309results about How to "Prevent overfitting" patented technology

The invention provides a multi-modal large model construction and operation method and system for reservoir dam safety. The method comprises the following steps: constructing a training set, wherein the training set comprises a single sample; multi-modal data as model input: sensor data arranged on the dam, historical hidden danger records, engineering technology documents formed in various stages of planning, design, construction and operation of the dam, information resources related to dam safety, geometric and material information of the BIM model, timestamps and spatial position information; the probability value of occurrence of the hidden danger event serving as the training label is obtained; the training set is adopted to train a multi-mode pre-training model based on a Transform framework; performing real-time prediction and early warning on potential safety hazards of the dam; and calculating the actual influence degree, the comprehensive risk score, the dynamically set early warning threshold and the emergency measure suggestion of the hidden danger event according to the probability value of whether the hidden danger event occurs or not output by the model. The scientificity and effectiveness of dam safety monitoring are comprehensively improved.

The invention relates to the technical field of model construction, and discloses a neural network-based frozen soil unfrozen water content detection model construction method and system. The method comprises the following steps: carrying out segmented sampling test on a clay sample in a temperature interval of 0 DEG C to-20 DEG C to obtain a multi-dimensional original data set, and carrying out standardization and principal component analysis on the data set to obtain a standardized feature vector; and performing three-stage classification marking on the feature vectors to obtain a training data matrix. A deep neural network is adopted to carry out five-fold cross validation training, and a prediction model of three hidden layers is constructed. And performing correction calculation on model output according to soil mineral composition and porosity parameters, and finally identifying an abnormal value through root-mean-square error analysis and a three-time standard deviation principle to obtain a standardized prediction result. According to the method, the mapping relation between the thermal parameters and the unfrozen water content is established, correction is carried out in combination with the soil body characteristic parameters, and the technical problem that the prediction precision of the unfrozen water content is not high in the prior art can be solved.

The invention discloses a large language model parameter fine tuning method, system and device, a storage medium and a product, and relates to the field of natural language processing. The fine tuning method comprises the steps that a pre-training model and a parameter matrix of the pre-training model are acquired, a parameter fine tuning module is arranged on a self-attention layer of the pre-training model, the parameter fine tuning module comprises LoRA modules of different task types and a router, each LoRA module comprises a general feature matrix and a specific feature matrix, and the general feature matrixes of the multiple LoRA modules are shared; obtaining a fine tuning data set; loading and freezing a parameter matrix of the pre-training model, and initializing parameters of a parameter fine tuning module; and performing fine tuning on the parameters of the parameter fine tuning module based on the fine tuning data set to obtain a fine-tuned large language model. Parameters of the MoE architecture are remarkably reduced, it is ensured that the model captures differences of various tasks to the maximum extent, and the cross-task generalization of the model is ensured.

The invention provides a zero-sample dynamic MRI reconstruction method based on global and local diffusion models, and the method comprises the following steps: generating an image data set based on original two-dimensional dynamic full-sampling heart magnetic resonance multi-coil data; designing a global and local time interlaced sampling strategy to construct two types of training data sets for generating a network model by adopting global and local cascade learning networks of a two-stage learning mechanism; performing prediction, correction, low-rank constraint and data consistency on data alternation by using an iterative solution algorithm for rapidly extracting prior information of a network model; and applying an iterative solution algorithm to the trained network model, testing, outputting a dynamic magnetic resonance reconstruction image, and evaluating the image quality. According to the method, the complete coding data is constructed through different time interlaced sampling strategies, the image is quickly and accurately reconstructed under the under-sampling condition in combination with the image prior information of the network model, and the quality and efficiency of the reconstructed image are remarkably improved.

The invention provides a hysteresis loop prediction model establishment method and device, electronic equipment and a storage medium, and the method comprises the steps: carrying out a hysteresis characteristic experiment on a ferromagnetic material, setting hysteresis loop related characteristics comprising magnetic induction intensity waveform data and magnetic field waveform data under the excitation of preset sine waves under different magnetic induction intensity peak values, establishing a hysteresis loop prediction model comprising an improved convolutional neural network model and a K-nearest neighbor algorithm model, training the improved convolutional neural network model, inputting a magnetic induction intensity peak value and magnetic induction intensity waveform data, and outputting magnetic field waveform data; and adjusting the magnetic field waveform data based on the feature data by using a K-nearest neighbor algorithm model to obtain a final magnetic field waveform result. According to the method, the hysteresis loop prediction model is established based on the improved convolutional neural network in combination with the K nearest neighbor algorithm, the overfitting phenomenon can be prevented, the training speed is higher, the resource consumption is lower, and rapid iteration and adjustment in practical application are facilitated.

