687 results about "Similarity matrix" patented technology

The invention discloses a multi-stage spatial-temporal clustering method and system based on a fused mahalanobis distance. The method comprises the following steps: acquiring a spatio-temporal data set, and determining a spatio-temporal neighbor relation of samples in the data set; calculating a space communication distance and a time decay distance of the sample; fusing the space communication distance and the time decay distance by using a mahalanobis distance to obtain a relative distance of the sample; selecting a class cluster center from the data set according to the local density of the sample and the relative distance; adopting a multi-stage distribution strategy to distribute non-class-cluster center samples to corresponding class clusters; wherein the multi-stage allocation strategy comprises an inevitable allocation stage based on space-time shared neighbor and a similarity allocation stage based on a weighted similarity matrix. According to the method, the key problems that an existing space-time clustering algorithm is insufficient in space-time attribute coupling processing and sensitive to distribution errors are solved, and the clustering accuracy, robustness and practicability in the fields of intelligent traffic analysis, seismic sequence recognition and the like are remarkably improved.

The invention relates to the technical field of automation and artificial intelligence crossing, in particular to an industrial monitoring and early warning method and system based on a multi-modal large model, and the method comprises the steps: obtaining multi-modal industrial data, carrying out the initial feature processing of the obtained multi-modal industrial data, and carrying out the initial feature processing of the obtained multi-modal industrial data; and performing deep feature extraction fusion based on the extracted preliminary features, taking a multi-modal fusion feature matrix as input to construct a multi-modal large model, performing model training on the constructed multi-modal large model, and performing real-time monitoring and abnormity early warning by using the trained multi-modal large model. According to the method, space-time alignment and feature association are realized through a defect feature matrix of a visual mode, a trajectory similarity matrix of a motion mode and a time sequence reconstruction matrix of a sensing mode and through dynamic time warping and cross-modal parameter sharing; and associated abnormalities of product quality, equipment actions and running states can be captured at the same time, so that the problem that the collaborative analysis capability of a traditional method on multi-source heterogeneous data is insufficient is solved.

The invention discloses a multi-modal offshore wind power ultra-short-term prediction method in the field of offshore wind power plant cluster power prediction, and aims to solve the technical problems of spatial-temporal feature splitting and insufficient dynamic dependency relationship modeling. The method comprises the steps of performing anomaly detection and restoration on fan data, and generating a corrected wind power cluster data set; extracting a mean value, a standard deviation and a latest value of core operation data of each fan through a dynamic time window, and constructing a multi-dimensional node feature; a static geographic similarity matrix is generated based on geographic coordinates, a basic wake effect matrix is generated in combination with real-time wind direction data, correction is carried out through the maximum mutual information quantization time-delay effect, and then a dynamic adjacency matrix is obtained through self-adaptive fusion; and integrating the multi-dimensional node features and the dynamic adjacency matrix into a space-time diagram sequence data architecture, inputting the space-time diagram sequence data architecture into a multi-scale wake flow perception diagram space-time prediction model, and outputting a multi-fan power prediction value. According to the invention, high-precision multi-fan power prediction can be realized.

The invention provides a self-adaptive semantic-driven data set field matching method and system, and the system comprises a data preprocessing module which is used for carrying out the cleaning, standardization and preliminary analysis of an input data set, and extracting a field name, a data type, a field description and a data sample; the deep semantic representation modeling module is used for constructing a field-level semantic representation vector; the multi-level similarity calculation module is used for comprehensively calculating the grammatical similarity, the semantic similarity and the statistical similarity among the fields, dynamically adjusting the weight of the similarity of each level by adopting a weighted fusion algorithm, and generating a comprehensive similarity matrix; and the matching result management and application module is used for generating a field matching mapping table and a fusion suggestion according to the comprehensive similarity matrix. According to the method, high-precision automatic matching of data set fields is realized by fusing deep semantic understanding, multi-dimensional similarity calculation and incremental adaptive learning, and the efficiency and accuracy of data set fusion are remarkably improved.

