57 results about "Hypersphere" patented technology

The invention discloses a hyperspectral image open set classification method based on prototype similarity distribution modeling. The method comprises the following steps: acquiring a training sample set and a test sample set; initializing a category prototype on the unit hyper-sphere and forming uniformly distributed prototype spaces through iterative optimization; training a deep learning network based on a joint loss function, guiding the samples to form high-discrimination similarity distribution in a prototype space, and calculating similarity vectors of the samples and prototypes of each category; training the OCSVM by using the similarity vector of the known class sample to learn the distribution boundary of the known class in the prototype space; in the test stage, a test sample is input into the deep learning network, the similarity vector of the test sample is calculated, and whether the test sample belongs to a known class or not is judged by the OCSVM; and if the class is a known class, determining the fine-grained classification according to the class prototype of the maximum similarity component in the similarity vector. According to the method, the interpretability and generalization of the open set classification model are enhanced, and the misrecognition risk between similar categories is reduced.

The invention provides a large language model generation code detection method and system, which is applied to the technical field of artificial intelligence, and comprises the following steps: obtaining a to-be-detected code; a to-be-detected code is input to a trained shared encoder, a code feature vector is obtained, the shared encoder is obtained through multi-target joint training, and the multi-target joint training is used for optimizing classification loss, domain confrontation loss, comparison loss and difficult sample loss at the same time; l2 normalization is carried out on the code feature vector, and the code feature vector is mapped to a hyperspherical space to obtain spherical embedding; the sphere is embedded and input into a sphere category classifier based on sphere logistic regression for classification processing, a detection result of the to-be-detected code output by the sphere category classifier is obtained, the detection result comprises AI generation and human writing, and a decision boundary of the sphere category classifier is an intersection line of a hyperplane and a hypersphere for classification. According to the invention, the AI generation code and the human compiled code can be accurately distinguished.

The invention discloses an LSTM-SVDD anomaly detection method and system based on rule guidance, belongs to the technical field of industrial nondestructive quality inspection, and aims to solve the technical problems that noise samples are sensitive and excessively depend on annotation of abnormal data, and expert experience and knowledge cannot be effectively introduced. Constructing a feature extraction model based on a bidirectional LSTM network, a rule template engine and a feature projection fusion layer; calculating a hypersphere radius through a quantile method, constructing an SVDD loss function as a basic loss function, introducing a rule penalty term to construct a rule loss function, carrying out weighted summation on the basic loss function and the rule loss function to construct a total loss function, and obtaining a trained feature extraction model and an optimized hypersphere radius through minimizing the total loss function; and inputting a to-be-detected sample into the trained feature extraction model, calculating an Euclidean distance between a fusion feature vector and the center of the hypersphere, and comparing the Euclidean distance with a dynamic detection threshold value.

The invention discloses a method and a system for monitoring pressure abnormity of a fuel oil common rail pipe of a marine main engine. The method comprises the following steps: acquiring historical operation parameter data and constructing a data set; the historical operation parameter data comprises historical host load data, historical host rotating speed data and historical fuel oil common rail pressure data; historical operation parameter data are preprocessed, an SVDD model is constructed and trained based on the historical operation parameter data, the model obtains a historical normal data hyper-sphere in a three-dimensional feature space, and the center and the radius of the hyper-sphere are obtained; similarly, real-time operation parameter data collected in real time are preprocessed and then input into the model, the distance between the real-time operation parameter data and the center of the hypersphere is obtained, and the value of the distance is compared with the value of the radius of the hypersphere in historical normal data; and when the numerical value of the distance is large, it is judged that the real-time operation parameter data is within an abnormal range, the fuel oil common rail pressure of the ship main engine is in an abnormal state, and an alarm signal is triggered. The method has the effect of low error rate.

