarse-grained region screening, fine-grained boundary enhancement and post-processing optimization segmentation. A mixed attention mechanism MLCA is introduced to enhance the focusing ability of the model on detailed features and the semantic robustness in complex backgrounds. The overall consideration of the real-time requirements of the inspection task and the accuracy of the pixel-level semantic modeling provides theoretical support and method foundation for intelligent identification of large structure surface cracks in high-resolution scenarios.

The invention discloses a defect lightweight detection model and method based on multi-scale dynamic fusion and application, and the model comprises the steps: constructing a multi-scale feature enhancement module, improving the receptive field and multi-scale modeling capability through large kernel convolution, multi-scale cavity convolution and structural re-parameterization, and combining with a multi-scale-cavity rate mapping mechanism to obtain a multi-scale feature enhancement module; through predefined cavity convolution configuration, a network receptive field can quantitatively cover defect areas with different physical sizes, and self-adaptive matching of the cavity rate and the defect scale is realized. According to the invention, an adaptive feature fusion module is optimized, context sensing feature fusion is realized again through dynamic multi-path convolution and introduction of a convolution gating mechanism, the calculation overhead is reduced while the precision is maintained, lightweight deployment is realized, and the method is suitable for large-scale popularization and application. The technical problems that in an existing wheel set tread defect detection model, the small-scale defect omission ratio is high, complex working condition false detection is frequent, and light weight and precision are difficult to balance are solved.

The invention discloses a cerebrovascular reactivity evaluation method for non-invasive calculation of a breath-holding index, and belongs to the field of hemodynamics numerical calculation and multi-scale modeling. The method comprises the following steps: constructing a one-dimensional (1D) arteriole hemodynamic model based on an allotactic scale law; constructing a main artery three-dimensional (3D) model, and coupling the main artery three-dimensional (3D) model with the arteriole one-dimensional model and the microcirculation zero-dimensional (0D) model to form a whole-brain 3D-1D-0D geometric multi-scale hemodynamic model; based on blood flow related data obtained in the experiment process, parameterization description is conducted on arteriole radius changes and microcirculation resistance changes; a breath holding index (BHI) is calculated by numerical solution of the geometric multi-scale model. According to the method, the non-invasive numerical calculation of the breath-holding index can be realized without directly implementing a breath-holding experiment or an external stimulation condition, and a stable and repeatable calculation means is provided for quantitative analysis and related engineering application of cerebrovascular hemodynamic behaviors.

The application discloses a protein sequence feature representation method based on multi-scale modeling, which comprises the following steps: (1) passing a protein sequence through a ProtBert model to obtain a protein initial feature matrix; (2) using a multi-size sliding window on the protein initial feature matrix, respectively assigning attention coefficients to residues at different positions in the window, and calculating a protein attention feature matrix through an attention mechanism; and (3) respectively performing convolution on the protein attention feature matrix using multi-scale convolution kernels to finally obtain a protein sequence feature matrix based on the attention mechanism and the convolution network.

The application belongs to the field of finite element modeling, and particularly relates to a finite element modeling method considering multi-scale characteristics of an assembly interface, structural data of a rough surface is collected, a rough surface height matrix is obtained after convolution coefficient calculation; an ideal surface corresponding to the rough surface is determined, and geometric information of the ideal surface is obtained; the rough surface height matrix is modeled and superimposed on the ideal surface to obtain superimposed rough surface information; finite element modeling is performed on different superimposed rough surface information of the multi-scale surface to obtain a finite element model containing multi-scale characteristics of the assembly interface. The technical problem that existing finite element modeling software and tools cannot directly realize multi-scale modeling of the assembly interface and can only establish a model based on ideal product size, resulting in an ideal smooth contact state of the assembly interface, is solved, and the multi-scale characteristics of the assembly interface are successfully included in the product finite element model, so that the model can contain multi-scale geometric information of the actual surface of the part.

The application relates to a kind of robot end effector multiscale clamping force control device and method based on DIC, wherein the method comprises the following steps: S1, obtaining multiscale signal, collecting multiscale speckle image: S2, processing multiscale data and extracting features, image is processed by power platform multi-thread;S3, mapping multiscale physical quantity and constructing model, power platform is associated with multiscale feature and parameter and builds layered model;S4, control multiscale closed loop and adjust real-time, power platform uses layered adaptive algorithm to adjust PWM parameter.The application can consider non-contact sensing, multiscale modeling and real-time control, and break the force control bottleneck of robot soft body end effector.

