34 results about "Normalized mutual information" patented technology

The invention is suitable for the technical field of smart power grids, and relates to a low-voltage transformer area undisturbed topology identification method and system, a medium and a terminal, and the method comprises the steps: S10, cleaning and screening original communication and measurement data; s20, the RSSI between the devices is converted into the distance between the nodes, and an undirected weighted graph is formed; s30, determining an affiliation relationship between the meter box and the electric meter by using a clustering algorithm; s40, performing constraint matching on the transformer, the line and the meter box node according to the power utilization information, and optimizing the topological structure by using the law of conservation of energy and the Kirchhoff's current law; and S50, evaluating the accuracy of the identification result by using the adjusted Lam index and the normalized mutual information. The method is simple in process and convenient to operate, a complete topological structure is constructed step by step on the premise of not adding extra hardware, uninterruptible power and not injecting characteristic signals by relying on electric power data of an existing intelligent electric meter, a communication signal and an electricity utilization information acquisition system, and high-precision and automatic identification of the topological structure of the transformer area is achieved.

The invention relates to the technical field of electroencephalogram signals, in particular to a method, a device and equipment for removing motion artifacts of electroencephalogram signals, and the method can obtain to-be-processed electroencephalogram signals, decompose the electroencephalogram signals into a plurality of mutually orthogonal intrinsic mode components by utilizing multivariable empirical mode decomposition, and obtain motion artifacts of the electroencephalogram signals. The real electroencephalogram signal and the artifact signal are decomposed into different components; standard mutual information between each intrinsic mode component and a preset class label is calculated; multiplying the standardized mutual information by a corresponding intrinsic mode component to obtain an optimized signal component, so that a component related to a real signal is enhanced, and an artifact signal is weakened; all the optimized signal components are accumulated, and the electroencephalogram signals with the motion artifacts removed are obtained. It can be understood that according to the technical scheme shown by the invention, through multi-level and multi-dimensional signal decomposition and artifact removal processing, the motion artifacts can be accurately recognized and removed, and the quality of the electroencephalogram signals is greatly improved.

The invention discloses a multi-temporal optical remote sensing image robust registration method based on multi-scale joint similarity measurement, and the method comprises the steps: 1, inputting a dual-temporal image, and obtaining uniformly distributed control point pairs through SIFT extraction and interactive manual auditing; 2, constructing a global dense displacement field by using a thin-plate spline, completing coarse registration and partitioning according to an overlapping strategy; 3, performing three-stage coarse-to-fine template matching on each image block, inheriting an initial displacement value step by step, and realizing radiation difference and geometric deformation synchronous self-adaption by taking a product of normalized mutual information and a normalized cross correlation coefficient as joint similarity measurement; and 4, performing sub-pixel-level deformation correction on the reliable displacement field by using a local thin plate spline, and outputting a registration image and metadata. According to the method, a global sparse-local dense framework and a multi-scale progressive + joint measurement + quality constraint mechanism are coupled, and a high-robustness and high-precision registration solution is provided for quantitative remote sensing application such as change detection and disaster assessment.

The present invention provides a method and system for monitoring wear state of milling tools for complex thin-walled components, comprising: taking monotonicity of feature vector as a first index, normalized mutual information of the feature vector and wear vector of the tool as a second index, and a ReLU function of Spearman correlation coefficients between feature vectors and the wear vector of the tool as a third index; according to the above indexes, determining behavior indexes corresponding to each feature vector; characterizing behavior characterizations of signal channels, comparing channel behavior indexes of each signal channel, determining input vectors of model, and training tool state recognition model according to the input vectors; and tool change is determined and performed timely by comparing output wear amount value of the tool by using the model with a threshold value.

The application discloses an automatic measurement and control intelligent box transformer control method and system, relates to the technical field of smart grid control, and comprises the following steps: collecting multi-source data by using intelligent sensors, simulating a group of virtual neurons for each intelligent sensor, calculating activation values, summing the activation values, generating aggregated features, calculating the normalized mutual information between the aggregated features, calculating the convolution feature vector by using Chebyshev polynomials, and generating a dimension reduction feature matrix; calculating the speed of a window by using a particle interaction model, calculating the abnormal score of the window, calculating the KL divergence of adjacent windows by using a KL divergence formula, defining a target function by using a weighted sum, and generating a control sequence. The simulation of intelligent sensors and virtual neurons, the Chebyshev polynomial method and the dimension reduction feature matrix generation technology are introduced, the precision and efficiency of feature extraction are improved, the particle interaction model and the KL divergence analysis are adopted, and the accuracy of the control sequence is improved.