The application discloses an iris recognition method and device, system and storage medium, and performs preprocessing such as polar coordinate unfolding on an input original iris image in sequence; multi-level image features are extracted through an inverse residual module; subsequently, a global depth convolution GDConv operator is introduced to replace a global average pooling layer, so that spatial structure distribution features of iris textures are retained under the premise of extremely low calculation complexity; finally, an ArcFace loss function is used to optimize the distribution of features in a hyperspherical space, and the class separability of the features is enhanced by introducing an additive angle interval. By using the technical scheme of the application, the technical problems that spatial topological information is lost due to a global average pooling GAP operation in iris feature extraction by using an existing lightweight convolutional neural network, and the feature discrimination is reduced due to network lightweight are solved.

This invention discloses a continuous vector discretization representation method, system, and application, comprising: constructing an end-to-end unified architecture including an attention encoder, a binary spherical quantization module, and an attention decoder; utilizing the attention encoder combined with a block causal masking mechanism to uniformly extract high-dimensional feature vectors from single-frame images or multi-frame videos; performing dimensionality reduction, spherical normalization, and binary quantization based on a learnable hyperplane on the high-dimensional feature vectors through the BSQ module, mapping the features to a unit hypersphere and generating binary discrete codes without an explicit codebook; and using the attention decoder combined with spatiotemporal position coding to perform high-fidelity reconstruction of the discrete codes. This invention solves the problems of low parameter efficiency, incompatibility with image and video processing, imbalance between reconstruction quality and computational efficiency, and unstable training caused by the reliance on explicit codebooks in existing technologies. It achieves lightweight, high-fidelity, wide compatibility, and easy training, making it particularly suitable for efficient storage, transmission, and accurate reconstruction of multimodal medical images.

This application belongs to the interdisciplinary field of artificial intelligence, computer vision, and network information security. It discloses a deepfake detection method and system based on orthogonal subspace decomposition and hyperspherical metric. By applying singular value decomposition to the weight matrix of a pre-trained visual model, it explicitly constructs a frozen principal subspace that preserves general semantic knowledge and a trainable orthogonal residual subspace that captures specific forgery traces, achieving orthogonal isolation of the parameter space. Simultaneously, hyperspherical metric learning is introduced into the feature space, performing L2 normalization on the features and applying alignment and uniformity losses. Combined with spherical linear interpolation, latent space data augmentation is performed while preserving the Riemannian geometric structure. Through the synergistic constraints of the parameter and feature spaces, this application can reduce the interference of fine-tuning on pre-trained general visual knowledge and improve the feature discrimination stability and cross-forgery generalization ability in deepfake detection tasks.

The invention belongs to the technical field of aero-engine gas path fault diagnosis, and discloses a gas path fault hyper-spherical clustering method based on angle interval constraint, comprising the following steps: S1, acquiring a gas path fault data sample and a fault-free data sample; s2, extracting features in the gas circuit fault data sample and the fault-free data sample, and obtaining sample features; s3, calculating a characteristic residual error according to the sample characteristics, and calculating a residual error difference according to the characteristic residual error; s4, constructing an angle interval constraint loss function, an L2 norm constraint loss function and a class center aggregation function; s5, constructing a total loss function; and S6, inputting the sample features, the feature residuals and the difference of the residuals into a total loss function to obtain a hyperspherical clustering result. The accuracy and robustness of gas path fault diagnosis are improved, and assistance is provided for health management and gas path fault analysis of an engine.

The application discloses a sea surface weak target detection method based on a residual network and a hypersphere constraint. The method can solve the problems of training underfitting or overfitting caused by the fact that the target sample is much smaller than the sea clutter sample under the sea clutter background, and the specific steps include: 1, performing short-time Fourier transform on each piece of data obtained after dividing the radar echo signal sample to obtain a time-frequency spectrum; 2, constructing an anomaly detection network model combined with a support vector data description based on a variational autoencoder of a residual network; 3, inputting the time-frequency spectrum into the anomaly detection network model for end-to-end training; 4, performing online detection of the target based on the trained model; and 5, evaluating the performance of the application from the constructed multi-dimensional index. The application can effectively enhance the separability between the target and the sea clutter, and improve the weak target detection precision under the condition of a real small sample.