The invention belongs to the technical field of image information processing, and discloses a scene guidance prompt representation enhanced infrared and visible light image fusion method, which is divided into two stages: a first stage, constructing a scene prompt generation network, and learning global visual semantic information covering a source image through a semantic segmentation task; in order to further enhance the prompt representation capability, a visual perception context prompt module is designed, interaction is performed by using a correlation matrix between modal specific features and text features, and the text features are refined in a dynamic weighting mode, so that scene prompt representation with richer semantics is obtained. In the second stage, a cross-modal alignment fusion network guided by prompt is provided, and infrared and visible light features are mapped to a unified shared embedding space by utilizing learned scene prompt. In the process, a pixel-text similarity matrix is established through a prompt driving feature alignment module, and accurate alignment of cross-modal features is realized, so that a fusion result of semantic consistency and detail fidelity is obtained.

The invention discloses a high and cold arid region slope soil stability safety risk evaluation system and method, and relates to the technical field of slope engineering. The state of each side slope under multi-dimensional indexes such as freeze-thaw cycle frequency, dry-wet alternation index, wind erosion strength, shear strength, water content, porosity and fracture density is quantified, the feature similarity between any two side slopes is calculated, and then a side slope feature similarity matrix is formed. Based on the matrix, an unsupervised clustering algorithm (such as spectral clustering, similarity propagation and the like) can be adopted to divide a plurality of side slopes into similar subsets with structural characteristics similar to environmental response, and category attribution of risks is achieved. On the basis, structural variation analysis, historical instability statistics and central risk difference extraction are performed on similar slope samples, so that the internal instability tendency of the slope can be identified, and a risk prediction model suitable for the type of slope can be constructed through feature training.

The invention relates to a photovoltaic bearing capacity assessment method and system for a distributed photovoltaic transformer area, belongs to the field of intelligent power distribution network and distributed energy access, and solves the problem that the photovoltaic bearing capacity of a low-voltage power distribution transformer area under a distributed photovoltaic high-proportion access scene is difficult to assess in a refined manner in the prior art. Comprising the steps of obtaining a physical adjacency matrix and a behavior similarity matrix of a transformer area based on collected voltage time sequence data of users in a non-photovoltaic output time period and a photovoltaic output time period in the transformer area; fusing the physical adjacency matrix and the behavior similarity matrix to obtain a user clustering result; based on the user clustering result and the observation time sequence data of the multiple operation conditions of the transformer area, obtaining an impedance matrix between transformer area clusters; and based on the clustering result and the impedance matrix, establishing a photovoltaic bearing capacity assessment implicit optimization network which aims at maximizing the total photovoltaic capacity of the transformer area and meets multiple constraint conditions, and solving by adopting an energy function and implicit layer optimization to obtain a photovoltaic bearing capacity assessment result of the transformer area.

Systems, methods, and other embodiments associated with quasi-supervised clustering for activity pattern characterization and anomalous activity detection are described. In one embodiment, a method generates a first sparse similarity matrix for nearest neighbors of a plurality of data points. The data points each characterize a pattern of activity associated with an account. The method generates a second sparse similarity matrix for random neighbors of the plurality of data points. The method recursively clusters the plurality of data points based on the first sparse similarity matrix. The method quasi-supervises the recursive clustering based on the second sparse similarity matrix to stop the iterative clustering when the data points are split into N clusters. The value of N is not pre-determined. The method detects that the individual data point has changed clusters, indicating anomalous activity. And, the method generates an electronic alert that the anomalous activity is associated with the account.

