292 results about "Moving average" patented technology

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 field of intelligent control of electromagnetic appliances, and provides a control method and system for online fault diagnosis of a magnetic latching relay, and the method comprises the steps: responding to an external control instruction, collecting real-time bus voltage and real-time environment temperature data, constructing environment state characteristics through a magnetothermal coupling algorithm, and generating an initial PWM drive duty ratio; pulse modulation voltage is loaded, a current sampling sequence is collected, a current change rate sequence is obtained through moving average filtering and electromechanical coupling decoupling, extreme value searching is executed to extract counter electromotive force distortion characteristics, and a dynamic current sequential sequence is generated; a fusion input tensor is constructed, armature state probability distribution is output through a gating circulation unit, and a real-time energy compensation instruction is generated in a displacement interval before closing; and according to the real-time energy compensation instruction, respectively generating an enhanced driving waveform, a delayed driving waveform or a vibration driving waveform. According to the method, multi-mode physical characteristic perception and depth time sequence deduction are fused, and online fault diagnosis and active intervention closed loop of the magnetic latching relay are realized.

The invention relates to the technical field of distribution network unmanned aerial vehicle autonomous tour-inspection navigation, in particular to a distribution network unmanned aerial vehicle tour-inspection autonomous navigation method based on GPS information and visual information, which comprises the following steps: acquiring a GPS and smoothing projection coordinates, constructing a Thiessen region random index, optimizing a route and acquiring a flight image, evaluating a risk and adjusting exposure parameters, and generating a light environment value. According to the method, an environment model is constructed by fusing GPS information, Gaussian projection is adopted to eliminate earth curvature errors, moving average filtering is adopted to suppress signal noise, Thiessen polygon is utilized to quantify routing inspection point and obstacle distribution, a tornado optimization algorithm is combined to generate an optimal route, the shortest path without repetition and obstacle avoidance are realized, and obstacles are identified based on a YOLOv12 model. A risk avoiding strategy is made according to the risk value, a route is dynamically optimized, a light environment fingerprint database is constructed, illumination change is predicted in combination with LSTM, aperture shutter sensitivity and white balance are adjusted in a self-adaptive mode, image brightness and definition are optimized, and inspection efficiency and data reliability are improved.

The invention provides a gait monitoring system based on multi-modal sensing fusion, which comprises the following steps of: establishing an initial base line to form a personalized gait base line model; baseline updating is adopted, and after a system obtains new gait data, a weighted moving average algorithm is adopted for self-learning updating, so that the historical data weight is enhanced, and the baseline stability is improved. The technical problems of insufficient individual difference adaptation, lack of dynamic threshold adjustment, poor real-time performance and the like in the prior art are solved.

The invention provides a comprehensive noise reduction performance evaluation method for a nonlinear ultrasonic detection signal noise reduction algorithm, and the method comprises the steps: firstly constructing a multi-working-condition noise reduction performance pre-screening mechanism based on error band analysis based on early-stage experimental data, and then introducing a radar map as an evaluation tool for the comprehensive noise reduction effect of the signal noise reduction algorithm. The noise reduction effects of a moving average method (MA), a spectral subtraction method (SS), a short-time Fourier transform method (STFT), a wavelet transform method (WT) and an orthogonal matching pursuit algorithm (OMP) are compared, and finally a signal noise reduction algorithm with the optimal comprehensive efficiency is screened out. On the basis of algorithm optimization, quantitative mapping rules between microcrack three-dimensional geometric parameters and relative nonlinear coefficients are analyzed through regression modeling, and the significant level of the correlation degree of the relative nonlinear coefficients and microcrack size parameters is effectively improved; and a high-confidence theoretical support is provided for quantitative nondestructive detection of the microcracks in engineering practice.

The invention is applicable to the field of exhaust control, and discloses an exhaust control method, system and device and a medium, the method comprises the following steps: acquiring exhaust operation state data to obtain a historical data set; a predicted pressure value is obtained by combining the historical data set with the weighting factor and adopting a weighted moving average prediction model; obtaining a predicted pressure change rate through the predicted pressure value, and comparing the predicted pressure change rate with a preset threshold value to obtain an exhaust triggering condition; on the basis that the exhaust triggering condition is met, exhaust parameters are generated through a digital twin simulation model, and exhaust is executed according to the exhaust parameters; based on the pressure value and the sealing state in the exhaust process, the exhaust state is judged in combination with the recovery threshold value, and an execution instruction is obtained; and exhaust abnormity detection is carried out based on the execution instruction to complete exhaust control, the process safety is guaranteed, whole-process data are recorded and uploaded, an exhaust closed loop is realized, and the accuracy, stability and safety of exhaust control are improved.