The invention discloses a steel cable defect detection method based on a CIA-YOLO model of YOLOv11, belongs to the technical field of machine vision and defect identification, and provides a method for fusing a CBAM attention mechanism in YOLOv11 to enhance the feature extraction capability of small-size defects in order to solve the problems of small feature scale, fuzzy boundary and the like in steel cable defects. A dynamically-scaled Inner-IoU loss function is introduced to improve the bounding box positioning precision, and an AKConv structure is optimized and combined with an improved C3k2 module to improve the multi-scale modeling effect. On a broken wire defect data set, a corrosion defect data set and a wear defect data set, CIA-YOLO is superior to an original model in the aspects of mAP and Recall, and especially has a remarkable effect in small defect identification. The method is suitable for real-time online detection of steel cable defects and has a good application prospect.

The present application belongs to the field of multi-scale modeling, and particularly relates to a parameterized modeling method for particles adhering to a rough surface; the steps are as follows: firstly, input modeling parameters, generate and calibrate an original rough surface model to meet a target roughness; then input a particle adhering set containing parameters such as particle size, protruding height and target coverage rate, and determine the number and center position of particles; further, embed the particle geometry into the original surface, and after reconstruction, perform feasibility verification on the reconstructed model in terms of geometric interference, coverage rate consistency and mesh quality; if the verification fails, automatically perform repair and cyclic verification until a preset criterion is met, and finally output a finite element input file; through the orderly connection of parameter driving and automatic repair, the present application realizes full-process automation from surface generation to particle layout, effectively avoids the risks of geometric self-intersection and mesh failure caused by manual operation, and significantly improves the stability and batch construction capability of the model.

The application discloses a kind of reconstruction method and system of marine internal wave three-dimensional structure, which comprises the following steps: obtaining multi-modal original remote sensing data and carrying out data perception to obtain multi-modal feature map;Obtain the physical characteristic vector of temperature-salinity profile and combine multi-modal feature map, carry out physical feature extraction and multi-scale modeling, obtain preliminary three-dimensional temperature-salinity profile and stratification parameter distribution;Three-dimensional structure reconstruction and uncertainty quantification are carried out on multi-modal feature map and preliminary three-dimensional temperature-salinity profile and stratification parameter distribution, to obtain the final full-water-depth three-dimensional grid field;Based on the final full-water-depth three-dimensional grid field, digital twin visualization and interaction are carried out, to realize internal wave three-dimensional structure reconstruction.The application can map two-dimensional sea surface observation signal to full-water-depth three-dimensional internal wave field with physical consistency.The application can be widely applied to the field of internal wave three-dimensional structure reconstruction as a kind of reconstruction method and system of marine internal wave three-dimensional structure.

The application discloses a kind of overburden and surface fissure network reconstruction and inversion method for similar simulation experiment, it is related to network reconstruction and inversion technical field, including: carrying out simulation test, laying experimental model;Carrying out test, collecting three-dimensional point cloud data;Data is preprocessed, and feature vector is extracted;Based on feature vector reconstruction point cloud, get fissure feature, reconstruct fissure network;Optimize fissure network, carry out three-dimensional visualization and analysis;Model is imported into simulation software, iteratively optimized parameter, and the evolution process of fissure is inverted.The application uses three-dimensional point cloud imaging technology to realize high-precision data acquisition and processing, accurately extracts fissure feature;Based on fault image or point cloud data, three-dimensional reconstruction algorithm is used to reconstruct the three-dimensional model of fissure network, and multi-scale modeling is supported;Through numerical simulation, the evolution process of fissure is inverted, combined with machine learning algorithm to optimize inversion model, improve prediction accuracy.

The application provides a typhoon path prediction method and system fusing physical constraints and path mutation identification. The method comprises the following steps: obtaining a multidimensional input feature tensor through a feature tensor generation module; performing time modeling through a multiscale modeling module to obtain a deep feature tensor, performing significance guidance through a significance guidance module to obtain a prediction path sequence; identifying a bending angle through a bending event identification module to obtain a plurality of path anomaly scores, introducing a plurality of preset disturbances through a counter disturbance analysis module to determine path confidence, and outputting a path prediction result through a path prediction output module. According to the technical scheme of the embodiment of the application, the typhoon path can be finely modeled, the PINNs physical constraint module is used to ensure that the prediction result meets the physical conservation principle, the bending event identification is used to enhance the path mutation detection capability, the disturbance analysis is used to realize the uncertainty evaluation, and the typhoon path prediction result is more accurate, reliable and physically consistent.