The invention discloses a multi-sensor fusion processing method, perception method and equipment based on LiDAR point cloud pseudo image conversion, a multi-view projection strategy is adopted to convert a preprocessed LiDAR point cloud into an aerial view, a front view and a side view, the aerial view, the front view and the side view are subjected to feature coding and then fused through a SENet attention mechanism to generate a multi-view fusion pseudo image, and a large amount of space information is reserved. Meanwhile, in order to effectively improve the multi-sensor feature fusion efficiency, after a visual image is preprocessed, an improved CNN network is adopted to extract the features of a LiDAR pseudo image and the visual image, feature fusion is achieved through a cross-modal attention mechanism, attention weights are generated by calculating a similarity matrix, and the feature fusion efficiency is improved. And after the features are enhanced, the features are fused in a channel splicing and element-level adding mode. The method can effectively improve the precision and robustness of automatic driving environment perception, enhances the performance in a complex scene, and is suitable for tasks such as target detection and semantic segmentation in automatic driving.

The invention discloses a multi-modal big data intelligent cleaning and fusion method based on a knowledge graph and deep learning, and relates to the technical field of artificial intelligence and multi-modal data processing. According to the method, a cross-modal semantic similarity matrix is constructed through a multi-modal semantic association mining algorithm, the problem that in a traditional method, an effective semantic association mechanism is lacked between modal data is effectively solved, the quality and consistency of the data are remarkably improved, and a hierarchical feature extraction and weighted fusion strategy is further adopted, so that the method is more efficient and efficient. The multi-modal data is subjected to feature extraction by using a convolutional neural network and a Transform model, and the fusion weight of each modal feature is dynamically adjusted through an attention mechanism, so that the information island phenomenon is effectively solved, the quality and effectiveness of the fusion feature are remarkably improved, and through a semantic consistency verification mechanism and a continuous learning updating mechanism, the accuracy of the feature fusion is improved. High quality of fusion features is ensured, the model can adapt to new data in real time, and adaptability and expansibility of the model are remarkably improved.

The invention provides a video abnormal behavior positioning method and device based on a multi-modal large model, and a medium, and relates to the field of computer vision. The method is applied to computer equipment and comprises the following steps: acquiring a to-be-processed video, and determining a plurality of query videos and a plurality of support videos; obtaining a thinking chain text according to the visual language model; obtaining a first robust feature representation and a second robust feature representation according to a semantic time sequence pyramid encoder; obtaining subtitle text fusion features according to a text encoder; and further performing alignment processing to obtain an alignment similarity matrix so as to determine a prediction interval of the query video, determine an abnormal behavior positioning result of the to-be-processed video and determine a dynamic model fine tuning sample. The video abnormal behavior positioning method and device solve the technical problems that in the prior art, visual information is excessively depended on, deep semantic understanding of the action context and the target relation is lacked, misjudgment is easily generated in the alignment process, the positioning precision is insufficient, and the video abnormal behavior positioning efficiency is low.

The invention provides a wind power prediction method based on a frequency domain space-time diagram convolutional network. The method comprises the following steps: firstly, respectively constructing a spatial proximity weight matrix and a similarity matrix based on a fan position and historical feature data, and adaptively fusing the spatial proximity weight matrix and the similarity matrix into an adjacent matrix; performing high-dimensional embedding on the normalized features, and converting a time domain signal to a graph frequency domain by using graph Fourier transform GFT; designing a prediction and reconstruction double-branch network, extracting key frequency features through discrete Fourier transform (DFT), causal convolution and channel attention, and performing inverse transformation reduction; and finally, constructing a joint loss function of prediction and reconstruction loss weighting, and realizing end-to-end training through back propagation and an Adam optimizer. According to the method, spatial topology and time-frequency dynamic characteristics are fused, prediction and reconstruction advantages are combined, and the precision and robustness of wind power prediction are improved.