This application provides a learning embedding method, apparatus, electronic device, and storage medium for a dual-view knowledge graph. The method includes: performing a global mapping transformation on the entity embedding features of the triples to obtain global entity features; performing a local mapping transformation on the global entity features to obtain hypersphere features, hyperbolic features, and Euclidean features; inputting the hypersphere features, hyperbolic features, and Euclidean features into an attention fusion model to obtain entity fusion features; obtaining a score value for the triples based on the global entity features and the entity fusion features; and updating the relationships between entities in the initial knowledge graph based on the score value to obtain a target knowledge graph. This application can model entity embedding from both global and local perspectives, improving the accuracy of learning.

The invention provides a surge management system of a fluid machine, an early warning method, a model training method, a computer readable storage medium and a computer program product. The surge management system comprises a computing device which is configured to operate a surge early warning model; the surge early warning model comprises a deep support vector convolutional neural network, and the deep support vector convolutional neural network has a deep separable convolutional network structure and is used for mapping input data reflecting the operation state of the fluid machine to a feature space to obtain corresponding feature points; in addition, the surge early warning model predefines a hypersphere in a feature space; and the calculation device is configured to trigger surge early warning when it is judged that the feature points are located outside the hypersphere. The surge early warning model adopted by the system can complete reliable development training under the conditions of scarcity of surge samples and imbalance of data sets, and light weight of a deep learning model is realized.

The application relates to a flight trajectory anomaly detection method based on an LSTM-GBSVDD model. In order to cope with the length variation problem of flight trajectory data, the application uses LSTM to extract key features in a time sequence, converts variable-length trajectory data into fixed-length representation, enables anomaly detection to be carried out in a fixed-length feature space, and improves the universality of the model. On the basis of feature extraction, the SVDD algorithm is introduced to construct a multi-dimensional hypersphere classifier to model normal flight trajectories. Through the model, potential abnormal trajectories can be identified in an unsupervised framework, the dependence on data labels is avoided, and the bottleneck that abnormality cannot be identified in an unsupervised environment is solved. The application first realizes the joint optimization of the LSTM network parameters and the SVDD scoring function, and proposes a gradient-based training method. The method significantly improves the accuracy and calculation efficiency of anomaly detection and improves the detection performance.

The present application discloses a kind of soft rock zone high spectral remote sensing identification method for railway engineering geological survey, comprising: obtaining railway survey area hyperspectral image and constructing data set;Subsample set is constructed by random sampling, and the space division of multiple groups of isolated hyperspheres and residual space is generated based on the distance of pixel and its nearest neighbor;According to the position of pixel falling into each space division, generate isolated kernel feature vector;With all or part of pixel as background pixel set, calculate the average value of its isolated kernel feature vector as isolated distribution kernel mean model;The inner product of the isolated kernel feature vector of the pixel to be measured and the mean model is calculated to obtain the initial abnormal similarity, and the contiguous area is extracted as soft rock zone after spatial neighborhood regularization.The present application reduces the computational complexity from quadratic to linear, has data density adaptive ability, and can efficiently and accurately identify soft rock zone under complex geological background along railway by fusing spatial continuity constraint.

The invention discloses a high-voltage cable partial discharge fault accurate positioning method and system based on feature space topology mapping. The method comprises the following steps: synchronously acquiring three-phase high-frequency current signals through a high-frequency current sensor array; filtering and de-noising the signal; extracting a four-dimensional feature vector formed by the amplitude ratio, the polarity combination, the phase difference and the energy distribution; mapping the feature vector to a four-dimensional hyperspherical feature space and constructing a topological manifold; calculating a surrounding number as a topology invariant, and establishing a bijection mapping relation between the surrounding number and a physical position to realize accurate positioning; when the matching degree is insufficient, starting an auxiliary criterion based on time delay and attenuation characteristics to carry out secondary distinguishing; and the fault position is visualized through three-dimensional projection. According to the invention, the high-precision positioning of the partial discharge position of the high-voltage cable can be realized, and the robustness and diagnosis efficiency in a complex scene are effectively improved.