The invention relates to a transfer learning-based power distribution network measurement data super-resolution sensing method. The method comprises the steps that a high-frequency time sequence is acquired and preprocessed, and training sample data is obtained; a cascade model is constructed based on a graph attention network and a transformer network, the training sample data and an incidence matrix are input into the cascade model, the cascade model is pre-trained, and the incidence matrix represents the topological connection relation of the power distribution network; the cascade model after pre-training is divided into a learning device part and a decision-making device part, transfer learning training is carried out, model parameters reaching a training target are reserved, a final super-resolution perception model is obtained, the learning device part comprises all layers except a transformer linear output layer, and the decision-making device part comprises a transformer linear output layer; and inputting real-time measurement data into the super-resolution sensing model to obtain a super-resolution measurement generation result.

The invention provides a hydropower station floating object detection method based on global relation modeling, and aims to improve the detection precision and efficiency of hydropower station water surface floating objects, and the method comprises the steps: collecting hydropower station water surface scene data through a multi-view device; extracting shallow multi-scale features through a lightweight convolutional network; modeling a global relationship between the floaters through a multi-head self-attention mechanism; carrying out the classification and positioning of floating objects through a target detection head; network training is performed through a multi-task loss function to optimize the model; and reasoning by using the trained floating object detection network, inputting a to-be-detected water surface image into the network, outputting the category and the position of the floating object, and finally generating a detection result. According to the method, the global relation modeling method is introduced, the detection precision of the hydropower station water surface floating objects is effectively improved, detection errors caused by insufficient local features in a traditional method are reduced, and the method has low calculation complexity and is suitable for real-time detection in practical application.

The invention provides a Chinese electronic medical record named entity recognition method based on a multi-head attention mechanism and a pulse neural network (SNN), and the method comprises the steps: obtaining a dynamic word vector through a pre-training BERT model, obtaining a text code through a bidirectional long short-term memory network (BiLSTM) model, and screening the features of the code through the SNN network for a named entity in an electronic medical record; the multi-head attention mechanism calculates the dependency relationship between characters, enhances the relevance to the context, and finally outputs a final sequence through a conditional random field (CRF) module to obtain an entity recognition result. According to the method, the time cost is remarkably reduced, the efficiency and precision of identifying the named entities in the medical text are improved, and a more efficient and accurate technical solution is provided for the field of medical information processing.

The invention belongs to the technical field of medicine supply, and particularly relates to a personalized federal learning-based medicine purchase quantity prediction method, which realizes model data sharing on the basis of protecting the problem of user privacy disclosure. The drug purchase quantity prediction method comprises the following steps: S1, determining hospitals and drugstores participating in a region, and collecting and preprocessing data; s2, selecting and constructing a federal learning model, wherein the constructed model is a DCNN-LSTM hybrid model; s3, performing training, parameter cutting and transmission, aggregation and optimization on the model constructed in the step S2, wherein the data is not out of the local; s4, performing prediction and application on the optimization model obtained in the step S3; and S5, supply chain adjustment. According to the method, doctor-patient information leakage is prevented, information communication overhead is reduced, development of a medical intelligent system is promoted, and key support is provided for promoting sustainable development of the intelligent medical system and further development of the artificial intelligence industry by utilizing the federated learning technology and accurately predicting the drug demand quantity.

The invention belongs to the field of CT image processing, and particularly relates to a CT image reconstruction method based on a three-dimensional diffusion model and a related device.The method comprises the steps that a medical CT scanner is adopted to obtain three-dimensional CT voxel data, and the three-dimensional CT voxel data is preprocessed; extracting a medical object surface point cloud from the preprocessed three-dimensional CT voxel data by adopting a voxel-level reconstruction algorithm, and generating three-dimensional sparse point cloud data; processing the three-dimensional sparse point cloud data by adopting a 3D Gaussian splashing technology, and converting the three-dimensional sparse point cloud data into three-dimensional Gaussian distribution data by using a pre-trained three-dimensional Gaussian kernel; gradually adding Gaussian noise to the three-dimensional Gaussian distribution data by adopting a three-dimensional diffusion model, and gradually removing the Gaussian noise to generate reconstructed three-dimensional Gaussian distribution data; and projecting the reconstructed three-dimensional Gaussian distribution data to a two-dimensional plane by adopting a rendering technology to generate a two-dimensional CT reconstructed image. According to the method, the advantages of the 3D Gaussian splashing technology and the three-dimensional diffusion model are combined, efficient, real-time and denoising medical image reconstruction can be achieved under the low-dose CT imaging condition, and meanwhile the problem of white floater artifacts possibly occurring in the reconstruction process is effectively solved.