The application provides an intent recognition model training method and device based on pseudo labels, and an intent recognition method and device, which comprises the following steps: inputting sample text into a language model to extract a feature vector; clustering the sample text based on the feature vector, taking the clustering result as a pseudo label, and calculating the normalized mutual information between the real label of the labeled sample text and the pseudo label; determining the confidence score corresponding to each sample; the confidence score is used to quantify the noise in the pseudo label, filter high-confidence samples, and take the corresponding pseudo label as a self-supervised signal to iteratively optimize the language model until convergence; after iteration, reinitialize the clustering, update the clustering result, the normalized mutual information, and the iteration number; when the iteration number reaches an upper limit or the normalized mutual information increment is less than a threshold, terminate the training and determine the language model as an intent recognition model; the problem that noise pseudo labels continuously spread and accumulate, leading to a decline in the ability of the model to recognize new intents, can be solved; and the ability of the model to recognize new intents is improved.

The invention provides an abnormal behavior detection method and system based on flight mission context awareness, and is applied to the technical field of data processing. The method comprises the following steps: processing flight dynamic data and task protocol data of an unmanned aerial vehicle, obtaining a task type by analyzing a MASSIONITEM field in a task protocol, and generating a dynamic tag; combining the GPS positioning data with normalized mutual information and random forest importance score to screen task sensitive features, and generating a task related feature subset; processing the task related feature subsets based on a random forest model, training an exclusive classification model for each task type, and generating an exclusive classification model of each type of tasks; and predicting the real-time flight state data of the unmanned aerial vehicle based on the exclusive classification model of each task to obtain probability distribution of each task type, performing judgment according to a set threshold, and generating an unmanned aerial vehicle abnormal behavior judgment result.

The application discloses a UAV multi-modal image registration method and device, medium and product, relates to the technical field of remote sensing image processing and computer vision, and comprises the following steps: calculating the normalized mutual information value of each wave band of the image to be registered and a reference image, selecting a wave band with the maximum value as a reference wave band, solving a global affine transformation matrix by using an enhanced correlation coefficient algorithm, performing global geometric coarse correction, and obtaining the image to be registered after coarse registration; calculating the local principal direction of the image to be registered after coarse registration and the reference image, constructing a Gaussian derivative filter, extracting structural features based on the turning theorem, generating respective structural feature maps and dividing the structural feature maps into regular grids, performing grid-by-grid template matching to obtain a sparse displacement field, performing outlier elimination and interpolation smoothing to generate a dense linear deformation field, performing nonlinear geometric correction and applying the dense linear deformation field to all wave bands of the image to be registered after coarse registration, and obtaining a final registration result. The application realizes high-precision registration of multi-modal images.

The application discloses a network security event report generation method and device based on a dynamic resolution community discovery algorithm, equipment and a medium, and relates to the field of network security. The method comprises the following steps: extracting target information in original network security alarm data as nodes, and constructing a time sequence relationship graph based on the nodes and the directed edges between the nodes; enumerating a plurality of resolutions in a preset resolution interval, performing a preset community discovery algorithm on the time sequence relationship graph according to the plurality of resolutions to obtain corresponding community division results, and determining an optimal community division result based on the normalized mutual information between the community division results; subdividing or merging the sub-communities in the optimal community division result to obtain a target community division result; identifying a core attack node in the target community division result; performing a depth-first search on the target community division result starting from the core attack node to extract a path set having a causal association with the core attack node, and generating a security event report based on the path set.

The invention discloses a normalized mutual information calculation method based on a KSG algorithm, and the method comprises the following steps: 1, estimating original mutual information MI (X, Y) through a k-nearest neighbor distance; 2, estimating variable entropy values delta H (X) and H (Y) based on the same k neighbor principle; 3, normalized mutual information is calculated, the problem that mutual information dimensions are not comparable is solved, an NMI value which is comparable to [0, 1] is output, the NMI value is expanded and applied to feature importance sorting in high-dimensional feature selection, similarity measurement in sample clustering analysis and conditional independence test (using a threshold value delta) in causal discovery, and the method has the advantages of being high in calculation efficiency (O (n log n)) and high in noise resistance.

