124 results about "Difference vector" patented technology

The invention discloses a double-seat cooperative safety control method and system for engineering machinery, relates to the technical field of engineering machinery control, and provides the following scheme that the method comprises the steps that a first operation instruction stream of a main driving seat and a second operation instruction stream of a co-driving seat are collected in real time, and a working condition parameter set of the engineering machinery is synchronously obtained; the operation instruction stream comprises all degree-of-freedom motion parameters of the engineering machinery, inputting the first operation instruction stream into a pre-constructed long-short-term memory model, outputting a prediction operation instruction stream in a future time window, and based on a difference vector of the first operation instruction stream and a second operation instruction stream on each degree-of-freedom motion parameter. According to the method, through the index curve fitting model and Simpson integral calculation, the defect that the co-pilot intervention weight is difficult to quantify is overcome, the dynamic intervention weight based on the physical equidistant interval absolute integral value is realized, and the accuracy of the control signal of the engineering machinery is improved.

The invention relates to the technical field of intracranial pressure monitoring, in particular to a craniocerebral injury patient secondary brain injury early warning method based on multi-modal fusion, which comprises the following steps: acquiring intracranial pressure, cerebral blood flow, electroencephalogram signals and brain tissue oxygen partial pressure signals, carrying out denoising and normalization processing, constructing a multi-modal normalized time sequence data set, and carrying out multi-modal fusion on the basis of the multi-modal normalized time sequence data set; and inputting a linear discriminant analysis model to generate a time sequence difference vector, constructing a difference driving map based on a Bayesian dynamic network model, identifying abrupt change nodes, outputting a risk early warning section set, and carrying out trend stability analysis and reconstruction optimization. According to the method, the abnormal state recognition sensitivity is improved through normalized processing of multi-modal signals, enhancement of signal fusion capability, quantitative capture of direction change, reconstruction of spatial relationship between signals, modeling of dynamic transition probability, recognition of abrupt change nodes, combination of time difference vector change, evaluation of signal linkage trend, positioning of risk sections and extraction of parameter features; and the early warning and intervention precision is improved.

The invention relates to the field of social security management data processing, and discloses a social security information intelligent analysis system based on dynamic closed-loop regulation, and the system comprises an analysis prediction unit which calculates a risk score based on a feature weight; the subthreshold residual cutoff unit captures weak score data lower than a trigger threshold value; the space-time entropy aggregation unit calculates the Shannon entropy of the weak data; the agility bias unit reduces the trigger threshold according to the low-entropy aggregation state; the self-adaptive damping correction unit utilizes a predicted and feedback difference vector to reversely correct a feature weight, the environmental background pressure is sensed by calculating a space-time entropy value of discarded residual data, a trigger threshold value is dynamically biased according to the environmental background pressure, a damping mechanism correction model is introduced into a feedback closed loop, and a damping mechanism correction model is introduced into the feedback closed loop. The problem of weak risk report omission caused by a fixed threshold value is solved, and self-adaptive dynamic balance of system sensitivity and false alarm rate is realized.

The invention provides a project cost intelligent auditing method and system based on artificial intelligence, and the method comprises the steps: carrying out the structural analysis and project item extraction of an original document related to the project cost, and forming a standardized project item structural unit set; based on the project item structure unit set, constructing a track chain of each project item in three key stages, namely a budgeting stage, a contract stage and a settlement stage, recording attribute state sequences of the track chain in different stages, and calculating a difference vector to represent the variation amplitude of a key field; performing abnormal behavior identification based on the trajectory chain, identifying a potential abnormal trajectory through a multi-dimensional abnormal scoring function combining numerical value change intensity and clause unit matching missing risk, and generating a structured interpretation prompt unit; and a unified explanation prompt structure is constructed according to the structured explanation prompt unit, and the explanation prompt structure summarizes the serial numbers of the abnormal items, the change fields and the clause missing rate to serve as a finally output structured audit result.