The invention relates to the field of defect detection, in particular to a multi-dimension-based defect detection labeling quality automatic evaluation method and system, and the method comprises the following steps: constructing an initial domain knowledge base, initializing a severity weight, a dynamic reliability weight and a multi-dimensional smoothing coefficient, and loading a pre-training defect detection model; inputting a defect sample batch to be evaluated, iteratively training the defect detection model, and calculating a multi-dimensional original evaluation index; obtaining an evaluation dimension set through dimension reduction and standardization processing, and generating a dynamic tracking result of each dimension index by adopting an index moving average algorithm in combination with a smoothing coefficient; in combination with the severity weight and the dynamic reliability weight, a comprehensive mark quality score is obtained through weighted fusion, and suspected error mark samples are screened out; and performing iterative training on the defect detection model, and outputting a final result. According to the invention, continuous optimization of evaluation parameters is realized through a man-machine cooperative feedback closed loop, and the practicability and stability of the technical scheme are further enhanced.

The invention relates to the technical field of intelligent control, and discloses a dexterous hand motion planning and control method based on a visual language motion model, and the method comprises the steps: S1, collecting data, and carrying out the time sequence synchronization and consistency verification of multi-source data; s2, performing dynamic adaptive exponential moving average filtering and calculation processing on the joint sequence, and reconstructing a processing result into adaptive structured features; s3, fusing joint states, inputting the fused joint states into an action head network, splicing online and offline samples into a training batch according to a preset proportion, and determining results to jointly optimize model parameters; s4, performing model reasoning to obtain target action information, performing processing in sequence to generate a smooth low-jitter control sequence, and sending the smooth low-jitter control sequence to the dexterous hand for execution; and S5, the deployment thread issues a control instruction by aligning the annular buffer and the timestamp, triggers a rollback track and allows manual intervention when the control instruction exceeds a threshold value, and writes manual intervention data into an offline buffer. According to the method, efficient, stable and smooth motion control of the dexterous hand can be realized under the condition of extremely few teaching data sets.

The invention belongs to the technical field of industrial automation control, and relates to a seasoning filling PID parameter adaptive optimization control method and system. The method comprises the steps that the real-time temperature of fluid in a filling pipeline, the static pressure of a valve inlet, the instantaneous flow and a valve opening instruction are collected, and moving average filtering preprocessing is carried out; constructing a self-adaptive forgetting factor calculation model based on the viscosity-temperature characteristics, and calculating a current optimal forgetting factor according to the fluid temperature change rate, the flow prediction error and the fluctuation condition of the flow data; a recursive algorithm is used for identifying the system process gain online, and the proportional gain of a PID controller is calculated in real time according to the identified process gain and the ratio of the hydrostatic pressure to the standard reference pressure; and the gain is input into a controller, and a final valve opening instruction is generated in combination with integral and differential terms. According to the method, the dynamic adjustment of the algorithm memory length can be realized, the viscosity change can be quickly adapted, excessive filling and slow response are eliminated, and high-precision filling under complex working conditions is ensured.

The invention discloses an experiment teaching system based on augmented reality and machine vision, which relates to the field of laboratory teaching systems and comprises an augmented reality visualization module, a machine vision recognition module, a task flow control module, an error judgment and prompt feedback module, a teaching data acquisition and evaluation module and a teacher control terminal. The experiment steps and the three-dimensional model of the equipment are overlaid in an actual scene in real time through the augmented reality technology, student operation images are collected through a machine vision algorithm, a state vector is generated, matching degree calculation is conducted on the state vector and a standard vector, misoperation is recognized, and visual prompt is given; the recognition robustness is improved by adopting a moving average mechanism, and the teaching process is dynamically controlled through a task graph; and the scoring module comprehensively evaluates student performance based on indexes such as a misoperation rate, completion time and a prompt response rate, updates a personalized knowledge graph and realizes a teaching feedback closed loop. The visual, normative and intelligent level of experiment teaching is improved, and good practicability and popularization value are achieved.