The invention discloses a high-capacity low-power-consumption internet of things communication de-noising and de-interference method, which relates to the field of wireless communication physical layer signal processing, and comprises the following steps: carrying out short-time Fourier transform on a received Chirp signal to generate a time-frequency diagram, and decomposing the time-frequency diagram into independent time-frequency units through a sliding window; learning noise distribution through forward diffusion and reverse de-noising processes by using a diffusion generative model, and performing iterative de-noising on the noisy time-frequency matrix to obtain a pure time-frequency graph; recording multi-scale hidden layer features in the denoising process, and obtaining a fusion feature map through spectrum normalization and attention weighted fusion; calculating normalized mutual information based on the pure time-frequency graph and the fused feature graph, and if the normalized mutual information is lower than a threshold value, de-noising again; and finally, mapping the reconstructed time-frequency graph into symbol representation through a decoder to complete adaptive demodulation. According to the method, the communication reliability and demodulation accuracy in a complex electromagnetic environment are improved by effectively solving the problems of low signal-to-noise ratio and multi-packet conflict in a high-density Internet of Things scene.

This invention discloses a multimodal anti-counterfeiting verification method and terminal based on microstructure transspectral scattering stability and reliable computation. The method includes: establishing a physical bus access mechanism for direct sensor access through a secure execution environment; extracting dark current fixed-mode noise for hardware fingerprint verification; acquiring visible light images and lidar point clouds and performing spatial mapping and alignment; extracting the visible light texture gradient field and infrared reflectivity distribution field, and calculating local normalized mutual information as a transspectral scattering invariant; inferring the main light source direction based on the point cloud surface normal and a lightweight feature extraction network, and calculating the macroscopic illumination residual; combining the above hardware fingerprint, transspectral microscopic and macroscopic illumination features, and using an adaptive compensation function to generate a comprehensive judgment result. This invention solves the problem of difficulty in defending against low-level data stream hijacking and generative forged images, significantly improving the physical interpretability and reliability of anti-counterfeiting verification.

This invention relates to the field of smart grid technology, specifically a method and system for non-disruptive topology identification of low-voltage distribution areas, including a medium and a terminal. The method comprises: S10, cleaning and filtering raw communication and measurement data; S20, converting the RSSI between devices into distances between nodes and forming an undirected weighted graph; S30, using a clustering algorithm to determine the affiliation between meter boxes and meters; S40, constraining and matching transformers, lines, and meter box nodes based on electricity consumption information, and optimizing the topology using the law of conservation of energy and Kirchhoff's current law; and S50, using adjusted RAND exponents and normalized mutual information to evaluate the accuracy of the identification results. This invention features a simple process and convenient operation. Without adding extra hardware, interrupting power supply, or injecting feature signals, it gradually constructs a complete topology based on existing smart meters, communication signals, and power data from electricity consumption information collection systems, achieving high-precision and automated identification of distribution area topology.

The invention relates to a mutual information and Shapley value fused power system operation reliability diagnosis method and equipment. On the basis of the power system operation reliability agent model, the mutual information and Shapley value fused power system operation reliability diagnosis method is provided, and the influence degree of uncertain factors on operation reliability indexes can be effectively quantified. According to the scheme, firstly, a normalized mutual information index is applied, the interpretation ability of each uncertainty factor for a reliability index is quantitatively evaluated, and key factors are screened out; and calculating the contribution of each key factor to the reliability index through a Shapley value to realize interpretable decomposition and responsibility attribution of the operation reliability of the power system. According to the scheme, interpretable decomposition and quantitative attribution of the influence of the uncertainty factors on the reliability indexes are achieved, an innovative analysis framework is provided for power grid operation risk identification and scheduling decision making, and the method can be used for assisting in further diagnostic analysis of the operation reliability risk of a power system.