The invention belongs to the technical field of artificial intelligence, and discloses a large model self-iterative optimization method and system based on feature distribution difference feedback, and the method comprises the steps: collecting a human-generated text data set and a model-generated text data set, and carrying out the preprocessing of the data sets; extracting features of multiple dimensions from the data set, and calculating a difference vector; based on the difference vector, using a Softmax function to distribute an optimization weight for each feature dimension to obtain a weight vector; based on the weight vector, using a joint loss function to perform fine tuning on the large model; and performing iterative fine tuning until a preset convergence condition is met. According to the method, the naturalness of the generated text and the human similarity are remarkably improved, automatic and low-cost continuous optimization is realized, a refined and explainable optimization direction is provided, and the method has good universality and expandability. According to the method, an automatic iteration closed loop integrating analysis and optimization is constructed, the problem of a'model cavity 'is solved, and quantifiable alignment of microcosmic language features is achieved.

The invention relates to the technical field of data processing, and discloses a multi-source data difference attribution system and method based on tensor decomposition, and the method comprises the steps: obtaining an original engineering quantity list file, and then carrying out decoupling to generate a standardized list item text sequence and an initial value set, constructing unit prices to form a tetrahedral geometric model set, mapping the tetrahedral geometric model set to a holographic feature vector space, and generating a multi-dimensional list feature vector set; constructing a standard cost reference vector field, identifying a difference vector set and generating a visual cost difference topological data model; a natural language query instruction is responded, a target difference attribution tensor is locked, a logic verification chain is constructed, and an intelligent diagnosis report is output; according to the method, a computable logic distance is established in multi-source heterogeneous data by reversely reconstructing a high-dimensional price-rate feature space, so that the problem of attribution blocking caused by dimension collapse in traditional cost auditing is effectively solved, and accurate traceability and compliance automatic diagnosis of cost differences are realized.

The invention provides an electrocardiograph fault self-diagnosis method and system based on lead wire automatic identification, and the method comprises the steps: triggering a frequency sweep generator according to an equipment starting signal, carrying out the pre-conduction detection of lead wires, then applying an excitation signal in a continuous frequency range to each lead wire, recording the signal collection time length of each frequency point in real time, and carrying out the real-time detection of the signal collection time length of each frequency point; obtaining response voltage and current data to determine a real-time impedance spectrum; performing frequency point alignment processing on the real-time impedance spectrum and a corresponding standard fingerprint, selecting frequency points in a key frequency band of the electrocardiosignal, and comparing amplitudes and phase deviation values of the real-time impedance spectrum and the corresponding standard fingerprint on the key frequency points to obtain a spectrum difference vector; and if the amplitude deviation of the spectrum difference vector exceeds a preset disconnection threshold value, judging that the lead wire has a disconnection fault, performing reverse signal back test on an interface end corresponding to the faulted lead wire, and recording a judgment result in a fault log to obtain a disconnection identification mark.

The invention discloses a synthetic lethal prediction method based on non-generality forgetting and low-rank interaction multi-mode adaptive fusion, and relates to the technical field of bioinformatics, computational biology and drug target screening. The method comprises the following steps: firstly, constructing a multi-view gene embedding expression based on a gene graph structure, a hypergraph structure and a knowledge graph; then constructing a difference vector and an interaction vector for any gene pair so as to characterize the difference and potential complementary relationship between the genes; multi-source information fusion is realized by using a deep learning model containing interactive attention, a forgetting gate and a multi-modal fusion mechanism; on this basis, multi-task optimization is carried out through joint classification loss, modal interaction loss and redundancy suppression loss, and optimal discrimination is realized through a dynamic threshold search strategy; and finally, outputting a sorting result of the candidate synthetic lethal gene pairs according to the prediction score. According to the method, the accuracy and generalization ability of synthetic lethal relationship prediction can be effectively improved, and a high-reliability calculation auxiliary tool is provided for anti-cancer drug target screening and gene therapy strategy design.