The invention discloses a nickel-based superalloy component design and optimization method based on quantum machine learning, and belongs to the field of material technologies and machine algorithms, and the method comprises the following steps: firstly, constructing a data set containing nickel-based superalloy components, oxidation environment conditions and oxidation resistance results, and executing preprocessing operation; based on the preprocessed data and a quantum computing framework, designing a quantum feature mapping circuit, extracting high-dimensional nonlinear features of alloy components and environmental parameters, obtaining a mixed kernel function, and performing training and verification; and carrying out optimization design on the alloy components by adopting a multi-objective genetic algorithm for the verified prediction model, and introducing a hyper-volume for evaluation in order to quantitatively evaluate the overall effect of multi-objective optimization. According to the method, the quantum feature enhancement technology is introduced to be combined with the improved multi-target genetic algorithm, so that the complex nonlinear relation is more effectively captured, and a better alloy design scheme is obtained.

This application relates to a text classification method, apparatus, computer device, and storage medium. The method includes: inputting sample texts into a text multi-classification model to be trained to obtain text feature vectors; for each text binary classifier in the text multi-classification model, treating sample texts of the target class as positive samples and sample texts of non-target classes as negative samples, and normalizing the classifier weight vector and the text feature vectors of the positive and negative samples to a unit hypersphere; determining an angular feature loss function; substituting the angle between the normalized text feature vectors of the positive and negative samples and the classifier weight vector into the angular feature loss function to obtain a sub-loss value corresponding to the text binary classifier; determining a target loss value based on the sub-loss values ​​corresponding to each text binary classifier to be trained, and iteratively optimizing the training in the direction of minimizing the target loss value to obtain the trained text multi-classification model. This method can improve the classification accuracy of the model.

The application discloses a CRM sales dialogue intelligent analysis method based on natural language processing. The method constructs a non-Euclidean emotional space, uses hyperspherical embedding to map the emotional state of the customer into a point in a high-dimensional space, further extracts the potential features of the customer's emotion through a graph neural network, and generates personalized sales dialogue by combining a multi-layer attention mechanism and a conditional decoding structure. The method can dynamically identify the emotional changes of the customer and optimize the dialogue generation in real time according to the emotional state of the customer, improving the adaptability and accuracy of the sales dialogue. Through deep reinforcement learning optimization pruning strategy, the application can effectively improve the computing efficiency, reduce redundant data processing and improve the system response speed.

The application provides an electrical appliance operation anomaly unsupervised detection method and system for a virtual power plant, and relates to the field of equipment operation state monitoring. The method comprises the following steps: collecting operation power sequences and time stamp sequences for preprocessing, splicing to form a feature vector set. A single classification model is constructed, the samples are mapped to a feature space through a Gaussian radial basis kernel function, and a parameterized hypersphere model describing normal data distribution is trained. The Lagrange dual problem of the single classification model is solved to obtain a hypersphere center vector, a radius parameter, and a Lagrange multiplier set. Real-time operation data of a target electrical appliance are collected to generate real-time feature vectors, and the judgment distance between the real-time feature vectors and the hypersphere center is calculated. It is determined that the target electrical appliance is in an abnormal state, the abnormal state equipment is removed from the resource scheduling pool of the virtual power plant, and a health warning information is sent. The technical problems of relying on scarce labeled abnormal data, high training cost and poor generalization ability in the prior art are solved.

The application relates to a multi-source domain industrial fault diagnosis method based on generalized zero sample learning and belongs to the field of industrial fault diagnosis. The method realizes fault diagnosis of source domains and target domains through joint learning of multi-classifiers and domain alignment consistency, and constructs a latent hypersphere space with orthogonal constraints for connecting a feature space and a semantic attribute space, so that discriminative information is extracted and visible and unseen fault diagnosis is realized. The application realizes efficient and accurate diagnosis of visible and unseen faults in the case of missing unseen samples, and improves the generalization performance and practicability of a model.