The invention discloses an image registration method and device. The method comprises the following steps: acquiring a reference image and a moving image, and extracting significant gradient features of the reference image and the moving image by adopting phase consistency; in response to a measure function constructed based on the significant gradient features of the reference image and the moving image, achieving global optimum, and obtaining a coarse registration parameter; and on the basis of the coarse registration parameters, performing local search operation on the reference image and the moving image by adopting a differential total change method to realize further registration so as to obtain fine registration parameters. According to the method, image layered registration is realized based on significant gradient normalized mutual information and differential total change, so that the registration effect is improved.

The application discloses a kind of high-stability spine ultrasound and X-ray image registration method, set from low to high multi-resolution registration grid, adopts gradient descent method, with the reciprocal of normalized mutual information similarity measure function as objective function, according to the order from low to high resolution, respectively, control point in different resolution registration grid is mutually optimized and calculated, and registration result X-ray image is obtained, registration result X-ray image obtained by registration of preceding low-resolution registration network is used as input X-ray image of subsequent high-resolution registration network, finally, registration result X-ray image generated by highest resolution registration grid is defogged using fuzzy perception attention network, and clear X-ray image after registration is output.The application can improve registration accuracy while ensuring registration stability, and greatly improve the visibility of image.

This invention proposes an optimization method for parallel deep convolutional neural networks based on Winograd convolution, comprising: S1, the model batch training stage, employing the feature filtering strategy FF-CSNMI based on cosine similarity and normalized mutual information, which eliminates redundant feature computation by filtering and then fusing, thus solving the problem of excessive redundant feature computation; S2, the parallel parameter update stage, employing the parallel Winograd convolution strategy MR-PWC, which reduces the computational cost of convolution in big data environments by using parallelized Winograd convolution, thereby improving the performance of convolution operations and solving the problem of insufficient convolution operation performance in big data environments; S3, the parameter combination stage, employing the task migration-based load balancing strategy LB-TM, which reduces the average response time of each node in the parallel system by balancing the load among nodes, improving the efficiency of parallel parameter merging, thus solving the problem of low efficiency in parallel parameter merging. This invention significantly improves both parallel efficiency and classification performance.

This invention discloses a defect monitoring model training method, a GIS equipment defect detection method, and related devices, applied to the field of GIS equipment mechanical defect detection. The method employs an improved adaptive noise complete set empirical mode decomposition to perform feature decomposition on the mechanical vibration signal of GIS equipment under variable frequency current excitation. It then uses normalized mutual information calculation to effectively screen the intrinsic mode functions and reconstruct the GIS equipment mechanical vibration signal. Finally, it extracts the feature matrix of the reconstructed GIS equipment mechanical vibration signal for model training to obtain a defect detection model. This invention achieves feature decomposition and signal reconstruction of the GIS equipment mechanical vibration signal under variable frequency current excitation through mode decomposition and normalized mutual information calculation. By extracting and training the feature of the reconstructed GIS equipment mechanical vibration signal, the defect detection model is trained. Compared with existing technologies that detect GIS mechanical defects based on a single power frequency current, this invention improves defect detection accuracy.

The invention relates to a zero sample industrial fault diagnosis method based on prototype prediction, and belongs to the field of industrial fault diagnosis. Comprising the steps of performing multi-scale adaptive sparse coding feature extraction on a visible fault sample, and constructing a visible fault prototype; a variable contribution degree is quantitatively calculated by adopting a variational auto-encoder in combination with SHAP value analysis, and a variable correlation matrix is obtained based on normalized mutual information; constructing a contribution-weighted fault similarity matrix, and quantifying the similarity between visible faults and non-visible faults; using the similarity matrix and the visible prototype to predict an unseen fault prototype, and determining the fault category of the test sample through clustering and cosine similarity. According to the method, the average accuracy rate of TE process diagnosis tasks is high, high-precision unseen fault diagnosis can be achieved only by depending on simple fault description, the problems that an existing method excessively depends on domain knowledge and is insufficient in generalization ability under complex working conditions are solved, accurate diagnosis of unseen fault samples and detailed semantic attributes is not needed, and an effective solution is provided for industrial fault diagnosis.

The invention provides a self-adaptive normalized mutual information context learning optimization method and system oriented to unbalanced data, and relates to the technical field of natural language processing. The problems of numerical instability, reference distortion and unbalanced data in context learning are effectively solved by constructing a domain background pseudo context to replace an empty character string, quantifying inherent prejudice of the domain background, calculating a smooth normalized mutual information score, applying an adaptive regularization weight and screening an optimal example arrangement; the method has the advantages that the stability of the model in unbalanced data context learning is improved, numerical explosion is avoided, field background prejudice is accurately quantified, and real label distribution is adaptively aligned.