The invention discloses an automatic driving system extreme environment decision-making method based on a neural evolutionary game. The method comprises the following steps: S1, collecting and constructing a real-time environment data set around a vehicle through an automatic driving system; s2, generating an environment feature vector; s3, based on the environment feature vector, analyzing the environment feature data through an extreme environment identification module; s4, activating a corresponding multi-target dynamic reward mechanism based on the extreme environment label, and setting a dynamic reward weight for the neural evolution process; s5, gradually evolving to generate an optimal neural network structure and parameter configuration suitable for different extreme environments; s6, generating a multi-target game strategy in combination with the environment feature vector; s7, generating an optimal driving path, vehicle speed adjustment, obstacle avoidance mode and direction control decision; and S8, inputting real-time feedback data into the neural evolution model, and further adjusting the structure and parameter configuration of the neural network. The automatic driving system can make a safer, more accurate and more efficient driving decision in an extreme environment.

The invention provides a pipeline defect assessment method and system based on multi-source data fusion, and the method comprises the following steps: collecting the data of the operation state of a pipeline body through a related collection device, and marking the state and fault type of the pipeline, so as to construct a training set and a test set; and preprocessing the static data of the pipeline, wherein the preprocessing comprises data cleaning, standardization and feature extraction. And optimizing the extreme learning machine by using a bird flock algorithm, analyzing static data and evaluating pipeline defects. Preprocessing the image data of the pipeline body, wherein the preprocessing comprises image conversion, enhancement and denoising; and evaluating the circumferential weld defect based on the pre-trained VGG16 model. And performing decision-level fusion on a defect evaluation conclusion obtained by static data analysis and circumferential weld image analysis by using a D-S evidence reasoning algorithm to obtain a final evaluation result. The method effectively improves the accuracy and reliability of pipeline defect evaluation, and has wide application prospects and important practical value.

The invention provides a grounding grid corrosion defect identification method based on an edge enhancement model, a medium and equipment, and relates to the technical field of grounding grid corrosion defect identification. On the basis of a traditional LeNet-5 network structure, two channels with different scales are connected in parallel to form a three-channel parallel module, macroscopic and microscopic information on different scales can be extracted, and the mutual relation between the macroscopic and microscopic information can be captured through cross-scale convolution, so that the information between the macroscopic level and the microscopic level is fused, and the information fusion efficiency is improved. The expression capability of the network on corrosion defects in different shapes is improved; by adding the edge extraction module, a traditional full connection layer is improved, network parameters can be greatly reduced, the overfitting problem is effectively avoided, meanwhile, the network structure is simplified, and the identification accuracy of the corrosion defect of the grounding grid is effectively improved.

The invention relates to the technical field of remote sensing data processing and earth surface parameter inversion, and discloses an earth surface matrix parameter hyperspectral data inversion method and system based on machine learning. Comprising the following steps: constructing a multi-source heterogeneous hyperspectral data set; performing feature screening on the preprocessed hyperspectral data set based on an adaptive band selection algorithm, constructing a dynamic weight matrix by calculating mutual information entropy and inter-class distance measurement between spectral bands to realize intelligent screening of key feature bands, and combining spectral derivative conversion and spectral index calculation to generate an enhanced feature vector; and a multi-task transfer learning neural network model is constructed, and an output layer realizes multi-parameter collaborative inversion based on a multi-task learning architecture. And performing preprocessing and feature enhancement operation which is the same as that of the training data on the hyperspectral image data of the to-be-inverted region, inputting the trained neural network model, and outputting a surface matrix parameter inversion result.