The invention provides an actuator configuration optimization objective function design and sampling verification method considering optimization index uncertainty. The method is suitable for vibration control of a spacecraft truss structure. The traditional method does not consider the uncertainty of the control force or the initial state, so that the optimization result may deviate from the global optimum. According to the method, an optimization index is expressed as a quadratic form containing an uncertain vector, the expectation and variance of the optimization index are deduced, a robust objective function is constructed, and the optimal actuator layout is solved by adopting a genetic algorithm. Furthermore, a large number of unit vectors are generated through uniform sampling on a hyperspherical surface, statistical distribution of optimization indexes is calculated, and correctness and effectiveness of an objective function are verified. Examples show that the method is superior to the prior art in the aspect of optimizing index distribution, and has good universality and engineering applicability.

The present application provides a prototype evidence sequence regression data processing method and system based on geometric ordering manifold learning, comprising: constructing a unit hypersphere feature space; based on the unit hypersphere feature space, constructing a learnable prototype vector; calculating the Euclidean distance between the feature vector and the learnable prototype vector to obtain an evidence vector; mapping the evidence vector to the concentration parameter of the Dirichlet distribution, combining the Dempster-Shafer theory reasoning framework to obtain a prototype evidence network model, and outputting a prediction result; based on the prediction result, introducing a ordinal manifold constraint based on the bulldozer distance, combining the Bayesian evidence loss to construct a loss function; based on the trained prototype evidence network model, processing new input data to obtain a final prediction result. The present application not only realizes accurate prediction of ordered data and geometric interpretability of feature space, but also has automatic rejection ability for out-of-distribution samples, significantly improving the robustness of the model in long-tail distribution and high-risk scenarios.

The present application discloses a kind of soft rock zone high spectral remote sensing identification method for railway engineering geological survey, comprising: obtaining railway survey area hyperspectral image and constructing data set;Subsample set is constructed by random sampling, and the space division of multiple groups of isolated hyperspheres and residual space is generated based on the distance of pixel and its nearest neighbor;According to the position of pixel falling into each space division, generate isolated kernel feature vector;With all or part of pixel as background pixel set, calculate the average value of its isolated kernel feature vector as isolated distribution kernel mean model;The inner product of the isolated kernel feature vector of the pixel to be measured and the mean model is calculated to obtain the initial abnormal similarity, and the contiguous area is extracted as soft rock zone after spatial neighborhood regularization.The present application reduces the computational complexity from quadratic to linear, has data density adaptive ability, and can efficiently and accurately identify soft rock zone under complex geological background along railway by fusing spatial continuity constraint.

The invention relates to the technical field of graph neural networks, in particular to a community structure constrained unsupervised graph node representation learning method, which comprises the following steps: firstly, initializing node representation on a unit hypersphere, and obtaining node embedding through graph neural network coding; and constructing a loss function of a fused community structure, wherein the community dispersion loss enhances the discrimination between communities through interval punishment, and the node stability loss keeps representation consistency through historical memory and community center fusion. The node memory representation is updated by adopting an index moving average mechanism, and the training stability is remarkably improved. According to the method, the over-smoothing problem in a traditional method is effectively solved, the discrimination and robustness of node representation are remarkably improved while the calculation efficiency is guaranteed, and the method is suitable for unsupervised representation learning of large-scale graph data.

The invention provides an airspace interference suppression method and device based on support vector data description and a medium, and belongs to the technical field of airspace interference suppression. Processing the received signal by using a self-adaptive spatial filtering algorithm to obtain an output signal representation based on the self-adaptive spatial filtering algorithm; solving real number frequency domain representation; the target function is minimized under the constraint condition, a new constraint is obtained by setting a variable partial derivative as zero, a Lagrange multiplier is obtained by optimizing the target function, and the reciprocal of the Euclidean distance between a signal sample in the low-dimensional feature space and the center of the hypersphere is solved to represent a numerical solution; and determining an output signal of the optimal adaptive spatial filtering algorithm. According to the method, the SINR of the output signal is obviously improved, and the interference suppression performance is obviously enhanced. According to the method, the problem of fuzziness of spatial domain filtering algorithm convergence in a spatial domain high-dynamic scene is solved.