The present invention discloses a method for adjusting zoom mismatch of heterogeneous image fusion based on edge gradient mutual information, which belongs to the field of heterogeneous image fusion. The present invention extracts edges from pre-processed heterogeneous input images, performs normalized mutual information statistics on heterogeneous edge grayscale images, and then performs gradient weighted statistics on edge images, and uses it as the weight of edge normalized mutual information, and uses edge gradient mutual information as the evaluation index of heterogeneous field of view matching; adopts a hill climbing search method improved by optimizing the search direction change strategy, uses three adjacent images as criteria, searches for the maximum value of the evaluation index, that is, the most matching position, and solves the problem of slight mismatch of field of view of continuous zoom heterogeneous fusion imaging system. The present invention improves the grayscale correlation of heterogeneous images, constructs gradient weighting that is sensitive to slight changes in field of view, and enhances the unimodality and sensitivity of the evaluation index; the present invention can effectively suppress the influence of local extreme values ​​and improve zoom registration accuracy.

The invention discloses a personality classification method and device, electronic equipment and a storage medium. The method comprises the following steps: obtaining resting state functional magnetic resonance image data of a plurality of individuals; performing spatial independent component analysis on the resting state functional magnetic resonance image data of each individual to obtain a spatial independent component of each individual; determining a standardized mutual information similarity matrix of a cross-individual spatial mode based on the spatial independent component of each individual; and determining a personality label of each individual based on the standardized mutual information similarity matrix of the cross-individual spatial pattern. According to the technical scheme, the independent component analysis method and the resting state functional magnetic resonance technology are combined, completely objective data-driven personality classification is achieved, the inherent limitation that a traditional personality classification method depends on subjective questionnaire measurement is broken through, and personality traits are classified more objectively and accurately.

The application discloses a pathological section image lesion grading and classification detection method, and comprises the following steps: step one, using an automatic label labeling method based on normalized mutual information registration to label the cropped image, and generating a data set; step two, improving a DenseNet121 network, and introducing a focal loss function to balance the weights between multiple tissue samples; step three, adding a hybrid attention module between each dense block and conversion layer in the improved DenseNet121, and adding a local supervision function to the loss function of the modified network; step four, combining the two improved network systems in steps two and three to form a lesion grading and classification detection scheme. The application can improve the detection speed and accuracy of lesion images, thereby reducing the cost of manual identification and expanding the application prospect of the medical image classification field.

The invention provides a landslide change detection sample enhancement method considering topographic factors, and relates to the technical field of image processing, and the method comprises the steps: cutting a landslide sample, obtaining image element clusters of which the true value labels are landslides, taking the image element clusters as landslide Mask samples, and carrying out the X-axis and Y-axis random overturning and / or random angle rotation; selecting a scene sample which does not contain a landslide mark, calculating normalized mutual information of a landslide post image corresponding to a landslide Mask sample and a post time phase image of the scene sample, obtaining a candidate scene, calculating the slope direction and gradient of each pixel, and obtaining an easy-to-develop slope area of the landslide; and acquiring a collage candidate area in a slope area where the landslide is easy to develop, calculating an average slope direction, judging the slope direction, and collaging the converted landslide Mask sample on the collage candidate area to obtain a new landslide change detection sample. The landslide pattern spots are spliced to the slope areas obviously different from the landslide pattern spots, the landslide change detection samples are quickly generated, and reasonability is guaranteed.

The invention discloses a network security event report generation method and device based on a dynamic resolution community discovery algorithm, equipment and a medium, and relates to the field of network security, and the method comprises the steps: extracting target information in original network security alarm data as nodes, and constructing a time sequence relation graph based on directed edges between the nodes; enumerating a plurality of resolutions in a preset resolution interval, executing a preset community discovery algorithm on the sequential relation graph according to the plurality of resolutions to obtain corresponding community division results, and determining an optimal community division result based on normalized mutual information among the community division results; subdividing or combining the sub-communities in the optimal community division result to obtain a target community division result; identifying a core attack node in the target community division result; and executing depth-first search on a target community division result by taking the core attack node as a starting point so as to extract a path set which has causal association with the core attack node, and generating a security event report based on the path set.