The invention discloses a CNN-XGBoost power load prediction method based on an attention mechanism, and the method comprises the steps: obtaining related historical load data and feature data, carrying out the data correlation analysis, employing the CNN to construct a feature extraction module based on the advantages of the CNN in feature extraction, and carrying out the calculation of the data correlation. Aiming at the problem that a traditional convolutional neural network cannot distinguish whether the extracted features are effective for prediction, Attention is introduced, different weights are given to the features, the features with different output contribution degrees are distinguished, the extracted features are emphasized consciously, the relation between the extracted features and labels is explored, and an XGBoost algorithm is combined to determine whether the extracted features are effective for prediction or not. According to the method, the advantages of the CNN algorithm in the aspect of feature extraction and the advantages of the XGBoost algorithm of high nonlinear load prediction precision and high prediction convergence speed can be integrated, and the accuracy and stability of short-term load prediction of power consumption of the whole network are improved.

The invention provides a skin state classification method based on progressive transfer learning and modal domain adaptation, relates to the technical field of skin state recognition based on a neural network, and aims to realize a skin health state classification method which is accurate and efficient and is not limited by labeled data volume and not influenced by clinical modals. Comprising the following steps: constructing a plurality of intermediate data sets through unlabeled samples in different types of skin state fields; adjusting color space and feature distribution of the plurality of intermediate data sets based on HSV color space conversion and a CycleGAN model; on the basis of modal domain adaptation, a progressive transfer learning strategy is adopted, parameters of an identification model are finely adjusted step by step through a plurality of intermediate data sets, and data sources of unlabeled and label-free skin images of different sources and different types are adopted as labels of the skin images; and adjusting the identification model according to the marked target data set through a self-adaptive unfreezing method. The method has the advantages of improving the classification accuracy and reducing the dependence on the marked data.

The invention belongs to the technical field of information science, and particularly relates to a disease survival prediction method based on a multi-scale pathological image and multi-modal data, which comprises the following specific steps: digital pathological image data processing, gene representation data processing, multi-modal data interaction and survival analysis. According to the method, image blocks corresponding to two magnification factors of x40 and x20 are used as input of a model, and integration of multi-scale information is realized by adopting an early fusion strategy; compared with a method of repeatedly calculating the same position by adopting a multi-view WSI, the method provided by the invention adopts a non-overlapping image segmentation and blocking mode, so that the over-fitting of the model can be prevented to a certain extent, and the efficiency of the model is improved.

The invention relates to a sensing signal processing method and system based on a metasurface and a complex valued neural network, and the method comprises the steps: processing a coefficient S11 and a coefficient S21 based on a trained and dynamically optimized complex valued neural network, and outputting the dielectric constant of a sample or the concentration of a solution; the complex valued neural network comprises an input layer, a preprocessing layer, a plurality of hidden layers and an output layer; the S11 coefficient and the S21 coefficient are obtained by microwave signals sequentially passing through the sample and the metasurface; the system comprises a transmitting antenna, a receiving antenna, a vector network analyzer, a metasurface and a sensor signal processing module, and is used for realizing the method. Compared with the prior art, the device has the advantages of capability of realizing measurement of various substances, high measurement precision and the like.

The invention relates to the technical field of intelligent monitoring, in particular to an electronic guide vehicle steering system fault prediction method based on time sequence data analysis, which comprises the following steps: mapping a steering system into perception, decision and execution virtual computing nodes to construct a digital twin model; extracting inter-node time lag correlation through multivariable Granger causality operation, and constructing a virtual causality matrix; fault parameters such as current, angle and bias are injected into the model for numerical simulation, and a response sequence is generated through time domain iteration; using a dynamic time warping algorithm to calculate the morphological difference between the time sequence and the reference sequence so as to generate a time sequence feature; and constructing a digital twinning dynamic graph, inputting the dynamic graph attention network, and extracting spatio-temporal evolution characteristics to generate a fault prediction result. According to the method, accurate deduction of the full-life-cycle fault evolution process of the steering system is realized through digital twinning.

The invention relates to the field of data processing, in particular to a hydraulic valve operation state monitoring method and system, and the method comprises the steps: sorting the flow data of a hydraulic valve collected within a set time according to a time sequence to obtain a data sequence; calculating a noise influence coefficient of each flow in the data sequence; obtaining an adjustment weight of each flow according to the noise influence coefficient; and filtering and denoising the data sequence by using the weight corresponding to each flow to obtain an optimized data sequence, and monitoring the operation state of the hydraulic valve based on the optimized data sequence. The improved filtering algorithm is used for filtering the data sequence, the filtering accuracy and efficiency are improved, the operation state of the hydraulic valve is monitored based on the optimized data sequence, and due to the fact that data are more accurate and reliable, the actual working condition of the hydraulic valve can be reflected more accurately.