The invention relates to the technical field of power equipment monitoring and fault diagnosis, and discloses a power transformer fault early warning method and system. The method comprises the following steps: acquiring a load current sequence, an environment temperature sequence and a top oil temperature sequence of a transformer; calculating an initial actually measured temperature rise based on the top oil temperature and the environment temperature, and determining thermal response lag time through cross-correlation analysis; carrying out translation correction on the load current according to the lag time, and constructing a time-aligned excitation input set and an actually measured temperature rise sequence based on a common effective time window; substituting the excitation input set into a thermal circuit model based on a thermal balance differential equation to calculate a theoretical reference temperature rise, calculating a residual error in combination with an actually measured temperature rise sequence, performing moving average filtering, and constructing a fault feature vector; and inputting the fault feature vector into a fault diagnosis classification model to determine a fault type and severity, and generating a grading early warning instruction according to the fault type and severity. According to the method, thermal hysteresis interference can be eliminated, and accurate quantitative early warning of the latent thermal fault of the transformer under the dynamic working condition is realized.

The invention discloses a large language model control fine tuning method and system based on multi-task cooperative regulation and control, and belongs to the technical field of large language models. Generating a gating coefficient and an initial dynamic evaluation signal through the intelligent regulation and control network; a task difficulty index is obtained by combining multi-index fusion and historical moving average, and the sampling probability and the exclusive learning rate are dynamically adjusted; weighting the fusion gradient and carrying out back propagation to update parameters; and closed-loop feedback monitoring is carried out and parameters of the regulation and control network and the scheduling policy device are optimized. The system comprises a data coding module, a collaborative intelligent regulation and control module, a dynamic balance control module, a joint optimization module and a closed-loop feedback module. According to the method, gradient conflicts among tasks are relieved, the problems of convergence instability and performance imbalance are solved, the multi-task training efficiency, convergence stability and generalization ability of a large language model are improved, and the method can be widely applied to multi-class multi-task learning scenes.

The invention provides an encrypted traffic anomaly detection method based on unsupervised learning, and relates to the field of artificial intelligence network security. In order to solve the problems of dependence on a large number of labeled abnormal samples, poor model generalization ability and poor dynamic adaptability of the existing encrypted traffic anomaly detection method, the invention provides an encrypted traffic anomaly detection method based on unsupervised learning, which comprises the following steps: extracting effective features of an encrypted traffic data packet, generating a state sequence by using a KMeans clustering algorithm, and carrying out unsupervised learning on the state sequence; calculating the occurrence probability of the state sequence in combination with an n-order homogeneous Markov chain model; a dynamic adaptive threshold is constructed based on exponential weighted moving average (EWMA) and a sliding window mechanism, and abnormality judgment is realized by comparing the occurrence probability of a state sequence with the dynamic threshold. According to the method, only normal encrypted traffic is utilized for training, and effective detection of all encrypted traffic including abnormal traffic can be realized.

The invention discloses a double-dynamic alternating consensus teacher network semi-supervised medical image segmentation method, and relates to the field of computer vision and medical image processing. The method comprises the following steps: firstly, constructing a system comprising a student network and two differentiated initialized teacher networks, predicting a label-free sample by using the double teacher networks at the same time in a training stage, obtaining a consensus degree value by calculating the pixel-level consistency of a prediction result in a foreground category, and dynamically adjusting the learning rate of the student network according to the consensus degree value, and the training intensity is adaptively controlled. Meanwhile, the activation and freezing states of the two teacher networks are controlled by adopting a periodic alternating strategy, and only the teacher network in the activation state is subjected to index moving average updating at any phase. Through a time dimension parameter decoupling mechanism and a consensus degree-based reliability feedback mechanism, confirmation bias errors in semi-supervised learning are effectively suppressed, and the segmentation precision and generalization ability of the model under the condition of low label data are improved.

The invention discloses a quantitative calculation method for analyzing injection-production dynamic response by using time sequence interference. The quantitative calculation method comprises the following steps: S1, acquiring time sequence data of output and water injection rate; s2, taking the change of the water injection rate as an intervention event, and establishing a virtual interference variable; s3, taking the oil well output as a target variable and substituting the target variable and the virtual interference variable in the S2 into an SARMAX time sequence model for fitting; s4, judging the significance degree of the model by using the residual error of the model, and if the residual error is a random time sequence, ending; if the residual sequence of the model is non-white noise, going to S5; s5, determining autoregression and moving average orders by using an ARMA (p, q) model; s6, substituting order information into the model in the S3 for re-fitting; and S7, repeating the steps S3-S6 until the model is significant. The method can quantitatively evaluate the influence degree of the injection well on the output. Quantifiable data are provided for evaluation of production and injection dynamic influences in oil reservoir dynamic management, reasonable injection and production adjustment countermeasures are formulated accordingly, oil reservoir production dynamic management is guided, and the oilfield exploitation effect is guaranteed.

