776 results about "Moving average" patented technology

The invention discloses a self-adaptive multi-dimensional adjustment metering box based on the Internet of Things, and relates to the technical field of intelligent monitoring of power equipment. The problem of measurement distortion caused by high fixed threshold false triggering rate, weak anti-aliasing capability and multi-source data asynchronization in a high-dynamic industrial scene in the prior art is solved. Multi-parameter acquisition is realized through the sensing acquisition module and the annular buffer area; a reconfigurable FIR filtering module is adopted to dynamically switch a low-pass / band elimination mode to suppress aliasing interference; noise features are extracted by combining spectral kurtosis analysis and concept drift detection, and a dynamic threshold value is generated by using exponentially weighted moving average and a sliding window standard deviation; a multi-channel time sequence is aligned through an IEEE 1588 protocol, and cloud parameter closed-loop optimization is realized based on extended Kalman filtering and a particle swarm algorithm; according to the method, the threshold fault tolerance, the high-frequency transient signal capturing precision and the multi-source heterogeneous data fusion reliability in a high-noise environment are remarkably improved.

A control method for fire-fighting false alarm of an energy storage system comprises the following steps of: calculating a fluctuation period, a derivative and a probability density function by using fast Fourier transform and the like by acquiring data streams such as current, voltage, power and heat energy of the energy storage system, determining a system state and predicting a related trend. And predicting power by using an autoregressive moving average algorithm and a support vector machine, and setting a threshold value according to a probability density function to judge abnormity. For heat energy data, through derivative calculation and signal analysis, states are identified and early warning is carried out in combination with models such as a deep convolutional adversarial neural network and a support vector machine. The normal range of monitoring data is determined by means of the long-short-term memory neural network, the change trend and physical quantity relation of abnormal data are analyzed and predicted through a cubic exponential smoothing method and Gaussian process regression, potential risks are judged through weighted fusion and a self-adaptive Simpson algorithm, fire-fighting false alarms are effectively reduced, fire risks are accurately recognized, and the safety of an energy storage system is guaranteed.

The invention discloses a marine environment short-term prediction method and system based on a conditional adversarial network, and particularly relates to the field of marine environment short-term prediction, and the method comprises the steps: S1, obtaining multi-dimensional original data covering a marine physical environment and a remote sensing field; s2, performing space-time registration and abnormal value elimination on the multi-dimensional original data, and filling missing values by adopting a moving average method to obtain standardized data; s3, fusing the standardized data into coupling features reflecting the ocean dynamic process, and constructing spatial-temporal feature codes; s4, constructing a conditional adversarial network model based on the spatio-temporal feature codes, driving the conditional adversarial network model to learn a real data distribution rule through a loss function, generating prediction features meeting spatio-temporal conditions, and completing training; according to the method, the three-dimensional spatial-temporal feature tensor reflecting the ocean dynamic process is obtained through multi-source data fusion and coupling feature engineering, and the characterization capability of data and the description precision of a physical mechanism are improved.

A multi-domain object detection method includes generating a teacher model and a student model from a pre-trained model, inputting an image with weak augmentation applied to a target image, for which an object is to be detected, to the teacher model, determining whether a pseudo label generated by the teacher model is below a preset threshold, performing negative learning for a class corresponding to the pseudo label when the pseudo label is determined to be below the threshold, inputting an image with strong augmentation applied to the target image to the student model, calculating an unsupervised loss by comparing a first prediction generated by the student model with the pseudo label, updating the teacher model using an exponential moving average (EMA) predetermined in the student model, and detecting an object in an image from another domain using the teacher model.