The invention discloses an atmospheric ocean multi-source data high-precision assimilation method and device, and relates to the technical field of ocean multi-source data assimilation, and the assimilation method comprises the steps: collecting multi-source observation data of a typhoon region of the South China Sea; preprocessing the data; decomposing the observation field into different scales by using wavelet transform; acquiring a background field from the numerical model, interpolating the background field to a grid consistent with observation, projecting the background field to an observation space, and performing multi-scale decomposition to enable the background field to correspond to an observation scale; calculating a difference vector of observation and background fields on each scale, and dynamically calculating an adaptive weight of each point on each scale according to an observation error, a background error and a current deviation; and constructing a cost function, solving through an optimization algorithm to obtain an analysis field, and outputting a final analysis field. According to the method, a multi-scale decomposition and self-adaptive regularization cooperation mechanism is introduced, so that high-precision dynamic fusion of the multi-source observation data and the numerical model can be realized.

The invention provides an engineering robot positioning method and system fusing a visual system and a laser system, and the method comprises the following steps: collecting a preset reflection array label image through the visual system, and rapidly solving the initial six-degree-of-freedom posture of an engineering robot; based on the initial attitude, a laser system is guided to actively detect beacons on a label, attitude angle drift is accurately corrected by calculating spatial difference vectors between the beacons, and the micro-Doppler effect analysis of reflected laser is utilized to determine the micro-vibration state of the tail end of the robot in real time. And calculating respective data confidence of vision and laser according to image quality and laser signal stability, fusing the initial attitude and the drift correction through a nonlinear optimization algorithm, and outputting a final high-precision six-degree-of-freedom attitude. According to the invention, high-precision robot positioning can be realized by fusing global perception of vision and local accurate measurement of laser.

The invention discloses a sight tracking method and system based on cascade deep learning and distance correction, and a storage medium. The method comprises the following steps: processing an eye image through a cascade deep learning model, and obtaining pupil center coordinates and cornea reflection light spot coordinates of two eyes; calculating a binocular pixel distance as a distance correction factor based on the pupil center coordinates, and calculating a relative position difference vector based on the pupil and light spot coordinates; establishing a three-dimensional nonlinear polynomial regression model taking the relative position difference vector and the distance correction factor as input; obtaining calibration data through a multi-distance sampling strategy and solving a model coefficient; and in the real-time tracking stage, calculating the coordinates of the fixation point of the screen by using the solved model. According to the method, the distance correction factor and the three-dimensional mapping model are introduced, so that geometric errors caused by forward and backward movement of the head of the user are effectively eliminated, meanwhile, the feature point positioning precision is ensured through cascaded deep learning detection, and the precision and robustness of a sight line tracking system are remarkably improved.

The invention relates to the technical field of converter transformer state monitoring, and discloses a converter transformer internal defect detection method based on multi-physics field coupling monitoring, and the method comprises the following steps: collecting and confirming electric field, magnetic field, temperature field, stress field, sound field and flow field data of a converter transformer without internal defects under a normal working condition; taking the data as multi-physics field reference data; collecting real-time multi-physical field data of the converter transformer to be detected; performing feature extraction on the reference data and the real-time data to obtain reference features and real-time features; calculating a feature difference vector between the real-time feature and the reference feature; and comparing the feature difference vector with a preset corresponding relation library of multi-physical field signal features and internal defect types, and determining the internal defect types. Through coupling analysis and difference comparison of multi-physical field data, cross validation of multi-source information is realized, and the accuracy and reliability of defect diagnosis are improved.

The invention belongs to the technical field of large language models, and particularly discloses a large language model illusion detection method and system, a terminal and a medium. Comprising the steps of obtaining a middle hidden state of each token in a process of generating an answer by a large language model, and organizing and forming a hidden state sequence according to a generation sequence; performing time sequence preprocessing on the hidden state sequence to complete the preprocessing of the sequence; inputting the pre-processed hidden state sequence and a dynamic feature formed by a difference vector, a trend drift degree or local fluctuation intensity into a dynamic sequence modeling network, and enabling the network to extract a time dependence feature reflecting an internal state change mode based on an evolution relation of a hidden state along with a generation process; and generating an illusion risk score according to the comprehensive time sequence characteristics output by the dynamic sequence modeling network, and determining that the answer contains the illusion content when the score exceeds a preset threshold. The reliability of the output content of the large language model can be improved.