The invention relates to a multi-agent-based personalized opera action teaching assistance method and system. The method comprises the steps of inputting a student portrait into a planning agent to generate a personalized learning plan; collecting student follow-up training videos in real time, extracting to obtain a key point sequence, inputting the key point sequence into the lightweight action matching model, calculating the similarity between student follow-up training actions and standard actions in real time, and outputting an action similarity score; inputting key point data obtained after the follow-up training is finished into an execution agent to obtain opera action characteristics, performing deviation judgment and reason analysis through a large language model, generating correction suggestions, and realizing teaching assistance; and after a learning cycle is completed, inputting deviation judgment and reason analysis results into the reflection agent, carrying out statistics on a deviation trend and a skill improvement amplitude, outputting a portrait updating suggestion, updating the student portrait through an index moving average algorithm, and triggering a new round of learning planning. Compared with the prior art, the intelligent level and effectiveness of Chinese opera action teaching assistance are remarkably improved.

The invention discloses a periodic settlement data prediction method fused with nonlinear dynamic lag modeling, which comprises the following steps: counting the number of manual operation times in unit time as an intervention frequency, comparing the intervention frequency with a preset threshold value, and judging whether a service scene is a periodic scene or a non-periodic scene; for the scene which is determined to be periodic, calibrating a monthly prediction result based on scale change of historical same-period data; for the scene judged to be aperiodic, parameters are updated in combination with a moving average method, and processing delay is dynamically adjusted by incorporating holidays and festivals and resting arrangement, so that sequence prediction is generated and completed; and dynamically updating model parameters and predicted values. According to the method, manual intervention intensity and sequence period feature analysis are fused, the universality and accuracy of service scene judgment are remarkably improved, the recognition limitation of a traditional method on mixed services is broken through, and lagging behavior prediction in the fields of energy, finance and the like can be adapted through the three stages of triggering, processing and completing.

The invention discloses an AC insulator leakage current simulation and DC extraction method and system, and relates to the technical field of power system insulation detection and signal processing, and the method comprises the steps: building a controlled wetting experiment platform based on a data collection device, a slope method test device and a sensor; collecting a leakage current signal under a constant alternating current voltage, and eliminating high-frequency noise by adopting moving average filtering; identifying arc intervals through time domain threshold segmentation, spectrum energy distribution analysis and dynamic feature matching in a combined manner, and removing the arc intervals; sliding window median filtering is performed on the remaining signal to eliminate baseline drift, pure wetting leakage current is extracted, and a direct current component is extracted through FFT analysis and content is estimated. According to the invention, high-precision simulation, signal separation and direct current extraction of the leakage current are realized, the arc identification accuracy and the direct current detection sensitivity are remarkably improved, and a reliable basis is provided for quantitative diagnosis of wetting, pollution and aging states of the surface of the insulator.

The invention relates to the technical field of wireless communication synchronization and gate valve action collaboration, in particular to a multichannel audio master-slave sound role intelligent conversion system and method.A master control unit broadcasts a time calibration packet to synchronize unit clocks, and an automatic calibration protocol is executed to generate an inherent processing delay baseline value; the subordinate gate valve unit records an instruction receiving and action completion timestamp through a delay detection module, calculates actual delay containing wireless transmission and local processing delay, and packages the actual delay into a structured feedback data packet for real-time feedback; a dynamic adjustment module of the main control unit filters noise through a weighted moving average algorithm, and calculates individual compensation delay parameters in combination with the dynamic buffer compensation amount of the environment interference monitoring module and delay trend prediction of an adaptive learning engine; and when abnormity occurs, historical data is called for replacement and self-inspection is triggered, an instruction distribution time sequence is adjusted according to compensation parameters, a gate valve action execution confirmation module verifies the displacement precision, and the fluid control precision and stability are improved.