The invention relates to a neural network anomaly detection method based on multi-scale feature extraction and time-frequency feature alignment. The method comprises the steps of preprocessing time sequence data; performing multi-scale transformation on the data through a preset sliding window strategy, and extracting features of three scales by using two layers of BiLSTM; calculating the similarity of the inter-scale feature subsequences and constructing a similarity matrix; fusion weight calculation and feature fusion of different scales are carried out based on similarity, and weighted fusion and normalization are carried out; the fused features are converted to a frequency domain through short-time Fourier transform, and alignment is carried out; and inputting the aligned data into a neural network decoder to realize anomaly detection of the intelligent operation and maintenance time sequence. According to the method, fine features of time series data can be greatly improved, and the model anomaly detection capability is enhanced; realization is easy and precision is high; the method can be used in the field of time sequence anomaly detection and prediction, and realizes monitoring of the system and industrial equipment.

The invention provides a video anomaly detection method and system based on semantic and amplitude depth collaboration, and the method comprises the steps: extracting a visual feature sequence of a video frame; enhancing the feature vector of the corresponding time step in the visual feature sequence by using the amplitude and energy of each time step to obtain an enhanced feature sequence; calculating a semantic similarity matrix of the enhanced feature sequence; calculating a sparse affinity matrix according to the amplitude of each time step in the visual feature sequence; fusion is carried out to obtain a final attention weight; and performing weighted aggregation on the value vector of the enhanced feature sequence by using the final attention weight, and inputting the value vector into a classifier to obtain a frame-level anomaly score so as to realize video anomaly detection. According to the method, through one-time parameter-free feature engineering, deep collaboration of semantics and amplitude information is forcibly realized on a model architecture level, and the problem that an existing weak supervision video anomaly detection method is insensitive to feature amplitudes is solved.

The invention belongs to the field of electroencephalogram signal classification detection, and particularly relates to a neuropsychiatric disease electroencephalogram diagnosis method based on iteration polar coordinate attention, which comprises the following steps: acquiring a multi-channel EEG (electroencephalogram) signal and preprocessing the multi-channel EEG signal; inputting the EEG preprocessing signal into a pre-trained improved LaBraM model to obtain a plurality of node features; calculating a Pearson's correlation coefficient matrix and a cosine similarity matrix according to the EEG preprocessing signal, and then constructing a brain function fusion connection matrix as an adjacent matrix through an iteration polar coordinate attention mechanism; according to the method, the large-scale pre-trained EEG model and polar coordinate attention are used for brain graph structure construction for the first time, and collaborative optimization of time feature generalization and space structure modeling capacity is achieved.

The invention provides a multivariable time series prediction method and system based on structural representation learning, and belongs to the technical field of multivariable time series prediction.The method comprises the steps that a similarity matrix between variables is constructed according to multivariable time series data, and variable community tags are obtained through community detection; carrying out weighted binarization to obtain a binary community sensing adjacency matrix; inputting the initial variable feature matrix and the adjacent matrix into a graph auto-encoder to obtain a variable structure embedded matrix; variable value embedding, time step position embedding and structure embedding are fused to obtain a unified input matrix, a time sequence prediction model based on a structure information guiding attention mechanism is input for modeling, and a prediction result is obtained; wherein structure embedding serves as an auxiliary feature to participate in modeling, and explicit structure constraint is not applied. According to the method, variable association is learned and mined through structural representation, and the accuracy and generalization ability of multivariable time sequence prediction are improved.

The invention discloses a multi-view clustering method based on tensor feature extraction, and the method comprises the steps: inputting a multi-view data matrix, constructing a similarity matrix of each view through a K-NN algorithm and a Gaussian kernel function, and carrying out the spectral clustering to obtain a sample embedding matrix; performing singular value decomposition on an original data matrix of each view, taking first c left singular vectors to construct a feature embedding matrix, applying 2, 1 norm group sparse constraint on the feature embedding matrix, and connecting a sample embedding matrix through a bigraph to extract features; and normalizing the sample embedded matrix, and reconstructing a block diagonal matrix into a third-order tensor. And integrating the sample embedding matrix, the feature embedding matrix and global tensor learning to construct a target function, and optimizing through an alternating direction multiplier method until convergence. And finally, the normalized samples are embedded into the matrix to form block diagonals to form a consistent similarity graph, and a clustering result is obtained by using an N-Cut or k-means algorithm.