The present application relates to a digital twin-based water turbine unit operation system and method, comprising a water turbine unit top cover, a plurality of soundprint collection arrays are arranged on the water turbine unit top cover, each soundprint collection array comprises a plurality of soundprint collectors arranged in a vertical direction and having a fixed interval. The soundprint collector is used to collect original soundprint data at different positions on the water turbine unit top cover, and send the collected original soundprint data to a data center. After receiving the original soundprint data, the data center pre-processes the original soundprint data, establishes a digital twin benchmark model based on the mapping relationship between real-time working condition parameters and soundprint signal characteristics through digital twin technology, constructs an expected feature vector based on a soundprint collector position compensation vector and a mapping matrix, and adjusts the difference vector using a correction factor. The system uses multiple orientations and multiple soundprint collectors configured in each orientation, and the soundprint collection accuracy is high under extreme water flow conditions.

This invention discloses an artificial intelligence-based method for managing and controlling power grid water dispatch data, specifically relating to the field of power grid water dispatch data management technology. The method involves collecting multi-source heterogeneous water dispatch data and its metadata to construct a unified format dataset; generating a credibility score based on metadata and data features to form a credibility matrix; constructing a multi-dimensional quality assessment tensor and extracting the dominant factor subspace; inputting this tensor into a clustering perceptual neural network to identify abnormal data, outputting an abnormal dataset and abnormal labels; combining a large model enhanced by RAG retrieval with historical similar data to complete abnormal data repair; constructing a data quality improvement report using the difference vectors before and after repair and label information; and finally pushing the report to a data management platform and generating a traceable rectification work order. This method achieves automatic identification, intelligent repair, and feedback tracking of power grid water dispatch data, improving data credibility, governance efficiency, and security, and is suitable for data management needs in complex scenarios under new power systems.

This invention discloses a method, device, and medium for correcting human motion posture based on AI vision, relating to the field of AI vision technology. The method includes: acquiring multi-channel image sequences of human posture using multiple cameras; performing temporal alignment and normalization on the multi-channel image sequences; obtaining a human probability map using a pre-trained human semantic segmentation network; performing morphological cleanup to obtain a human silhouette sequence; performing back-projection according to camera intrinsic and extrinsic parameters to obtain a 3D joint sequence from a 3D AI visual volume; extracting the human motion manifold to perform contact-sensing neural motion priors and outputting prior signals; and using posture feasibility scores and joint importance weights to weightedly fuse the geometric difference vector and the center of gravity correction component to obtain a comprehensive joint correction vector. This invention generates a 3D AI visual volume through back-projection and extracts the central skeleton, using constant bone segment lengths for constraint and correction, thus enhancing the reliability of posture stability analysis and alignment with standard action templates.

The invention discloses a vector calculation-based water meter water quantity and water rate calculation method, and aims to solve the technical problems of I / O bottleneck, high calculation delay and poor expansibility in existing large-scale water meter charging. The method comprises the following steps: constructing a water meter data column vector, calculating an original difference vector based on a current-period reading vector and a previous-period reading vector, and performing Boolean mask processing to obtain a meter reading water volume vector; actual water consumption accounting, total and score table allocation and allocated water amount calculation are achieved through vector operation; a billing water volume vector is obtained in combination with a preferential strategy, and an actual water fee is finally obtained through mixed water splitting, stepped billing matrix operation and water fee preferential adjustment. According to the method, vectorization and matrix operation are adopted to replace traditional line-by-line calculation, so that the operation efficiency can be improved through parallel calculation, the vector algorithm is designed to adapt to complex water affair billing logic, the processing performance is effectively improved, efficient and large-scale billing is achieved, the intelligent water affair development requirement is met, and practicability and popularization value are achieved.