The invention discloses a soybean plant pod-related phenotype extraction method, device, medium and equipment, and relates to the technical field of agronomy, and the method comprises the following steps: obtaining front and back images of a soybean plant in a to-be-detected area; constructing an improved twin network model; based on an original twin network model, replacing a feature extraction part with a compressed excitation residual network SE-ResNet50, increasing an attention mechanism, and adding a Dropout layer in a network classification layer; inputting the front and back images of the soybean plant into a target detection model to obtain coordinate information of stem nodes; according to the coordinate information of the stem nodes, the coordinate information of the pods is extracted, the Euclidean distance is adopted to obtain the distance between each pod and each stem node, and the pods and the stem nodes are matched; and after the pods are matched with the stem nodes, inputting the soybean plant image with the front and back images containing the pods into the improved twin network model, performing similarity judgment on the pods in the image, and obtaining the soybean plant pod related phenotypes.

The invention relates to a power distribution network intelligent monitoring method based on a data middle station and edge calculation, and the method comprises the following steps: S1, arranging a sensor in a power distribution network, collecting key data in real time, and carrying out the denoising processing and normalization processing of the collected data; s2, performing preliminary anomaly detection on the processed data at the edge end by using a lightweight GMM model; s3, storing the preprocessed data from the edge nodes in the data center set, and performing secondary detection by adopting a variational auto-encoder based on a preliminary anomaly detection result to obtain a final detection result; s4, constructing a causal relationship graph of the power distribution network, and performing anomaly analysis based on a final detection result; and S5, generating a maintenance report according to the final detection result and the abnormal analysis result. According to the invention, by integrating the edge calculation and the data center, and through multi-stage detection and comprehensive analysis, the abnormity in the network is found and predicted in time, and a reliable guarantee is provided for intelligent management and optimization of the power distribution network.

The invention discloses a hot upflow positioning method based on data driving. For a long time, unmanned aerial vehicles are always indispensable in the industries of monitoring tasks, geographic surveying and mapping, target reconnaissance and the like. Flight endurance is a key limit for wide use of unmanned aerial vehicles. However, although researchers effort to modify and optimize the aerodynamic shape to reduce glider drag or replace battery reactants, use of these methods to increase flight endurance still has limitations. The invention aims to utilize the energy of the nature by collecting wind energy. According to the method, upflow estimation is considered, and then the position of the wind field is estimated through a sparse recognition algorithm of nonlinear dynamics. The thermal positioning is realized through the SINDy-based method and the characteristic of accurately identifying the mathematical physical specification model, so that the capability of identifying the hot upflow is improved, and the energy of the aircraft is increased.

The invention relates to the technical field of medical image analysis, in particular to a training method and system for a cataract recognition model and a storage medium. The method comprises the following steps: acquiring cataract eye image samples and non-cataract eye image samples in different stages; lesion region segmentation and image synchronous simulation enhancement are carried out on the cataract eye image samples and the non-cataract eye image samples in different stages to obtain cataract lesion enhanced image samples and non-cataract enhanced image samples; extracting local detail features of the lesion and extracting structural features of a mask pair, and constructing a multi-task convolutional neural network learning framework to train a cataract recognition model and optimize focus weighted loss; and obtaining a cataract related key feature set and carrying out online incremental learning updating to generate a cataract recognition online incremental updating model. According to the invention, the accuracy and robustness of cataract recognition model training can be improved.

The invention relates to a pet health discrimination method and device based on a gradient lifting xgboost model. The method comprises the following steps: S1, acquiring a historical detection data record of an intelligent pet cat litter basin; s2, performing data set division and xgboost model verification and training on the historical detection data by adopting a data set division method of ten-fold cross validation, and optimizing the xgboost model in each cross validation; s3, performing identification model training based on an xgboost algorithm according to the historical detection data to obtain a cat health condition discrimination model corresponding to the intelligent pet cat litter basin and health index importance corresponding to the pet cat litter basin; s4, monitoring the performance of the xgboost model, and performing incremental updating of the pet health discrimination model and continuous optimization of feature selection according to historical detection data and new detection data; and S5, smoothing the pet health index to obtain a warning threshold value of the intelligent pet cat litter box, and when the real-time detection data of the intelligent pet cat litter box exceeds the warning threshold value, performing pet health abnormity warning on a pet owner.