This invention discloses a method and system for detecting solder joints on PCB circuit boards based on image recognition, belonging to the field of image processing technology. The method first acquires and preprocesses 3D point clouds of PCB solder joints, and calculates the local geometric probability density. Non-uniform downsampling is performed using density inverse weighting, prioritizing the retention of sparse key features such as solder joint pins and removing redundant substrate data, outputting a set of physical feature points. This set is then input into a feature extraction network, and strong correlations in feature dimensions are eliminated through local and global forging mapping technology to construct isotropic hypersphere spatial feature vectors. Finally, a multi-coverage discrimination model containing defective hyperellipsoids and normal hyperspheres is constructed based on local density and spatial features. The solder joint quality is comprehensively judged using an integrated voting strategy. This invention effectively solves the problems of false detection and missed detection caused by sparse feature loss and sample imbalance, and significantly improves the detection accuracy of minute defects.

The present application relates to the technical field of trajectory representation task processing, in particular to a trajectory similarity measurement black-box adversarial attack method and system based on model decision, which trains a trajectory representation model by using a trajectory data set, and takes the trained trajectory representation model as a target model to be attacked; obtains node vulnerability under a fixed disturbance intensity and an optimal disturbance direction under single-point disturbance based on the target model and single-point vulnerability scanning, and constructs a single-point disturbance sample; aggregates the single-point disturbance sample to a high-dimensional hypersphere based on the node vulnerability and the optimal disturbance direction, determines a trajectory iteration direction according to the gradient of model decision in a limited iteration space, and obtains and outputs an optimal target trajectory adversarial sample. The present application finds gradually smaller adversarial disturbances around the target trajectory, which makes the adversarial trajectory sample satisfy the premise of being very close to the target trajectory in the data domain, and makes the target model output an erroneous similarity measurement result, so as to test the adversarial vulnerability of the trajectory similarity measurement model.

The invention discloses a strong generalization classification method combining boundary feature generation and hyperspherical constraint, and the method comprises the steps: training a prototype network through employing known model samples in a library, and learning a prototype for each class, so as to initialize a feature space; the depth features extracted by the feature extraction network are sorted according to the distance between the depth features and the prototype, feature points close to a classification boundary are screened, directional disturbance is applied, and difficult-to-classify sample features located near a decision boundary are generated; hyper-spherical constraint is applied to original features and generated difficult-to-classify sample features, the constraint explicitly limits similar features in a compact area in a hyper-sphere, the compactness of intra-class features and effective control over inter-class intervals are guaranteed in the aspect of geometric structure, an accommodating space is provided for potential unknown similar special-shaped target features, and the classification accuracy is improved. The generalization ability of the model to unknown similar special-shaped targets is improved; and carrying out collaborative optimization on the feature extraction network and the learnable prototype in combination with prototype comparison loss, and completing classification for unknown similar special-shaped targets.

The invention discloses a CRM sales talk skill intelligent analysis method based on natural language processing. According to the method, a non-Euclidean emotional space is constructed, emotional states of customers are mapped into points in a high-dimensional space by adopting hyperspherical embedding, potential features of customer emotions are further extracted through a graph neural network, and a personalized sales verbal skill is generated in combination with a multi-layer attention mechanism and a condition decoding structure. The method can dynamically recognize the emotional change of the customer, optimizes the verbal skill generation in real time according to the emotional state of the customer, and improves the adaptability and accuracy of the sales verbal skill. The pruning strategy is optimized through deep reinforcement learning, the calculation efficiency can be effectively improved, redundant data processing is reduced, and the system response speed is increased.

The present application relates to the technical field of log anomaly detection, in particular to a log anomaly detection method based on structure perception heterogeneous graph Transformer. The present application adopts Drain algorithm to parse original logs, extracts log events and component information, and constructs a directed weighted log heterogeneous graph containing node types, edge types and edge weights; uses a pre-trained language model BERT to perform context perception semantic coding on each node; constructs a structure perception heterogeneous graph Transformer network, dynamically modulates the heterogeneous mutual attention calculation and message passing process through a structure anomaly perception factor; adopts a joint loss function composed of hyperspherical minimum volume constraint and orthogonal regularization for model training, and uses the Euclidean distance between the graph-level deep representation and the feature space center as the anomaly score to complete detection. The present application can effectively improve the accuracy, robustness and generalization performance of log anomaly detection.