The application discloses a shield tunnel excavation disturbance data intelligent identification method based on dynamic time sequence prediction, which comprises the following steps: in the case of excavation disturbance, collecting various time sequence data of shield tunnel monitoring in real time, carrying out denoising, interpolation and normalization processing on original data, and inputting time window samples into a DFCM algorithm; calculating a membership matrix based on the distance between samples and cluster centers, and updating the membership in real time, and constructing an LSTM model of the membership vector output by the DFCM clustering algorithm; predicting the membership of the next time step, outputting a predicted difference vector, and determining a difference dynamic threshold value by using an adaptive threshold method, and when the predicted difference exceeds the threshold value, it is judged that the data is disturbed. The method can realize real-time identification and uncertainty quantification of disturbed data in a tunnel structure monitoring environment with strong interference and high noise, improve the accuracy and stability of underground structure health monitoring, and is suitable for urban shield tunnel operation monitoring and maintenance.

This invention discloses an efficient offline state data caching method for water edge smart gateways based on multimodal data, belonging to the field of water edge smart gateway technology. The method includes the following steps: When a sensor temporary restart is detected, a modal acquisition reference matrix is ​​constructed using the multimodal data structure of historical sampling periods. A modal difference vector is generated by combining this matrix with the actual acquisition content of the current period to determine the modal data gap state under the temporary sensor restart condition. Based on the determined modal data gap state, the image data acquisition record at the current time point is matched and analyzed with the modal data gap vector at the current time point to identify modal inconsistencies between the image and the sensor cached at the same time point. This invention solves the problem of the inability to identify and mark modal gaps in multimodal data under offline conditions, achieving modal consistency identification and structural integrity assurance.

The invention discloses a rapid whole-network coordinate resolving method for a large-scale GNSS (Global Navigation Satellite System) reference station network, which comprises the following steps of: dividing a GNSS observation station into a plurality of sub-networks, and applying loose constraint to each sub-network to resolve and generate a clock error and a track file; independently carrying out PPP floating point calculation on each subnet by using the clock correction and the track file; combining the PPP floating point solution results of all the subnets, calculating the PPP solution of the whole net, and obtaining the coordinates and covariance matrixes of all the stations under the original framework; an independent baseline set is constructed, a relative coordinate difference vector is generated, and the independent baseline set converts the station coordinate vector into a baseline difference vector through a coefficient matrix; and introducing a known coordinate of the reference station under the new reference frame, and reversely deducing coordinates of other stations of the whole network based on the relative coordinate difference vector and the known coordinate of the reference station to complete reference frame conversion. According to the method, the independence and robustness of the conversion process are enhanced, and the coordinate precision and reliability of a large-scale network under a unified reference framework can be ensured.

The present application relates to the technical field of aviation medical monitoring and flight protection, in particular to a perianal health risk assessment and dynamic intervention management system for military flight personnel, comprising: a data sensing step: synchronously collecting environmental stress data and real-time physiological sensing data; a benchmark reconstruction step: calculating ideal physiological response data based on environmental stress; a pathological simulation step: injecting pathological flow parameters to generate pathological simulation data; a difference extraction step: constructing a double-track difference vector, and respectively calculating a real residual vector and a theoretical residual vector; a coupling judgment step: calculating vector similarity to determine the risk type; and a dynamic intervention step: generating a physical intervention strategy to adjust the equipment state in response to the risk type; the present application effectively distinguishes physiological congestion from pathological disorders by dynamically stripping environmental noise, thereby reducing the false positive rate in a high dynamic environment.