The invention relates to the technical field of hydrogen peroxide production, in particular to a hydrogen peroxide production monitoring and early warning method based on big data, and the method comprises the steps: calculating a material imbalance vector of an equipment node based on the law of conservation of mass; adjusting a smoothing coefficient of an exponentially weighted moving average algorithm according to the material imbalance vector variation of the equipment node, and calculating a target smoothing coefficient of the equipment node to obtain a corrected sensor reading of the equipment node at each moment; taking the weighted sum value of the corrected sensor readings of the equipment nodes in the equipment node upstream adjacent node set at one moment as the theoretical sensor readings of the equipment nodes at the moment; the upstream and downstream influence of the equipment node at each moment is determined on the basis of the difference between the theoretical sensor reading and the corrected sensor reading of the equipment node at each moment, so that abnormal monitoring and early warning of the production process are realized, and the accuracy of monitoring and early warning is effectively improved.

The invention discloses a civil engineering structure monitoring data fusion processing and visual analysis method based on a cloud computing architecture. In the invention, through standardized integration and dynamic cleaning of multi-source heterogeneous data, in a data access stage, an edge node carries out real-time preprocessing on vibration strain displacement and other multi-type sensor data, and in combination with a moving average filtering and abnormal value elimination mechanism, the stability of original data is ensured; after entering the cloud, heterogeneous data output by different devices are integrated into a standard column storage structure through unified format conversion and unit calibration, and meanwhile, missing values and sensor drift errors are processed by adopting linear interpolation and drift correction technologies, so that the data quality is remarkably improved. The hierarchical processing mode not only retains the real-time advantage of the edge end, but also realizes deep cleaning of large-scale data by means of cloud computing power, so that a more reliable data basis is obtained for subsequent fusion analysis, and structural state misjudgment caused by data exception is avoided as much as possible.

The invention provides a time series prediction method and system based on period-trend-residual decomposition, and relates to the technical field of time series data processing, and the method comprises the steps: obtaining time series data, inputting the time series data to a cycle period generation module for processing, and explicitly extracting and outputting a periodic component corresponding to a global periodic mode; inputting the time sequence data into an adaptive weighted multi-scale moving average module in parallel for processing so as to extract and output trend components corresponding to a sequence evolution trend under different time granularities; calculating and outputting a residual component representing high-frequency fluctuation and random disturbance through a residual sensing module based on the time sequence data, the periodic component and the trend component; and performing addition reconstruction on the periodic component, the trend component and the residual component to obtain a prediction result in a specific time range in the future. According to the invention, through the explicit decomposition and lightweight module, high prediction precision is ensured, the calculation complexity of the model is significantly reduced, and the lightweight of the model is realized.

The application discloses a kind of well leakage risk early monitoring and early warning method, in the selection of calculating drilling fluid outlet flow moving average time period, time-frequency variation method is determined, so that the period contains the fluctuation characteristics of signal main, more scientific;Based on the comprehensive alarm method formed by moving average method and dispersion integral method, the sensitivity of the algorithm to the outlet discharge signal is ensured Reduce trend monitoring, while eliminating the risk false alarm caused by normal data fluctuations;The minimum value of the previous period dispersion integral is used to automatically generate a threshold, reducing the error caused by manually setting the threshold based on experience, and having good recognition effect in the early stage of well leakage.

This invention discloses a vulnerability type prediction method based on a fusion of self-training and knowledge distillation of teacher-student models, belonging to the fields of software engineering and information security. The method includes: constructing a joint training data source using a small amount of labeled data and a large amount of unlabeled data; supervising and fine-tuning the teacher model using labeled samples and selecting the teacher model with the highest F1 score based on the validation set's F1 score; generating pseudo-labels by performing MC Dropout inference on unlabeled samples using the teacher model, filtering high-quality samples based on confidence thresholds, and performing class-balanced sampling; training the student model on real and pseudo-labels, learning the soft probability distribution and feature representation of the teacher model, using KL divergence, cross-entropy, and MSE as the joint loss, and introducing progressive weights to stabilize convergence; updating the teacher model parameters using an exponential moving average strategy to improve the quality of pseudo-labels; and finally selecting the student model with the highest F1 score to complete vulnerability type identification.

The invention belongs to the technical field of surface defect detection, and particularly relates to a defect detection method and system based on texture enhancement and dynamic pseudo label screening, and the method comprises the steps: obtaining an industrial surface defect image, constructing a data set, and dividing labeled and unlabeled data; a defect detection network is constructed, a texture enhancement module is introduced into backbone shallow layer features, and the reliability of pseudo labels is improved; constructing a teacher and student detector under a teacher-student self-training framework, and updating teacher parameters by index moving average; generating a weak / strong enhanced view for the unlabeled image, and generating candidate pseudo labels for the weak enhanced view by a teacher; determining a category adaptive threshold based on confidence distribution, and dynamically screening pseudo labels according to categories; training students by combining screened pseudo labels with labeled data, and iteratively updating parameters; and outputting a defect category and a positioning result after training is completed. Compared with the prior art, the method has the advantages that pseudo label noise can be inhibited, class imbalance can be relieved, the detection precision is improved under a low label proportion, and the method is suitable for industrial defect detection scenes.