The invention discloses a Beidou-based road slope disaster early warning method, and relates to the technical field of road slopes. The method comprises the following steps: determining and arranging Beidou receiving nodes in key subareas of a road slope for data acquisition to obtain road slope node data; performing displacement difference calculation on the three-dimensional coordinates of the road slope node data, and dividing key sub-region categories; performing feature extraction on the road slope node data to obtain a road slope node feature vector, constructing a road slope node similarity matrix, and based on the road slope node similarity matrix, obtaining a high-consistency track cluster through a density clustering method and contour coefficient screening; carrying out dimensionality reduction on road slope node feature vectors of high-consistency track clusters in the key sub-region categories by utilizing a principal component analysis method, and fitting track trends to obtain evolution indexes; euclidean distance is calculated through historical landslide event trajectory library indexes, a risk value is generated, and when the risk value exceeds a threshold value, early warning is performed and a risk level is established.

The invention provides a personalized federal learning method and system for multi-device heterogeneous data, and the method comprises the steps: transmitting a neural network structure to all clients, so as to enable all clients to carry out local model training; receiving the trained model parameters of each client, and dividing the trained model parameters of each client into non-BN layer parameters and BN layer parameters; all the non-BN layer parameters are aggregated; calculating distribution similarity among the clients according to all the BN layer parameters to obtain a similarity matrix; clustering the clients according to the similarity matrix by adopting an affinity propagation algorithm to obtain a client group with similar feature distribution; performing intra-group aggregation on the BN layer parameters corresponding to each group of clients; sending the aggregated non-BN layer parameters to all the clients, and sending the aggregated BN layer parameters in the groups to the clients in the corresponding groups; therefore, the training stability and the prediction precision in a multi-device heterogeneous environment are improved.

Systems, methods, and other embodiments associated with quasi-supervised clustering for activity pattern characterization and anomalous activity detection are described. In one embodiment, a method generates a first sparse similarity matrix for nearest neighbors of a plurality of data points. The data points each characterize a pattern of activity associated with an account. The method generates a second sparse similarity matrix for random neighbors of the plurality of data points. The method recursively clusters the plurality of data points based on the first sparse similarity matrix. The method quasi-supervises the recursive clustering based on the second sparse similarity matrix to stop the iterative clustering when the data points are split into N clusters. The value of N is not pre-determined. The method detects that the individual data point has changed clusters, indicating anomalous activity. And, the method generates an electronic alert that the anomalous activity is associated with the account.

The invention discloses a sugar net image grading method and system based on multi-modal decoupling knowledge distillation, and belongs to the technical field of image processing and artificial intelligence. The method constructs a teacher-student architecture: a teacher model generates an ordered DR language prototype and calculates an image-text similarity matrix through a ViT image encoder and a text encoder with a Prompt module; according to the student model, image features are extracted by using light-weight EfficientNet-B0. In the training stage, a decoupling distillation mechanism is designed, target class KL loss, non-target class KL loss and grade sorting loss generated by Prompt are jointly optimized, and migration of multi-modal ordered knowledge is achieved; in the reasoning stage, only student models are deployed for efficient grading. According to the method, the problems of boundary fuzziness, long-tail distribution and sequential modeling in DR classification are solved, lightweight deployment is realized while the classification precision is improved, and the method is suitable for clinical edge equipment.

The invention belongs to the technical field of computer big data, and discloses a relation graph construction and layout method, device and system based on spectral clustering and a storage medium, a clustering center is initialized through a genetic algorithm, the clustering center serves as genetic information and is coded into a character string, the operation time can be shortened, and the classification precision can be improved; furthermore, a weighted Euclidean distance is constructed as a distance function of a K-means algorithm, mutual relation weighting between the features can be reflected, features of different weights are counted into the distance, the classification precision can be effectively improved, the loss is reduced, and the classification efficiency is improved. According to the method, an initial similarity matrix, obtained through a traditional similarity calculation method, between XML documents is corrected through an affinity propagation algorithm, the similarity between the hidden similar XML documents can be reflected, on the basis, the correct clustering number and the correct clustering result are obtained by applying a multi-path spectral clustering method NJW, the method is irrelevant to the sequence of the XML documents, and the method has the advantages of being high in practicability and easy to popularize. The method is suitable for clustering the retrieval results of the XML documents arranged in any sequence.