The invention relates to the technical field of power system oscillation analysis and control, and provides a multi-scale modeling-based broadband oscillation feature extraction and suppression method for a new energy high-proportion access system, and the method comprises the steps: building a multi-scale model which is used for describing a power system and comprises an electromagnetic transient model, an electromechanical transient model and a dynamic feature model; based on the multi-scale model, constructing a digital twinborn model which operates synchronously with the physical power system, and updating the state of the digital twinborn model in real time by adopting a data assimilation algorithm; utilizing a digital twinborn model to extract broadband oscillation characteristics of the power system; and generating a suppression strategy according to the broadband oscillation characteristics, and regulating and controlling the physical power system through the suppression strategy verified in the digital twinborn model. Furthermore, accurate extraction and effective suppression of broadband oscillation in the new energy high-proportion access system are realized.

The application discloses a kind of multi-scale modeling and fracture simulation method for compressed air energy storage cover layer integrity evaluation, steps include: constructing multi-scale heterogeneous geomechanics model;Establish the fatigue damage constitutive relationship under rock loading and unloading condition;Establish fracture simulation control equation;Multi-scale calculation strategy and local adaptive grid are constructed;Cyclic loading condition simulation and integrity quantitative evaluation are carried out.The application is designed for the core damage mode of CAES gas storage low pressure fatigue, compared with general structure analysis software, the pertinence and accuracy of cover layer integrity evaluation are higher, the heterogeneous geological characteristics of cover layer are truly reflected, the process of crack initiation and propagation in cover layer under cyclic injection and production conditions is accurately simulated, and the advanced prediction and quantitative evaluation of cover layer integrity failure risk are realized.

The invention discloses a switch hand feeling multi-scale modeling method based on a physical information neural network, and the method comprises the steps: firstly building a one-dimensional nonlinear elastic damping partial differential equation PDE and an ordinary differential equation ODE which describe the overall deformation behavior of a key, and forming a hybrid PDE-ODE kinetic model; secondly, according to the dynamic model, designing a Bayesian physical information neural network B-PINN with an uncertainty quantification capability; and then the data sampling density is dynamically improved in a mutation region, training weight distribution is adjusted by adopting an event-driven strategy, and the adaptive training capability of the neural network B-PINN is enhanced. And finally, generating an elastic modulus and damping coefficient probability distribution map by using the trained network model, mapping the subjective hand feeling score of the user into a mechanical objective function, and inverting parameters. According to the method, the defects of poor generalization and insufficient physical consistency of a pure data driving method are overcome, and the reliability of switch design parameters is improved.

The application discloses a kind of sludge treatment quantity monitoring method and system, relating to shield construction and sludge treatment quantity prediction processing technical field, the method is by collecting the operation parameter of shield machine out of slag and conveying link, calculates sludge flow coupling index, assesses out of slag link coupling state and carries out optimization;Calculate shield machine load coupling index and judge energy distribution balance, for imbalance state carries out dynamic adjustment;Calculate energy consumption out of slag propulsion efficiency and carry out matching degree evaluation, optimize each link power and conveying parameter;Using the multiscale modeling method of combination of CNN and LSTM predicts future sludge treatment quantity, and carries out online correction by mutation adaptive network;Finally, standardized data set is fused with actual operation parameter, generates unit time sludge treatment quantity sequence.This method can improve shield machine out of slag and propulsion matching degree, optimize energy utilization efficiency, and realize high-precision prediction and real-time regulation of sludge treatment quantity.

The invention provides a multi-scale modeling power generation power assembly behavior recognition control system and method, and relates to the technical field of generator control. Taking the dynamic behavior feature library as a priori knowledge constraint, and constructing a multi-scale behavior classification model; performing multi-mode operation sensing to obtain a multi-physical field coupling signal; performing multi-scale feature extraction based on multi-scale frequency domain time-frequency domain features to obtain a layered feature set; dynamic working condition intelligent identification is carried out, and a multi-scale working condition state identification result is output; cooperative control optimization iteration based on model simulation is carried out, and a hierarchical cooperative control strategy is output; and full-working-condition self-adaptive closed-loop control is carried out. The technical problem that in the prior art, power generation power assembly control mostly depends on a model of a single time scale to conduct operation management, multi-physical field interaction behaviors under the complex load condition cannot be effectively processed, and consequently the power generation power assembly cannot make quick and accurate response under the complex working condition is solved.