The application relates to the technical field of intelligent identification, and discloses a gesture recognition method based on adaptive uncertainty-aware meta learning, which formalizes a gesture small sample regression framework, extracts features by using a linearized neural network and a neural tangent kernel, establishes a Gaussian process regression model by combining Bayesian inference, projects a covariance matrix in a reduced dimension by using a Fisher information matrix in view of high-dimensional characteristics, simultaneously constructs an adaptive weight generator, generates a dynamic threshold value by means of historical loss moving average, training progress and an uncertainty correction term, and assigns specific weights to tasks; a posterior mean value is output as a predicted angle in a meta test stage, and the posterior covariance is used to quantize uncertainty. The application can effectively cope with object symmetry ambiguity and feature loss, focuses the model on difficult tasks through an adaptive mechanism, significantly improves prediction accuracy and reliability in a small sample scene, and reduces computational complexity.

The invention provides an energy management dynamic planning method for a light storage system under time-of-use electricity price, and the method comprises the steps: building a user behavior feature matrix through collecting and normalizing the start-stop sequence, power change, frequency and other multi-dimensional behavior data of electric equipment, quantifying the uncertainty of a user load through a Bayesian inference and sequential Monte Carlo method, and carrying out the optimization of the user behavior feature matrix. The calculation of the regional level load uncertainty index is realized in combination with a weighted variable coefficient; based on an S-shaped response curve and a moving average mechanism, dynamically adjusting the weights of economical and stable targets, driving multi-target dynamic planning to solve an energy scheduling strategy, and considering both the economical efficiency and the scheduling stability; user behavior modeling and parameter adaptive adjustment are optimized through closed-loop residual analysis feedback, and the accuracy of the model for dealing with load fluctuation and equipment state change and the real-time performance and robustness of scheduling optimization are greatly improved.

The invention discloses a semi-supervised adaptive unstructured scene segmentation method, and relates to the technical field of computer vision. The method comprises the following steps: acquiring label-free image data of an unstructured scene; inputting the weakly enhanced image into a teacher model to generate a first prediction result; inputting the strong enhancement image into the student model to generate a second prediction result; calculating the JS divergence between the first prediction result and the second prediction result, and dynamically adjusting the attenuation coefficient of the index moving average according to the JS divergence; updating teacher model parameters according to the adjusted index moving average attenuation coefficient to obtain an unstructured scene segmentation model; and inputting a to-be-predicted unstructured scene image into the trained unstructured scene segmentation model, and outputting an unstructured scene segmentation result. According to the method, the teacher model weight updating rate is reduced in the model divergence significant stage such as the initial training stage or the difficult sample processing stage so as to suppress noise propagation, and the prediction precision of unstructured scene segmentation is effectively improved.

The invention provides a target detection method and device based on a meta universe, electronic equipment and a storage medium, and the method comprises the steps: obtaining an initial image, carrying out the AI creation of the initial image in the meta universe to obtain an enhanced image with a label, training a preset teacher model according to the enhanced image and the initial image, and obtaining a target detection result; and optimizing the weight of the teacher model through a gradient descent algorithm to obtain a trained target teacher model and a teacher model weight, performing weight updating on a preset student model through an index moving average and the teacher model weight to obtain an initial student model, inputting the initial image to the initial student model, and obtaining the target teacher model. Obtaining a prediction bounding box and prediction category information, calculating distillation loss based on the prediction bounding box and the prediction category information, performing knowledge distillation on the initial student model according to the distillation loss to obtain a target detection model, and performing target detection through the target detection model; the target detection method based on the meta universe is more practical and more accurate.

The present disclosure provides a method, medium, and device for processing meteorological data based on improved moving average filtering. The method includes: collecting and cleaning original meteorological data to obtain to-be-processed meteorological data; performing weighted moving average filtering on the to-be-processed meteorological data to obtain filtered meteorological data; conducting trend analysis and boundary processing on the filtered meteorological data; and reconstructing the filtered meteorological data after being subjected to the trend analysis and boundary processing to ensure the data continuity and integrity. The present disclosure provides higher precision and reliability in the meteorological data processing process.