The invention relates to the technical field of vibration monitoring, in particular to a slurry pump motion monitoring management system based on data analysis, which comprises a multi-source acquisition module, a period judgment module, a trend focusing module and a dynamic management and control module. According to the method, vibration data are collected through a three-axis acceleration sensor, a fundamental frequency amplitude is extracted through FFT, time-frequency characteristics and periodic distribution are fused in combination with an extreme value interval sequence, the data representation dimension is enhanced, the rising slope is processed through moving average, the periodic variation is calculated, random noise is restrained, and trend continuity is enhanced. Monotonicity test is combined with dynamic threshold judgment to improve anomaly recognition sensitivity; a sliding window is used for segmenting multi-cycle data; a sudden change interval is positioned based on a pressure amplification rate second derivative; the limitation of a frequency domain on transient feature recognition is broken through, a pressure amplification rate three-dimensional vector is constructed, cosine similarity is calculated, and a working condition difference degree is quantified to dynamically bind a maintenance instruction; and closed-loop feedback is formed to improve the early warning and decision matching degree.

The invention discloses a temperature collecting, monitoring and alarming method for palm fire fighting, relates to the technical field of temperature monitoring for fire fighting, and is used for solving the problem of inaccurate temperature monitoring and prediction. According to the method, environmental factor data are synchronously collected through temperature monitoring points, the correlation between each factor and the temperature is calculated after denoising, multi-factor characteristics are extracted to generate an output characteristic vector of temperature prediction, the output characteristic vector is constructed into a time sequence matrix to be input into a long-short-term memory network, and temperature trend characteristics are extracted; the method comprises the following steps: predicting a future trend and comparing the future trend with a safety threshold to judge whether early warning is triggered or not, meanwhile, identifying temperature changes of different frequencies by adopting fast Fourier transform, a weighted moving average method and an exponential smoothing method through short-term, medium-term and long-term scale temperature trend analysis, setting a three-level early warning mechanism and dynamically adjusting the threshold, and finally, carrying out early warning. Based on early warning feedback data, a Q-learning algorithm is adopted to optimize a threshold value and an alarm strategy, a self-adaptive learning closed loop is formed, and the early fire early warning capability is improved.

The invention discloses a visual target tracking method and system for an unmanned aerial vehicle, and relates to the technical field of image acquisition and processing.The method comprises the steps that firstly, a bridge area is scanned in a directional mode in multiple different sun incident angle periods (the interval is larger than or equal to 30 degrees), and stain interference is eliminated through natural light multi-angle changes; the geometric stability of the crack edge is quantified by constructing a gradient direction distribution matrix of the time dimension; secondly, dividing multiple groups of scanning data into at least six direction groups, calculating pixel proportion variances group by group, accumulating the pixel proportion variances into a total variance parameter, and effectively distinguishing cracks from surface stains by combining a direction consistency verification mechanism (the total variance lt is judged as cracks by a preset threshold value); and finally, establishing a dynamic feedback closed loop based on a historical variance parameter sample library, adaptively adjusting a threshold value through a moving average value deviation ratio formula, and solving the problem of false detection caused by material aging and environmental fluctuation.

The invention discloses a sparse dynamic graph convolution-based electroencephalogram signal decoding method and system. The method comprises the following steps of: acquiring a multi-channel electroencephalogram signal and preprocessing the multi-channel electroencephalogram signal; performing multi-band filtering on each channel signal, extracting statistical characteristics on each band signal, calculating a covariance matrix of a task electroencephalogram signal, constructing image electroencephalogram data by taking an electroencephalogram channel as an image node, the multi-band spliced statistical characteristics on the channel as a node feature vector, and the covariance matrix between the channels as an adjacent matrix; finally, a dynamic graph convolutional neural network model is constructed, the model constructs a graph convolutional neural network based on an autoregression moving average filter, graph electroencephalogram data is used as input, the category of electroencephalogram signals is used as output, an adjacency matrix is dynamically generated in combination with bilinear mapping, fuzzy label learning and sparse constraint are added to improve the decoding capacity of the model, and the dynamic graph convolutional neural network model is obtained. And the frequency domain response capability and robustness of the model to the graph structure are enhanced.

