37 results about "Feature synthesis" patented technology

The invention relates to the technical field of defect detection, and particularly discloses an electronic component defect detection method, system and device, and the method comprises the steps: collecting a multiband original image set based on multispectral imaging, inhibiting batch material difference and surface reflection through orthogonal complementary space projection, and obtaining a de-noised image set; performing feature extraction on the de-noised image set, and identifying and marking all fragmented defect areas; based on a graph neural network space attention mechanism, establishing an association relationship between fragmented defect areas to form a complete defect feature chain; gray frequency value sequences of the to-be-detected image and the standard sample image are constructed respectively, and a first fitting curve and a second fitting curve are obtained through fitting of a skewed distribution function; and finally, comprehensively judging whether the electronic component has defects or not by combining the defect characteristic chain and the difference characteristics of the two fitting curves. According to the method, interference can be effectively suppressed, fragmented defects can be accurately identified, and the detection accuracy and reliability are remarkably improved through multi-dimensional feature fusion.

The invention relates to the technical field of artificial intelligence, and discloses an artificial intelligence-driven multi-modal data analysis method and system, and the system comprises a data collection module which is used for setting collection nodes and terminals, collecting multi-modal data and equipment operation parameters in real time, and carrying out the format conversion; the multi-modal fusion module is used for constructing a cross-modal semantic association model, processing space-time characteristics, fusing characterization characteristics and performing association simulation; the feature extraction module is used for constructing image, text and voice related extraction algorithms to obtain corresponding features; the comprehensive analysis module is used for carrying out correlation calculation on the features after standardization processing, and comparing and analyzing semantic consistency with a preset threshold value; and the grading processing module is used for combining context calculation analysis when semantics are inconsistent, and analyzing target scene performance after secondary comparison. The system and the method can effectively process multi-modal data, improve semantic analysis accuracy and efficiency, and are suitable for various scenes.

The invention discloses a bearing fault diagnosis method based on feature mode decomposition and heterogeneous model fusion, and belongs to the technical field of mechanical fault diagnosis. An original vibration signal of the rolling bearing is collected; independent of any bearing geometric parameter or fault characteristic frequency information, a finite impulse response (FIR) filter is iteratively optimized to maximize a correlation kurtosis (CK) value, and a plurality of intrinsic mode functions (IMF) are adaptively decomposed; selecting first M IMFs with the highest CK value, extracting time domain, frequency domain and nonlinear complexity features of the IMFs, generating high-order derivative features by adopting depth feature synthesis DFS, and constructing an enhanced fusion feature vector; and finally, inputting a stacked integrated classifier consisting of six types of heterogeneous basic learners and an XGBoost element learner, and outputting a fault diagnosis result. According to the method, high-precision and high-robustness diagnosis can still be realized under the working conditions of speed change, strong noise and weak faults.

The invention discloses a focus auxiliary recognition method and device based on deep learning, and relates to the technical field of image recognition. The method comprises the following steps: acquiring a CT image of a target area, taking the CT image as a target image, and executing an image enhancement operation according to the target image to obtain a first feature map; performing image reconstruction operation according to the first feature map to obtain a first reconstructed image, and inputting the first feature map into a decoder to obtain a first continuous vector; the first continuous vector is a continuous vector obtained by decoding a first feature map; performing an abnormal feature synthesis operation according to the first continuous vector to obtain a discrete feature vector; inputting the discrete feature vector into an encoder to obtain a second reconstructed image; fusing the first reconstructed image and the second reconstructed image to obtain a final reconstructed image; performing feature classification network operation on the discrete feature vector to obtain an abnormal feature vector; and integrating the abnormal feature vector and the final reconstructed image to obtain a second abnormal feature graph. The method improves the definition of the image and the accuracy of diagnosis.

The invention discloses a sewer pipe network defect detection method based on multi-label zero sample learning. The method comprises the following steps: generating defect description corresponding to each pipeline defect category through a large language model; performing feature extraction and domain adaptation on the pipeline inner wall image and the defect description by using a representation guide module to obtain global and local image features and defect detailed description text features; synthesizing global and local image features of the image, respectively calculating global and local matching scores of the global and local image features and detailed description text features, and fusing to obtain a defect initial prediction score; constructing a semantic relationship adjacency matrix for displaying the relationship between different defect categories; and correcting the initial prediction score by using the semantic relationship adjacency matrix between the categories to obtain a final prediction score of each defect category. According to the method provided by the invention, knowledge migration from known defects to unknown defects is realized by constructing a guidance-fusion-correction network, and the difficulty in identifying types of unseen defects is effectively solved.