The invention relates to the technical field of converters, in particular to a power dispatching system based on a multi-level direct current chopping technology, which comprises a flux linkage trend analysis module, a duty ratio constraint module, a standby unit evaluation module and a path synchronization coordination module. According to the method, the power integral curve is constructed in real time, the flux linkage increment trend is analyzed, the relevance between the flux linkage direction and the current change is dynamically judged, the duty ratio amplitude limiting range is adjusted in combination with the exponential smoothing prediction model, and phase compensation correction is executed, so that the dynamic response capability of the system is remarkably improved; a sliding standard deviation algorithm is adopted to screen a standby unit group and analyze a current fluctuation coefficient, linear regression is combined to process a temperature change trend, the anti-interference capability of the system is enhanced, a dynamic time warping algorithm is adopted to calculate a phase offset and construct a voltage difference vector, accurate synchronous control of a multi-level architecture is realized, and the stability of the system is improved. Harmonic content and energy loss are effectively reduced, electric energy conversion efficiency and equipment operation stability are improved, and the service life of a power device is prolonged.

The present application relates to the field of reagent environment control, and particularly relates to a reagent storage environment control system for multi-classification reagent kits and a reagent kit, the control system comprising an initial-to-bin module, an environment pre-adjustment module, an opening and closing management module, a door closing recovery module, an execution action module and a database, and is matched with the classification information of the reagent kit in the database through a scanning mode, automatically outputs a reagent cabinet number and generates an opening cabinet signal, and constructs an internal and external environment difference vector, combines a digital twin model to simulate a candidate control strategy, and outputs an optimal control parameter vector through a comprehensive cost function to select the optimal output, realizes intelligent pre-adjustment of the internal environment of the cabinet before opening the door, and combines periodical convergence verification and maximum pre-processing time length double constraints to issue an automatic opening door instruction when all the environment parameters are compliant and there is a safety margin, and real-time monitoring of the cabinet door closing state is realized, convergence and fluctuation double determination is completed to realize steady state recovery, a suboptimal event is recorded when the time is up, and an alarm is triggered when the limit is exceeded.

This invention relates to the field of image recognition technology and discloses an image or video recognition method for extracting deterioration features of pre-prepared vegetables. The method includes: acquiring an image and generating a semantic index mask for solid, liquid, and specular reflectance subspaces based on semantic segmentation; constructing a dynamic illumination geometric benchmark using the specular region and calculating the relative color difference vector for the solid state and the relative texture signal-to-noise ratio index for the liquid state; constructing a differential feature manifold space and mapping normalized features to difference vectors; monitoring vector magnitude and direction changes based on topology preservation logic, and generating a deterioration recognition instruction when deviating from the normal topology threshold. This invention achieves relative feature extraction free from environmental illumination interference through a semantic anchor difference and manifold space mapping mechanism, solving the problem of identifying the physical property deterioration of solid and liquid components in pre-prepared vegetables under complex illumination.

The application relates to the technical field of protein engineering and bioinformatics, and specifically provides a protein fitness prediction method and device, a terminal and a storage medium. The method comprises the following steps: extracting wild-type sequence-level representation and mutant sequence-level representation by using a pre-trained protein language model, and calculating a representation difference vector between the two; deducing coevolution coupling information from multiple sequence alignment information, and constructing pairwise evolutionary constraint information reflecting spatial proximity relationship; then modeling the interaction between the representation difference vector and the pairwise evolutionary constraint information through an attention mechanism, so that the pairwise coevolution information guides the propagation and weighting of the representation difference vector between residues to generate enhanced representation; and inputting the enhanced representation into a downstream prediction head to regress a scalar value as a fitness prediction result. The application models the interaction between the sequence-level difference vector and the pairwise evolutionary constraint, so that the non-additive effect between mutations can be accurately captured without relying on any experimental structure.

The invention relates to the technical field of water treatment automation control, in particular to a water pollution detection and recovery treatment system and a control method. The method comprises the steps of data acquisition, model calculation, difference analysis and feedback control. The method comprises the following steps: collecting inflow, matrix concentration and sludge concentration by a system, and resolving ideal effluent matrix concentration by using an unsteady-state kinetic model; the method comprises the following steps: calculating a simulated effluent concentration based on a fault factor, and generating a fault projection matching index by calculating the projection of a real and simulated difference vector; when the index exceeds a preset confidence threshold value, the system generates a control signal based on the superposition correction amount; the dynamic reference zero point is used for automatically filtering unsteady fluctuation interference, hydraulic impact and biochemical reaction faults are accurately distinguished, and the problem of false alarm caused by inflow fluctuation is effectively solved.