The invention provides a financial risk intelligent early warning method based on multi-source data fusion, and the method comprises the steps: separating the periodic characteristics of income data from the abnormal transaction influence through a time series decomposition technology if the transaction volume fluctuation ratio in an income stability index exceeds a preset threshold value; a weighted moving average algorithm is adopted to smooth records with low data alignment precision, and a smoothed income fluctuation sequence is generated; according to the node level risk set, modeling is carried out on the periodic characteristics of the dynamic risk contribution degree, the change trend of an income fluctuation sequence in a future time period is predicted through a long and short-term memory network, and a dynamic risk prediction sequence containing abnormal transaction influence is generated; and if the risk value of a certain time point in the dynamic risk prediction sequence exceeds a preset warning threshold value, identifying a high risk point driven by the abnormal transaction influence through an isolated forest algorithm, and generating an early warning signal set containing a timestamp and a regional market difference. The financial risk management capability of an enterprise can be effectively improved.

The invention provides an intelligent road crack detection method, and belongs to the technical field of electrical digital data processing.The intelligent road crack detection method comprises the steps that firstly, a line structured light scanner is adopted to obtain road surface point cloud data, and attitude compensation processing is conducted in combination with an inertial navigation system; then noise is eliminated through bidirectional moving average filtering, and the depth matrix is converted into an image and segmented; gaussian background modeling is used to screen abnormal points, a connected graph is constructed through a minimum spanning tree algorithm, and pruning is carried out to obtain a crack skeleton; using a feature enhancement function to optimize the crack saliency; and finally, inputting a pre-trained multi-scale pavement damage identification network, and outputting a crack identification result and classification. According to the method, three-dimensional point cloud data, geometric feature analysis and a deep learning technology are fused, the technical problem of poor adaptability of a complex pavement environment is solved, and accurate identification of various crack types is realized.

Lowest cost usage scheduling of an energy system during a time interval of interest is achieved by utilizing two components of learning and optimization. First, a number of learning approaches including Linear Least Square Regression (LLSR), Auto Regressive Integrated Moving Average (ARIMA), and Multi-Layer Perceptron Deep Learning (MLPDL) are used to forecast energy production and storage of the energy producing components of a given group of energy components as a function of historical production data and weather data collected from weather and geo reports. Then, an optimization problem is formulated to create the optimal schedule of energy usage during time intervals of interests for the given group of energy components subject to scheduling and equipment constraints. Two alternative solutions, namely Simplex Branch and Bound (SBB) and Interior Point Branch and Bound (IPBB) are presented to solve the optimization problem. Accordingly, integrated iterative methods, programs, and systems are described aiming at minimizing the cost of energy consumption for the given group of energy components within time intervals of interest.

The invention relates to the technical field of distributed data processing, in particular to a hydraulic engineering distributed flood prevention prediction system and method.The system comprises a real-time monitoring module for calculating the mean value and variance of data of a water level sensor every five minutes, and an anomaly detection module for comparing the mean value difference with a threshold value, recognizing a sudden change point and sending the sudden change point to a cloud server; the window adjustment module adjusts the collection frequency according to the abnormal result, the trend analysis module predicts the flood trend by applying moving average and exponential smoothing, and the early warning generation module evaluates the water level and generates an early warning signal. According to the method, the mean value and the variance of the water level monitoring data are calculated in real time and compared with the threshold values, so that the recognition capability of the abrupt change point is improved, the data collection frequency and the window size are adjusted according to the anomaly detection result, the adaptability and efficiency of data collection are enhanced, and the water level trend is predicted through the moving average and exponential smoothing technology; and in combination with a geographic information system, the affected area is accurately evaluated, so that the early warning accuracy and the decision support effectiveness are remarkably improved.