This application relates to the field of computer technology, providing an intent prediction method, apparatus, electronic device, and storage medium. The intent prediction method includes: first, acquiring target user data, which includes data associated with a target product; then, determining comprehensive characteristics corresponding to the target user based on the target user data, including user profile features, sentiment features, and behavioral features; finally, obtaining a target probability based on the comprehensive characteristics of the target user and a target prediction strategy, whereby the target probability represents the probability that the target user will purchase the target product. The target prediction strategy is correlated with the amount of historical user data corresponding to the target product, and the amount of historical user data is positively correlated with the product's shelf life. This allows the target prediction strategy to evolve gradually over time, enabling it to match prediction needs at different stages and improving prediction accuracy.

The application discloses a kind of neurogenetic disease identification model construction method and system, it is related to neurogenetic disease identification technical field.The method includes: obtaining multiple identification results and multiple user feature sets of neurogenetic disease identification, and multiple inter-class difference degrees are obtained by analysis;Based on multiple identification results and multiple user feature sets, train first identification model;According to multiple inter-class difference degrees, configure user feature synthesis strategy, generate synthetic user feature set and synthetic identification result, carry out the reinforcement learning of first identification model, obtain second identification model;Test the combined identification error rate of second identification model, combined with multiple inter-class difference degrees again configure synthetic strategy, continue to generate synthetic user feature set and synthetic identification result, carry out reinforcement learning, obtain third identification model.The application solves the problem that neurogenetic disease identification model is poor in generalization performance, easily confused class and low in recognition accuracy under small sample condition.

The application discloses an automatic feature engineering method based on Monte Carlo tree search and a large language model, which comprises inputting feature data such as a table data set into a computer system; the computer system performs feature initialization, evaluates the score of each initial feature input, and constructs an initial feature tree structure; the steps of candidate feature selection, expansion based on a large language model, feature score evaluation and score parameter updating are iteratively performed until the termination condition is met; based on the selected features, the large language model is called in combination with the prompt word engineering to expand and generate new feature nodes; the new feature nodes are substituted into a predefined end-to-end machine learning pipeline for evaluation, and the feature score is comprehensively calculated; the related parameters of all ancestor nodes in the feature tree are updated from bottom to top along the feature synthesis path; and after the iteration is terminated, the optimal feature set capable of improving the prediction performance of a basic machine learning model is output. The application can significantly improve the accuracy of a machine learning model in a table data prediction task.

A method includes obtaining a first dataset including a first feature set, generating a first set of feature values by providing the first dataset to a set of feature primitive stacks, and determining a reduced set of feature values based on the first set of feature values by dimensionally reducing features of the first set of feature values. The method further includes generating an intermediate set of feature values by providing a value of the first dataset and a value of the reduced set of feature values to at least one feature primitive of the set of feature primitive stacks. The method further includes updating the reduced set of feature values by dimensionally reducing features of the intermediate set of feature values and storing a second dataset including features of the intermediate set of feature values in association with the first feature set.

At least a method and an apparatus are presented for efficiently encoding or decoding video using neural networks wherein the bitstream is adapted to hybrid machine / human vision applications. For example, the scalable decoding comprises applying to a tensor of reconstructed data a neural network-based feature synthesis processing to generate a tensor of input feature representative of a feature of image data samples, resizing the tensor of input feature to generate a tensor of output feature intended to be fed a neural network-based vision inference processing to generate a collection of inference results. Advantageously, resizing the tensor of input feature adapt at least a dimension of the tensor of input feature to the neural network-based vision inference processing.