The application discloses a horizontal double-runner impulse water turbine generator state monitoring method and system, relates to the technical field of intelligent state monitoring, and comprises the following steps: collecting double-runner multi-modal data of a water turbine generator for preprocessing, generating two groups of standardized data, calculating a difference vector, and constructing a multi-dimensional feature matrix; performing linear transformation based on the multi-dimensional feature matrix, generating a query, key and value matrix, performing near and far field division, obtaining a near field and a far field, calculating near field attention on the basis of the near field, constructing a binary tree on the basis of the far field, calculating far field attention, fusing the near and far field attention, obtaining a final output feature vector, performing state prediction by using a full connection layer, outputting a state label for parameter iterative training and feedback, and obtaining a prediction result; and the application improves the accuracy of state monitoring and identification through effective feature fusion.

The invention relates to the technical field of image recognition, and discloses an image or video recognition method in prefabricated dish degradation feature extraction, which comprises the following steps: acquiring an image and generating semantic index masks of solid, liquid and highlight reflection subspaces based on semantic segmentation; constructing a dynamic illumination geometric reference by using the highlight area, and calculating a solid-state relative color difference vector and a liquid-state relative texture signal-to-noise ratio index; constructing a differential feature manifold space and mapping the normalized feature into a difference vector; vector modulus length and direction changes are monitored based on topological structure retentivity logic, a degradation identification instruction is generated when a normal topological threshold value is deviated, relative feature extraction without environmental illumination interference is achieved through semantic anchor point difference and a manifold space mapping mechanism, and the method is suitable for the relative feature extraction. The identification problem of the physical property deterioration of the solid-liquid components of the prefabricated vegetable under complex illumination is solved.

The invention relates to the technical field of power frequency electromagnetic field intelligent identification, and discloses a power frequency electromagnetic field intelligent identification method and system based on deep learning. The method comprises the following steps: inputting a time-frequency spectrogram into a multi-branch convolutional network to extract a comprehensive feature vector, calculating and identifying uncertainty indexes, and marking a low-confidence sample; calculating a difference vector of the low-confidence-coefficient sample comprehensive feature vector and the high-confidence-coefficient feature center vector, positioning a deviation dimension set, and determining a deviation time-frequency spectrogram; and carrying out local resampling on the deviation time-frequency spectrogram corresponding to the original signal time period to generate a local time-frequency spectrogram, extracting a local deviation correction feature vector, calculating a fusion weight according to a difference vector, carrying out weighted fusion to generate a corrected feature vector, and executing secondary identification. According to the invention, the identification accuracy, stability and real-time response capability of the power frequency electromagnetic field intelligent identification method based on deep learning in a complex electromagnetic environment are improved.

The invention provides a method, device, medium and equipment for monitoring the deposition state of particulate matter in a particulate matter catcher, and the method comprises the steps: converting each ultrasonic signal into a preset dimension map, and carrying out the feature extraction of the preset dimension map, and obtaining a plurality of feature vectors; performing difference processing on the plurality of feature vectors to obtain a feature difference vector; training a particulate matter deposition state recognition neural network by using the feature difference vector, and outputting a particulate matter deposition state recognition neural network model; determining a current preset dimension map of the to-be-detected particulate matter trap, identifying the current preset dimension map by using the particulate matter deposition state identification neural network model, and outputting the particulate matter deposition state type of the to-be-detected particulate matter trap; thus, the preset dimension map can accurately quantify the particulate matter, the model obtained by training the particulate matter deposition state recognition neural network can accurately complete the classification and recognition task, and the recognition accuracy of the particulate matter deposition state is improved.