The invention relates to the technical field of data processing, in particular to an energy consumption data real-time monitoring method for a smart park, and the method comprises the steps: collecting power energy consumption data, and carrying out the preliminary prediction based on historical data through employing an ARIMA model; historical data are divided into a plurality of segments through mutation point detection, the cumulative change rate and the variation coefficient of each segment are calculated, and then the change trend indexes and trend types of the segments are determined; in combination with the similarity between the current data and each segment, judging the change trend type of the current data; dynamically adjusting an autoregression coefficient and a moving average coefficient of the ARIMA model according to a trend type and residual information; and finally, predicting by using the adjusted model, and realizing energy consumption abnormity monitoring through the deviation between a predicted value and an actual value. According to the method, the accuracy of energy consumption prediction in a changing complex scene is improved, and the accuracy of energy consumption monitoring is improved.

The invention discloses a computing power scheduling management method and system based on an algorithm. The method comprises the steps that an initial resource consumption data set is obtained, and the data set comprises the processor occupancy rate and the memory usage amount of a plurality of calculation tasks in different calculation stages; according to the initial resource consumption data set, a fixed time window is adopted to segment historical data, and a smooth resource consumption sequence is generated through moving average calculation; for the smooth resource consumption sequence, applying a trend decomposition method to separate long-term trend and periodic fluctuation features, and constructing a resource dynamic feature matrix; and according to the resource dynamic characteristic matrix, constructing a multi-layer long-short-term memory network model, and optimizing parameters through a gradient descent method to generate a resource demand prediction curve and the like. According to the method disclosed by the invention, the resource consumption data of the computing task is deeply analyzed and processed, so that efficient computing power resource allocation and scheduling are realized to cope with the change of the demand of the computing task on the processor and memory resources in different stages.

The invention belongs to the technical field of computer vision, and relates to a lightweight image feature point detection and matching method and related equipment thereof. The method comprises the following steps: performing feature extraction on a target image through a lightweight SuperPoint neural network model to generate feature information; performing super-pixel segmentation processing on the image to generate a super-pixel segmentation result; performing initial matching on the feature information through a lightweight SuperGlue neural network model to generate an initial matching pair; performing outlier filtering on the initial matching pair based on a superpixel segmentation result, and determining an optimized target matching pair; performing geometric consistency optimization on the optimized target matching pair by adopting an adaptive exponential moving average algorithm to generate a final matching result; the lightweight SuperPoint neural network model is generated by carrying out hierarchical progressive pruning processing on an original SuperPoint neural network model; the lightweight SuperGlue neural network model is generated by carrying out attention head pruning processing and graph network layer pruning processing on an original SuperGlue neural network model.

The invention belongs to the technical field of aero-engine state monitoring and intelligent fault diagnosis, and discloses a multi-modal information fusion bearing fault diagnosis method based on self-supervised learning. The method comprises the following steps: firstly, through mask reconstruction self-supervision pre-training, extracting stable feature representation insensitive to mask disturbance from an unlabeled multi-modal signal, and dynamically updating each modal feature reference point by using an index moving average algorithm; in a downstream fault diagnosis task, a multi-modal joint decision model comprising a pre-training encoder, a single-modal classifier and a fusion classifier is constructed, and adaptive weighted fusion of multi-modal decision is realized through contribution degree calculation based on a cooperative game Shapley value in combination with a deviation degree of modal features and a reference point. According to the method, the dependence of the deep neural network on fault labeling data is effectively reduced, the accuracy and robustness of the diagnosis system in a multi-modal signal diagnosis scene are improved through a dynamic fusion mechanism, and the method is suitable for industrial scenes with limited sample label resources.

The invention relates to the technical field of self-supervised learning, in particular to a target detection pre-training method based on Transform. A CL-MAE (Control Learning-Masked Autocoder) self-supervised pre-training method is designed, an original image and an enhanced image are processed by adopting a double-branch architecture, branch parameters of the original image are frozen, enhanced branch parameters are updated by using index moving average, and multi-view contrast learning is introduced, so that the problem of'lazy 'of an encoder is effectively prevented, namely, a reconstruction task is completed by depending on a decoder. The method comprises the following steps of: pre-training a target detection network based on a PVT (Pyramid Vision Transfer), transferring the weight of a Vision Transfer) encoder to the target detection network based on the PVT after the pre-training is completed, and realizing the conversion from self-supervised pre-training to target detection in cooperation with FPN (Feature Pyramid Networks) feature fusion and a special detection head. According to the method, the problem that a traditional target detection model depends on annotation data is solved, and meanwhile, the problem of lazy during self-supervision pre-training of a mask auto-encoder is avoided. Compared with a non-pre-training model, the method achieves better target detection precision and convergence speed.