The application relates to a driving risk identification method based on an automatic timing hyperparameter optimization model, aiming to construct an automatic timing hyperparameter optimization model based on space-time trajectory data to identify driving risk behaviors, reduce training costs and optimize model precision on the basis of an existing model. The method constructs an automatic timing hyperparameter optimization model, builds an automatic machine learning framework based on vehicle space-time trajectory data, realizes automatic optimization of sliding window parameters and model hyperparameters, first synchronizes the phase coupling relationship between multiple feature data based on a dynamic time warping algorithm; second, automatically generates and reduces the dimension of features based on a deep feature synthesis-sliding window algorithm and a t-distributed stochastic neighbor embedding algorithm; finally, automatic model selection and hyperparameter adjustment are realized through Bayesian optimization, and model integration is carried out; the method considers the time sequence characteristics of data in the automatic machine learning framework, realizes the reduction of training costs and the optimization of the precision of the driving risk identification model.

The application provides a digital prototype lightweight system based on feature extraction and adaptive grid optimization, and relates to the technical field of computers, comprising a model access and preprocessing module and a file analysis unit; the file analysis unit is used for receiving an original digital prototype model, and analyzing geometric topology data and design tree information of the model; a feature collaborative decision unit receives output results of the three units, realizes screening and classification of multi-dimensional features through feature influence weight calculation, and generates a feature mapping diagram containing key feature areas and non-key feature areas; through multi-dimensional collaborative extraction of geometric analysis, topology analysis and semantic analysis, and in combination with weight coefficient calculation of feature comprehensive influence degree determined according to downstream CAE analysis sensitivity, key areas having influences on product performance and assembly precision can be accurately identified, compared with an existing single-dimensional feature extraction scheme, loss of functional key features is avoided, and a basis guarantee for subsequent CAE analysis precision is provided.

The invention discloses an automatic feature engineering method based on Monte Carlo tree search and a large language model. The method comprises the following steps: inputting feature data such as a table data set into a computer system; the computer system executes feature initialization, evaluates the score of each input initial feature, and constructs an initial feature tree structure; iteratively executing the steps of candidate feature selection, expansion based on a large language model, feature score evaluation and score parameter updating until a termination condition is met; on the basis of the selected features, calling a large language model in combination with cue word engineering, and expanding to generate new feature nodes; substituting the new feature nodes into predefined end-to-end machine learning pipeline evaluation, and comprehensively calculating feature scores; related parameters of all ancestor nodes in the feature tree are updated from bottom to top along the feature synthesis path; and after iteration is terminated, outputting an optimal feature set capable of improving the prediction performance of the basic machine learning model. According to the method, the accuracy of the machine learning model in the table data prediction task can be remarkably improved.

The invention discloses a storage environment intelligent regulation and control method and system based on deep learning driving, and relates to the field of artificial intelligence, and the method comprises the steps: extracting the operation effect characteristics of a to-be-executed environment regulation and control operation through constructing a first storage region state association map; carrying out interactive calculation on the parameter characteristics of different characteristic domains of the driving unit to obtain multi-dimensional parameter characteristics of the driving unit; comprehensively generating a regulation and control instruction set for each storage area category (the area category is determined by storage state parameters and / or historical environment parameters) in combination with the operation parameter characteristics; and finally, the corresponding area is regulated and controlled. According to the method, accurate and intelligent regulation and control of the storage environment are realized through multi-dimensional feature fusion and a regional differentiation strategy, the regulation and control effect and the energy efficiency ratio are improved, and the method is suitable for intelligent management of complex storage scenes.

The application discloses a sewer network defect detection method based on multi-label zero-shot learning, and the method comprises the following steps: generating a defect description corresponding to each pipeline defect category through a large language model; using a representation guide module to perform feature extraction and field adaptation on a pipeline inner wall image and the defect description, so as to obtain global and local image features and defect detailed description text features; comprehensively considering the global and local image features of the image, respectively calculating global and local matching scores of the image and the detailed description text features, and fusing the global and local matching scores to obtain an initial defect prediction score; constructing a semantic relation adjacency matrix for displaying the relation between different defect categories; and using the semantic relation adjacency matrix between the categories to correct the initial prediction score, so as to obtain a final prediction score of each defect category. The method disclosed by the application realizes knowledge transfer from known defects to unknown defects by constructing a guide-fusion-correction network, and effectively solves the problem of recognizing unknown defect types.