This invention proposes a hyperspectral image classification method based on a multi-scale spiral scanning spatial-spectral dual-branch Mamba module, which mainly consists of four modules: a hyperspectral preprocessing module, a multi-scale spiral scanning spatial Mamba module, a bidirectional scanning spectral Mamba module, and an adaptive enhancement fusion module. It adopts a spatial-spectral dual-branch structure and achieves model lightweighting through a single Mamba layer. Specifically, the multi-scale spiral scanning spatial Mamba module integrates lightweight multi-scale convolution and center-first spiral scanning to effectively extract spatial features. The bidirectional scanning spectral Mamba module overcomes the unidirectionality of the state-space model through a bidirectional scanning mechanism, fully extracting spectral features. The adaptive enhancement fusion module achieves complementary enhancement and fusion of the two modules. This method systematically solves the problems of insufficient multi-scale modeling capability, high computational complexity, and poor adaptability to targets of different sizes in standard Mamba, achieving a significant improvement in classification accuracy and effectively reducing the computational burden.

The invention discloses a self-adaptive multi-scale seismic high-resolution processing method fusing time-frequency characteristics, and belongs to the technical field of seismic exploration data processing. According to the method, an encoder-decoder network architecture based on Transform is constructed, continuous wavelet transform is combined to realize time-frequency feature fusion, a self-adaptive multi-scale division strategy based on K-Means energy clustering and a double attention mechanism are adopted to capture local and global dependency relationships of seismic data, and finally, the resolution of the seismic data is improved. According to the method, the problems that a traditional method excessively depends on geological hypothesis and professional knowledge and an existing deep learning method is insufficient in time-frequency feature utilization and multi-scale modeling are solved, and the method has higher generalization ability and migration ability and is suitable for seismic data processing of complex geological structures.

This invention discloses an adaptive multi-scale high-resolution seismic processing method that integrates time-frequency features, belonging to the field of seismic exploration data processing technology. This method constructs a Transformer-based encoder-decoder network architecture, combines continuous wavelet transform to achieve time-frequency feature fusion, and employs an adaptive multi-scale partitioning strategy based on K-Means energy clustering and a dual attention mechanism to capture local and global dependencies in seismic data, ultimately improving the resolution of the seismic data. This invention addresses the over-reliance of traditional methods on geological assumptions and professional knowledge, and the shortcomings of existing deep learning methods in utilizing time-frequency features and multi-scale modeling. It possesses stronger generalization and transfer capabilities, making it suitable for processing seismic data with complex geological structures.

The invention discloses a three-dimensional ground penetrating radar dispersive medium discontinuous finite element forward modeling method and system and a storage medium, and the method comprises the following steps: S1, building a forward modeling model, and setting parameters at a boundary; s2, inputting parameters, and setting the positions of an excitation point and a receiving point; s3, performing mesh generation on the three-dimensional dispersive medium model; s4, solving numerical flux of six components of the three-dimensional electromagnetic field; s5, updating an auxiliary field variable of the three-dimensional area; s6, updating electromagnetic field components of the whole region; s7, time steps are added, and the steps are repeated until three-dimensional forward modeling simulation of the current time step is completed; s8, repeating the steps until all excitation is completed; and S9, outputting the data. According to the method, the three-dimensional full-region electromagnetic field evolution calculation logic is optimized, and the multi-scale modeling requirement of a large-scale complex underground structure is met. The finally output three-dimensional radar profile can present the depth, form and spatial distribution of the underground target in a three-dimensional manner, and detection blind areas are reduced.

The application discloses a short-impending precipitation prediction method based on a CAMS-Unet model, and comprises the following steps: acquiring original radar echo image data and performing pretreatment; constructing a CAMS-Unet short-impending precipitation prediction model; based on target radar echo image data, training and optimizing the CAMS-Unet short-impending precipitation prediction model to obtain a target CAMS-Unet short-impending precipitation prediction model; acquiring radar echo image to be predicted and inputting the radar echo image to be predicted into the target CAMS-Unet short-impending precipitation prediction model to output a predicted radar echo image. The radar echo image is pretreated through three stages of data set screening, abnormal value and vacancy value processing, denoising and data enhancement processing. Through the construction of the CAMS-Unet model comprising a global feature extraction module CAMM and a local feature extraction module MSFM, the problems of multi-scale modeling imbalance and insufficient capture of space-time heterogeneity are solved.