The invention belongs to the technical field of safety monitoring, and discloses a gas pipe network construction site safety monitoring method and related equipment, and the method comprises the steps: collecting the original inclination angle data of different positions of a construction region in real time through distributed monitoring nodes, and carrying out the filtering processing of the original inclination angle data through a moving average filtering algorithm, the differential settlement value is calculated according to the filtered inclination angle data of the adjacent monitoring nodes, the differential settlement value and the dynamic settlement difference threshold value are compared to determine the settlement risk and give an early warning, real-time monitoring of gas pipe network construction site ground settlement is achieved, and the settlement monitoring accuracy is improved.

The invention discloses a new energy power interval prediction method for mining correlation between seasonal change characteristics and meteorological characteristics. The method comprises the following steps: 1) obtaining original photovoltaic power data; 2) decomposing the original photovoltaic power data by using a moving average algorithm to obtain a trend component and a seasonal component; 3) acquiring meteorological data, and constructing a trend component prediction model and a seasonal component prediction model; 4) inputting the meteorological data and the seasonal component into the seasonal component prediction model to obtain a seasonal prediction result; 5) inputting the trend component into the trend component prediction model to obtain a trend prediction result; and 6) calculating the sum of the seasonal prediction result and the trend prediction result to obtain a final photovoltaic prediction result. According to the invention, weather information is allowed to influence seasonal components of photovoltaic power generation data without influencing trend components. The invention further provides a seasonal component prediction unit with a double-layer hierarchical attention mechanism, and attention to the relation among meteorological features, key time nodes and seasonal components is enhanced.

The invention provides a multi-factor correction and real-time early warning method and system based on bridge deflection, and the method comprises the steps: deploying a high-precision deflection sensor at a key position of a bridge, and collecting the deflection data of the bridge under different load and temperature conditions in real time; establishing an initial finite element model of a vehicle load and a temperature effect on the basis of the collected data, and correcting and optimizing the model through methods of static and dynamic load tests, weighted moving average and the like so as to accurately reflect the actual stress state of the bridge; combining the design standard, the safety margin and the historical monitoring data of the bridge to calculate a reasonable threshold value of the deflection of the bridge; in the real-time monitoring process, when actual deflection data exceed a set threshold value, the system automatically triggers an alarm mechanism, and alarm information is sent to a manager through a preset alarm channel; the functions of intelligent monitoring and early warning of the deflection of the bridge are achieved, and the safety and reliability of the bridge structure are improved.

The invention discloses a horizontal wind field inversion method and device based on the radial velocity of a wind profile radar, and relates to the technical field of wind profile radars, and the method comprises the steps: reading the radial velocity data of a wind profile radar in each direction wave beam, carrying out the rearrangement and splicing arrangement of the radial velocity data of each direction wave beam according to a preset sequence, and carrying out the reconstruction of the radial velocity data of each direction wave beam; forming a beam radial velocity profile; performing data identification, one-dimensional median filtering and one-dimensional moving average filtering on the beam radial speed profile in sequence so as to perform consistency processing on the data of the beam radial speed profile; carrying out time-space continuity processing, and then carrying out time consistency average processing; and obtaining an inverted horizontal wind field according to the processed radial speed data of the beams in all directions based on a space geometrical relationship between the wind field and the radial speed data of the beams in all directions. According to the invention, the problem of low accuracy during horizontal wind field inversion in the prior art is solved.