The invention relates to the technical field of data feature mining, in particular to a remote nursing management method based on data feature mining. The method comprises the following steps: acquiring specific conditions and demand data of a patient; performing data source selection and configuration according to the specific condition and demand data of the patient to obtain a data source configuration table; performing multi-dimensional data acquisition according to the data source configuration table, and performing multi-dimensional data preprocessing to obtain patient comprehensive data; carrying out individualized baseline modeling on the patient comprehensive data to obtain an individual baseline model; performing real-time data standardization on the patient comprehensive data to obtain standardized real-time data; performing deviation calculation according to the standardized real-time data and the individual baseline model to obtain a deviation value sequence; and performing index correlation analysis on the deviation value sequence to obtain an index correlation matrix. According to the invention, the utilization of data feature mining is improved, so that remote nursing management is more accurate and personalized, and finally the nursing quality and efficiency are improved.

The invention relates to the technical field of intelligent traffic, and discloses a deep learning-based expressway traffic flow detection method and system, and the method comprises the steps: obtaining multi-modal traffic data and environment data in real time from each key road section of an expressway, and carrying out the preprocessing of the multi-modal traffic data obtained in real time, including the time-space alignment of the multi-modal traffic data; inputting the preprocessed multi-modal traffic data and environment data into a pre-trained deep fusion network model, and outputting various vehicle parameters for representing expressway traffic flow characteristics; inputting the various vehicle parameters into a pre-trained high-speed traffic flow detection model, and outputting traffic flow grade information; dynamically updating the deep fusion network model at the key road section by adopting a model fusion mode according to the vehicle parameter change condition at each key road section within the preset time; and updating the high-speed traffic flow detection model according to external conditions introduced by the traffic flow grade information in actual use. And the accuracy and real-time performance of high-speed traffic flow detection are improved by combining deep learning.

The invention discloses a single-face array low-altitude radar target recognition system based on deep learning, and the system comprises a data collection module which obtains the radar echo data of different targets in various environments, a receiver has a specific sampling frequency and a dynamic range, the collection time is set according to the characteristics of the targets, and the data is compressed and stored; the data preprocessing module is used for denoising, clutter suppression and normalization; the feature extraction module extracts features by using CNN; the target recognition module classifies and recognizes a target; the model training and optimizing module is used for training a model and optimizing parameters; and the visual display module displays the result. According to the invention, data can be accurately collected and efficiently preprocessed, target features are accurately extracted and recognized by using deep learning, the model is optimized to improve the accuracy, the result can be visually displayed, the accuracy, real-time performance and reliability of low-altitude radar target recognition are effectively improved, and the monitoring requirements are met.

The invention provides an intelligent prediction method and system for hydroelectric energy consumption in a target area, and the method comprises the steps: S1, obtaining multi-modal data which comprises historical hydroelectric energy consumption data, weather data and passenger flow data; carrying out the preprocessing of the obtained multi-modal data, and obtaining the preprocessed multi-modal data; s2, performing feature extraction on the preprocessed multi-modal data by adopting a multi-layer perceptron (MLP), and splicing the extracted features to form a unified large feature vector; s3, an improved Transform model is constructed; s4, training optimization is carried out on the constructed improved Transform model, and the improved Transform model after training optimization is obtained; and S5, performing intelligent prediction on the water and electricity energy consumption of the expressway service area by using the improved Transform model after training optimization. The improved Transform model is used for realizing time sequence prediction based on a multi-head attention mechanism and position coding optimization.

The invention relates to the technical field of remote sensing target retrieval, in particular to a remote sensing target retrieval method, device and equipment based on generalized likelihood ratio metric learning and a computer storage medium. According to the remote sensing target retrieval method, hypothesis distribution is introduced in the training process of a remote sensing target retrieval model, a generalized likelihood ratio corresponding to statistical distribution parameter estimation is defined as a metric space, a loss function is constructed based on the metric space, and the remote sensing target retrieval model is trained; statistical information of global data is introduced in the training stage, overfitting of the neural network is avoided, the feature expression ability of the neural network is improved, and the corresponding target can be retrieved in the remote sensing image based on the content of the given target image with high precision.