The invention provides a disease recognition model training method and device and a disease recognition method and device, and belongs to the field of artificial intelligence. Each first electroencephalogram signal comprises electroencephalogram signals of multiple channels under the same time sequence and multiple frames of second electroencephalogram signals divided according to a preset sliding window in the time sequence direction; determining a three-dimensional standardized mutual information matrix corresponding to the plurality of first electroencephalogram signals as a training sample; the three-dimensional standardized mutual information matrix corresponding to the first electroencephalogram signal is obtained by combining two-dimensional standardized mutual information matrixes corresponding to multiple frames of second electroencephalogram signals; and training a deep learning model based on the attention mechanism at least by using the training sample after the attention weighting reconstruction. According to the method, the connection relation of electroencephalogram signals among different channels can be effectively captured, and the extraction accuracy of information for disease recognition is improved.

The application provides an adaptive normalized mutual information context learning optimization method and system for unbalanced data, and relates to the technical field of natural language processing. By constructing a field background pseudo-context to replace an empty string, quantifying inherent bias of the field background, calculating a smoothed normalized mutual information score, applying adaptive regularization weights, and screening optimal example arrangements, the problems of numerical instability, benchmark distortion and unbalanced data in context learning are effectively solved, and the method has the advantages of improving the stability of the model in unbalanced data context learning, avoiding numerical explosion, accurately quantifying the bias of the field background, and adaptively aligning the real label distribution.

The application relates to an electromagnetic signal classification method based on case-based selection and multi-target linear programming, which comprises the following steps: training a first classification model by using old class data of electromagnetic signals including multiple classes; reordering old class data training samples according to normalized mutual information between real labels and predicted labels in the training; and storing part of training samples in the old class data of each class according to classification difficulty; for new class data of the increased multiple classes, performing multi-target linear programming on the basis of the first classification model, updating the classifier, and obtaining a second classification model; and performing feature fine-tuning training of the second classification model by using the stored old class data and the training samples of the increased new class data, obtaining an incremental learning classification model, and using the incremental learning classification model to classify new class data and old class data. The application not only realizes high classification accuracy for new added classes, but also can maintain the classification ability for old class electromagnetic signals.

The invention provides a water diversion pipeline risk assessment method and system and a storage medium. The method comprises the steps of obtaining time sequence monitoring data of a preset monitoring node of a water diversion pipeline; for the time sequence monitoring data of any two monitoring nodes, calculating normalized mutual information in a preset time delay search domain, and generating a time delay mutual information spectrum; establishing an event tree analysis model, and determining an initial event occurrence probability through a probability mapping function when the event association feature vector and the GIS vulnerability level of the grid unit where the monitoring node is located meet a preset triggering condition together; for non-initial event nodes, the conditional transition probability is calculated based on the event association feature vector and the upstream path cumulative probability; for any scene path, the risk consequence value is calculated according to the GIS vulnerability grade, the real-time hydraulic working condition parameter and the failure mode of the corresponding grid unit; and traversing all scene paths of the event tree, multiplying the occurrence probability of each path by the risk consequence value, and summing to obtain a comprehensive risk index of the water diversion pipeline.

This invention provides a laboratory safety self-control method and system for fault self-diagnosis. The method includes: S1 periodically and actively generating health labels through self-checks; S2 dividing the time slice of a single processor into execution time slots and self-test time slots, calculating the CRC signature of key data in the execution time slot, executing test vectors and comparing them with historical signatures in the self-test time slot, and locking the output to a safe state via independent hardwired connection in case of a fault; S3 based on incremental causal graph and online parameter learning, collecting samples using fault labels, updating the conditional probability table through online EM, automatically expanding the causal graph structure through normalized mutual information, and outputting the fault type and confidence level using confidence propagation; S4 actuator closed-loop feedback and hardwired bypass protection; S5 dynamic reconstruction degradation fault tolerance; S6 emergency shutdown and solidification. This invention achieves high-coverage fault detection with a single processor, eliminates blind spots of common faults, and simultaneously achieves small-sample adaptive and accurate fault reasoning, significantly improving system reliability and intelligence.