The invention provides an industrial host control method and system based on artificial intelligence. The method comprises the steps that a white list process is set, and when a process accesses the industrial host, the industrial host monitors whether the process accessing the industrial host is abnormal or not through application layer data before and during running of the process; then, real-time syscale flow data is collected through an eBPF behavior collector of an inner kernel layer of the industrial control host, and the syscale flow data is analyzed through an AI decision engine to judge whether the syscale flow data is abnormal or not; and acquiring a process behavior entropy value, smoothing the behavior entropy value by adopting a moving average method, and controlling triggering of a fusing instruction based on monitoring of an application layer and a kernel layer. Through the method and the corresponding system, the capability of detecting the abnormal process can be improved.

The invention discloses a multi-scale time sequence prediction method and device, and belongs to the technical field of time sequence prediction analysis. According to the method, the fOU process, the LSTM model and the attention mechanism are combined, the long-term trend and the short-term fluctuation of the time sequence can be captured at the same time, and the prediction precision and the interpretability are improved. The method comprises the following steps: firstly, acquiring multivariate multi-scale time sequence data and preprocessing; then decomposing the target time sequence into a low-frequency component and a high-frequency component by adopting a moving average method and a residual error method; carrying out modeling and prediction on the low-frequency components by adopting an fOU-LSTM hybrid model; carrying out modeling and prediction on the high-frequency components by adopting a multi-scale LSTM model and combining an attention mechanism; and finally, fusing the low-frequency result and the high-frequency result by adopting a dynamic weighting method to obtain a final target variable prediction value. The method is suitable for time sequence prediction tasks with long memorability, multi-scale features and nonlinear features in the fields of finance, meteorology, medical treatment and the like.

The invention provides a driving risk early warning method based on long-term and short-term driving style features. The method comprises the following steps: constructing long-term and short-term driving feature vectors according to driving behavior data; based on the standardized long-term and short-term driving feature vectors, a clustering method is adopted to determine long-term style feature vectors and long-term and short-term driving risk score labels; constructing a multilayer driving risk assessment model considering long and short term driving styles, and performing fitting training on the assessment model based on the long and short term driving style sample set to obtain an optimal multilayer driving risk assessment model; determining a long-term driving style category according to a long-term driving feature vector collected in real time, inputting the long-term driving style category and a short-term driving feature vector collected in real time into an optimal multilayer driving risk assessment model, and outputting a moving average trend of a scoring time sequence and a fluctuation trend of a driving behavior in a time window according to a predicted driving risk score output in real time. And formulating a two-stage early warning strategy. According to the invention, highly personalized and adaptive risk early warning can be provided.

The invention discloses a filter intelligent backwashing control system based on a pressure sensor, and relates to the technical field of intelligent equipment, and the filter intelligent backwashing control system comprises the following steps: S1, the system firstly executes pressure difference reference initialization, and determines a pressure difference reference value; s2, constructing a corrected differential pressure offset model, and determining the current differential pressure offset of the system; s3, constructing a blockage risk assessment model, and pre-judging whether the system needs to start backwashing or not; s4, carrying out segmented pulse backwashing control; and S5, constructing a closed-loop judgment model, and evaluating the actual backwashing effect of each stage. According to the method, a high-credibility differential pressure reference is obtained through a reference calculation model; the blocking trend is tracked in real time through the moving average pressure difference offset; constructing a blockage risk coefficient fusing the offset, the offset rate and the duration, and judging the flushing opportunity in multiple dimensions; a three-section flow is adopted to distribute flushing strength and duration according to needs; and after each section of pulse is finished, the backwashing effect is evaluated in a closed-loop manner through the closed-loop judgment model.

The invention relates to a semi-supervised remote sensing image classification method based on hierarchical crossing, and belongs to the technical field of remote sensing image processing and computer vision. The method comprises the following steps: firstly, constructing a representation learning module, adopting a hierarchical cross pseudo-tag generation mechanism, randomly combining different hierarchical features in a main model and an index moving average (EMA) model to construct a cross model, and generating a high-quality pseudo-tag for a weakly enhanced image; secondly, a self-adaptive weight mechanism is introduced, the label loss-free contribution is dynamically adjusted in combination with the training progress and the pseudo label utilization rate, and the self-adaptability of the model in different training stages is enhanced; and finally, designing a label alignment strategy based on a training stage, adjusting pseudo label category distribution through a time decline function, guiding the model to pay more attention to minority categories at the initial stage, and improving small sample category performance. Through hierarchical cross random combination, an adaptive weighting mechanism and a label alignment module, pseudo label quality, training stability and minority class performance are improved.