According to the asynchronous federated prediction method for the integrated energy system fusing the multi-time scale characteristics, the federated learning can construct a higher-precision load prediction model under the condition of guaranteeing data privacy for a plurality of integrated energy systems with similar energy consumption characteristics. For the situation that the processing speeds of different nodes have significant differences, the execution efficiency of the existing federated learning method is generally low due to the influence of an adopted global synchronization mechanism. For two types of load data of a long time scale and a short time scale, the framework respectively constructs an asynchronous federated learning model, and each model can better adapt to the fluctuation rule of the emphasized load while the coupling relationship between different loads is fully used, so that the accuracy of the model is improved. According to the local model parameter weighted aggregation mechanism based on the comprehensive contribution rate, the appropriate aggregation weight can be autonomously given according to the importance degree of the local model, and then the performance of a global prediction model on a server side is improved.

The invention provides a revocable PPG template design privacy protection method based on a CNN network, and relates to the field of safe and efficient authentication based on revocable PPG signals, and the method comprises the steps: carrying out the noise reduction and time-frequency conversion of an original PPG signal, and obtaining a two-dimensional time-frequency image; performing feature extraction on the two-dimensional time-frequency image by using a pre-trained convolutional neural network model to obtain an original depth feature vector; and performing discontinuous segmentation, random rearrangement, feature synthesis and coding on the original depth feature vector to generate an irreversible, revocable and unlinkable binary template. According to the technical scheme provided by the invention, high accuracy is realized, so that the authentication efficiency is improved, and meanwhile, the security of PPG signal authentication is enhanced.

The present application relates to the technical field of intelligent evaluation of water conservancy dam, in particular to a reservoir dam safety intelligent evaluation system based on machine learning, comprising a data acquisition module, a preprocessing module, a multi-source feature analysis module, a state fusion evolution module and a dynamic evaluation output module. The data acquisition module acquires multiple types of monitoring time series data, the preprocessing module aligns multiple dimensions of data, and generates a unified time scale record sequence containing three types of exclusive sequences; the multi-source feature analysis module extracts structural behavior deep features, comprehensive load state spectrum and abnormal event feature coding respectively; the state fusion evolution module generates a safety comprehensive state atlas through feature fusion network interaction evolution; and the dynamic evaluation output module constructs an evaluation network based on the atlas to complete dynamic scoring and form a graded safety state list. The present application realizes accurate characterization and dynamic evaluation of dam safety state through multi-dimensional data regularization and multi-feature interaction fusion.

The invention discloses an industrial time series data prediction method and device based on multistage feature selection. The method comprises the following steps: acquiring multi-source heterogeneous time series data, and carrying out de-noising smoothing by adopting a rolling truncation mean value; constructing a displacement prediction data set, and establishing a mapping relation between a historical observation window and a future target moment; the feature comprehensive relevancy is calculated based on an entropy weight method, the features are divided into strong, medium and weak correlation subsets, and maximum correlation minimum redundancy screening, complementarity information screening and complementary screening based on gating attention are adopted respectively to form a final feature subset; and constructing a heterogeneous base learner group by using CNN-LSTM and CatBoost, performing probability fusion in combination with a Bayesian ridge regression element learner, and outputting a prediction result with a confidence interval. According to the method, the problems of noise interference, feature redundancy and lag in industrial time series data prediction are effectively solved, and the prediction precision and robustness are improved.

The embodiment of the application belongs to the technical field of voice recognition, is applied to a voice comparison recognition scene, and relates to a voiceprint comparison recognition method, device and equipment and a storage medium thereof. The voiceprint comparison recognition method comprises the following steps: obtaining to-be-compared voice; performing voiceprint feature extraction; performing linear mapping layer mapping to obtain high-dimensional feature vector representation; inputting the high-dimensional feature vector representation into a feature representation fusion model to obtain feature comprehensive representation corresponding to the to-be-compared voice; calculating a distance value between the feature comprehensive representation corresponding to the to-be-compared voice; and identifying voice segments of the same object from the to-be-compared voice based on the distance value. Compared with direct comparison of voiceprint features, comparison based on feature comprehensive representation can reduce the calculation of comparison quantities to a certain extent and improve comparison recognition efficiency. The voiceprint comparison recognition method of the application is applied to the technical field of biometric recognition, for example, a bank or financial service voice unlocking scene, and can more safely provide fund security guarantee services for customers.