The invention discloses a multi-user short-term power load prediction method for an improved Transform model, and relates to the technical field of novel power systems, and the method comprises the steps: S1, setting an internal decomposition block based on a moving average and Fourier transform thought, integrating the internal decomposition block into a frame, and decomposing an input data sequence into trend, season, holiday and festival variables and residual variables; s2, an Nystrom method is used for approximating a softmax matrix in a standard self-attention mechanism, and the softmax matrix is used for coping with challenges of secondary time complexity and memory use in Transform; s3, performing deep feature mining on the seasonal part after time sequence decomposition by using convolution operation; and S4, probability prediction based on Monte Carlo random inactivation. And designing an internal decomposition module of the depth model, and extracting an internal complex time sequence trend of a hidden state in the model by the module, so that the model has a progressive decomposition capability of a complex time sequence. Different from a traditional load prediction method for exploring time characteristics of a single user, a space-time attention mechanism is improved, and accurate multi-user short-term load prediction is provided.

The invention discloses a cross-view image geographic positioning method based on dynamic threshold pseudo tag self-training learning, and the method specifically comprises the following steps: introducing a difficult sample feature mining method, dynamically adjusting the loss weight of a sample according to the change of similarity, and building a dynamic difficult sample triple loss model; the method comprises the following steps: dynamically adjusting a confidence threshold value of a sample by adopting an index moving average weighting method, iteratively training and screening an unlabeled sample, namely a pseudo label, establishing a pseudo label self-training mechanism of a dynamic threshold value, mining and utilizing non-paired data, and solving the problem of high manual labeling cost; a reference image most similar to a query image is found through image retrieval, and the offset of a query position is predicted. Experiments on CVUSA and CVACT data sets show that as the distance threshold increases, the accuracy of the cross-view image geographic positioning method based on dynamic threshold pseudo tag self-training learning presents a stable rising trend, and the cross-view image geographic positioning method based on dynamic threshold pseudo tag self-training learning is superior to other methods under the same threshold condition.

The invention provides a bedding structure rock finite element model reconstruction method based on a rock core digital image, and belongs to the technical field of digital rock cores, and the method comprises the following steps: S1, obtaining a gray level image of a rock sample; s2, calculating a transverse mean value of the analysis area to obtain a longitudinal gray level distribution curve; s3, applying moving average filtering and smooth filtering to obtain a smooth gray curve; s4, first-order difference is carried out on the smooth gray level curve, the position with the absolute difference value larger than a threshold value is taken as an initial stripe boundary candidate, and starting and stopping pixels, the width and the average gray level of each stripe are determined; s5, mapping each clustering label interval into a physical interval based on FEM grid partition; and S6, constructing a sample finite element model, and generating a reconstructed finite element model with real partition information. According to the method, the macrostructure features of the rock can be efficiently recognized and extracted by directly utilizing the pictures and combining gray profile analysis, and efficient and low-cost reconstruction of the finite element model of the bedding structure rock is achieved.

A computing device may perform training of a set of machine learning models on a first data set associated with a first domain. In some examples, the training may include, for each machine learning model of the set of machine learning models, inputting, as values for a set of parameters of the respective sets of parameters and for an iteration of a set of iterations, a moving average of the set of parameters calculated over a threshold number of previous iterations. The computing device may select a set of model states that are generated during the training of the plurality of machine learning models based on a validation performance of the set of model states performed during the training. The computing device may then generate an ensembled machine learning model by aggregating the set of machine learning models corresponding to the set of selected model states.