The invention relates to the technical field of data processing, and particularly provides a student performance prediction method and system based on a Mama state space model, and the method comprises the steps: carrying out the preprocessing and vectorization representation of multi-source student original feature data, mapping different types of features into feature vectors of a unified dimension based on a feature type perception strategy, and carrying out the prediction of the performance of a student. Obtaining a feature sequence; inputting the feature sequence into a Mama state space model, processing the feature sequence through the Mama state space model, and obtaining a hidden state vector containing all feature comprehensive information; and constructing a regression prediction model, inputting the hidden state vector into the regression prediction model, and obtaining a student performance prediction result. According to the method, complex interaction between features is effectively captured through sequence modeling, so that better performance can be achieved on prediction accuracy, the parameter scale and calculation complexity are greatly reduced, the model is more difficult to over-fit on a small data set, and training is more stable.

The invention discloses a data asset classification and level-to-level management method and system based on artificial intelligence, and the method comprises the steps: firstly collecting enterprise comprehensive asset data and a data classification standard in a period, carrying out the filling of numerical data, carrying out the abnormal processing, carrying out the normalization, and carrying out the cleaning, error correction and filling of text data, so as to obtain preliminary comprehensive data; performing key feature analysis on each text in the preliminary comprehensive data by using an improved natural language processing technology, collecting word conversion vectors corresponding to feature words in all texts, and obtaining feature comprehensive data; performing clustering analysis on each text in the feature comprehensive data by using a K-means technology to obtain a similar hierarchical data set; performing grade matching on each group in the similar grading data set by using a grading scheme design model to obtain a data classification grade corresponding to each group; according to the data classification level corresponding to each group, protection measures in the data classification standard are used for management, and level-to-level management of enterprise data assets is achieved.

The invention discloses a handwriting dynamics cascade integrated learning screening system, and relates to the technical field of dynamics screening. The handwriting dynamics cascade integrated learning screening system comprises an acquisition and preprocessing module for preprocessing original dynamics signal data and risk feature data; the handwritten data quality evaluation and grading module is used for judging the acquisition quality according to a signal coupling evaluation analysis result, executing an interaction prompt mechanism and entering a risk grading process; the dynamic index judging and prompting module is used for executing a risk grading process according to a sensitivity amplification theory analysis result; the abnormal risk scoring module is used for training a shallow feedforward neural network model; and the risk feature integration and early judgment module is used for carrying out early dynamic anomaly judgment. The problems that an existing handwriting dynamics screening system is insufficient in early-stage and tiny abnormal feature extraction and judgment sensitivity, and the accuracy and reliability of disease early screening are difficult to achieve are solved.

The invention discloses a product parameter optimization method and system based on user experience, and relates to the technical field of artificial intelligence. The method comprises the following steps: acquiring user experience evaluation data for a plurality of product samples; on the basis of the user experience evaluation data, a variance fluctuation uncertainty multi-attribute weighting strategy is adopted, weights of multiple evaluation dimensions are calculated, and a comprehensive user experience evaluation value of each product sample is determined; training a radial basis function (RBF) neural network prediction model by taking the product parameter combination as an input feature and integrating the user experience evaluation value as a target variable, and optimizing parameters of the RBF neural network by adopting an improved sparrow search algorithm ISSA; and utilizing the trained RBF neural network prediction model to determine an optimal product parameter combination which maximizes the comprehensive user experience evaluation value.

The invention provides a student behavior tracking method based on a video stream and a computer program product thereof. According to the student behavior tracking method, a to-be-detected image containing student behaviors in a video stream is obtained, and the head of a student in the to-be-detected image is matched with a corresponding body. The behavior of the student can be analyzed according to the matching result of the head and the body of the student. When the head and the body of the student are matched, the head and the body of the student can be comprehensively matched by combining the central point coordinates of the head detection frame and the body detection frame and the features obtained by feature extraction of the to-be-detected image, so that the accuracy of the obtained matching result is relatively high; and the student behavior analyzed according to the matching result is more in line with the real behavior of the student. And parents and teachers of the students can specify personalized learning schemes according to the analyzed behaviors of the students.

The feature selection method, device and electronic equipment provided in the application relate to the field of data processing, and include the following steps: obtaining an input feature data set; preprocessing the input feature data set to obtain a plurality of standardized features; obtaining a pre-constructed feature weight coefficient module; inputting each standardized feature into the feature weight coefficient module to obtain a self-correlation score corresponding to each standardized feature; performing operation on each standardized feature according to an attention mechanism to determine a mutual correlation score between each standardized feature; multiplying and fusing the self-correlation score and the mutual correlation score to obtain a feature comprehensive score; and sorting each standardized feature according to the feature comprehensive score to obtain a sorted target feature set. The method provided in the application achieves the effect of improving the accuracy and adaptability of feature selection.

The invention relates to the technical field of intelligent prediction and chip design, in particular to a power load prediction method based on fused power features and a chip. The method comprises the following steps: acquiring a power load data set of a power system in a specified historical time period; performing feature extraction on the power load data set to obtain a high-dimensional power feature data set, and performing data dimension reduction on the high-dimensional power feature data set by using two dimension reduction methods to obtain a first low-dimensional power feature data set and a second low-dimensional power feature data set; performing linear fusion on the first low-dimensional power feature data set and the second low-dimensional power feature data set to obtain a fused power feature data set; training a machine learning model based on the fused power feature data set to obtain a power load prediction model; and performing power load prediction by using the power load prediction model. According to the invention, redundant information can be reduced, and key features can be reserved. The advantages of different dimensionality reduction modes are integrated, and the accuracy of power load prediction is improved.

The application discloses a counterfeit audio identification method for characteristic synthesis, relates to the technical field of audio authenticity identification, and is used for diversified identification of audio forgery in a deep level. The method comprises the following steps: preprocessing audio to be identified, obtaining human voice audio data, environmental audio data and real age information of human voice of the audio to be identified; establishing a sound quality comparison model, analyzing the human voice audio data and the environmental audio data to generate first comparison values and second comparison values, and performing classification processing according to the first comparison values and the second comparison values to generate a sound quality difference identifier; establishing a human voice matching model, analyzing and matching the human voice audio data to generate matching age information; analyzing and processing the matching age information and the real age information to generate an audio age matching identifier; and constructing a data correlation output model, integrating the sound quality difference identifier and the audio age matching identifier to generate a counterfeit audio target, a suspicious audio target and a real audio target.

The invention discloses an image feature extraction method and device based on an improved shielding self-encoding model, and relates to the field of computer vision, and the method comprises the steps: carrying out the feature extraction of a target image obtained through a pre-trained improved shielding self-encoding model, so as to obtain a feature vector; the pre-training process of the model comprises determining a mask pattern for an input image so as to input the mask pattern into the model to obtain a predicted complete image, the improved shielded self-encoding model comprises an encoder, and the encoder comprises an improved multi-head attention module constructed based on a preset feature point level-multi-window level feature comprehensive capture strategy; the encoder is used for outputting spatial scale features corresponding to each layer so as to obtain a predicted complete image, and model parameter correction and continuous training are carried out until a preset stop condition is reached. The improved shielding self-encoding model in the scheme realizes joint modeling of feature point level and dynamic multi-window scale features, and is beneficial to enhancing the perception capability of the model for different scales and coarse and fine granularity features.

The application discloses a multi-feature grouping Transformer neural network and method, and relates to the technical field of motor testing; the neural network comprises a multi-feature extraction module, a Transformer feature extraction module and a feature grouping fusion classification module which are connected in sequence; the multi-feature extraction module comprises n MFCC modules; the Transformer feature extraction module comprises n Transformer modules and an embedded full connection layer; the feature grouping fusion classification module comprises a splicing module, a feature comprehensive full connection layer, an average pooling layer and a binary classification full connection layer; the method comprises the following steps: S1, multi-feature extraction; S2, Transformer deep feature extraction; and S3, feature grouping fusion classification to obtain a binary classification result; the spliced embedded feature vector overcomes the problems of mixed information and strong interference, reduces the parameter quantity and lowers the calculation amount.