The invention discloses a flight flow intelligent prediction and air traffic collaborative optimization system based on big data, and relates to the technical field of air traffic management. Comprising the steps that a data acquisition module acquires multi-source data from a civil aviation database and aligns the multi-source data to generate a fusion data matrix; the traffic prediction module extracts traffic feature vectors, generates a feature similarity matrix by calculating distribution differences, and generates a basic traffic prediction result based on migration prediction model parameters; the disturbance correction module detects a sudden disturbance event based on real-time weather and airspace state data and corrects a basic prediction result; the capacity adjusting module generates a capacity adjusting scheme when the capacity deviation value exceeds a threshold value according to the correction result and the airport operation capability data; and the collaborative optimization module finally generates a multi-airport collaborative scheduling scheme through a game equilibrium algorithm based on the correction result, the capacity scheme and the regional coordination data. According to the invention, the accuracy of flight flow prediction and the efficiency of multi-airport collaborative management are effectively improved.

The application relates to the technical field of signal processing, data representation and classification, and discloses a radar signal spectrum clustering sorting method based on SOM anchor point extraction and graph fusion, which comprises the following steps: configuring a radar signal sorting cluster number and a radar pulse parameter, obtaining a normalized radar pulse data set, taking each normalized radar pulse in the radar pulse data set as a node, and constructing a KNN graph of the radar pulse; extracting an anchor point of the normalized radar pulse data set based on SOM, calculating the similarity between the extracted SOM anchor point and all nodes, obtaining a similarity matrix, and constructing an anchor graph adjacency matrix based on the similarity matrix, namely an adaptive anchor graph; weightedly fusing the KNN graph and the adaptive anchor graph to obtain a fusion graph; and performing spectrum clustering sorting based on the radar signal sorting cluster number and the fusion graph to obtain a sorting result. The application breaks through the limitation that a classical radar sorting clustering algorithm can only utilize distance information and density information, and improves the sorting performance under complex radar pulse distribution conditions.

The invention relates to the technical field of bioinformatics, in particular to a method for typing acute respiratory distress syndrome biological subtypes. The method comprises the following steps: a) acquiring multi-omics data and carrying out standardized preprocessing; the multi-omics data comprises transcriptomics data, proteomics data and metabonomics data of a biological sample source; b) constructing a similarity network of each group by using a similarity fusion network (SNF), and obtaining a uniform sample similarity matrix through multi-group network fusion and iteration; multiple collaborative principal component analysis (MCIA) is adopted to carry out dimension reduction on multi-omics data so as to realize visualization of a clustering result; carrying out multi-omics joint discrimination modeling under the guidance of SNF clustering by using a data integration analysis (DIABLO) method so as to identify key feature variables; and carrying out biological subtype classification based on the clustering result of the steps.

The invention relates to an unmanned aerial vehicle image automatic splicing method for fault photovoltaic module positioning, and the method comprises the steps: extracting a feature point of a photovoltaic slate point edge in an image through employing a SuperPoint deep learning model, and precisely positioning a significant position which can be used for matching in the image; constructing a key point association graph through a SuperGlue algorithm, calculating a feature similarity matrix in combination with an attention mechanism, and establishing a corresponding relation of feature points among different images; solving an optimal feature matching pair by applying a Sinkhorn algorithm to obtain an optimal feature point corresponding relation between adjacent images, and providing accurate matching data for geometric transformation; according to the method, a SuperPoint self-supervision framework is utilized, through a VGG encoder and a double-branch decoder, the photovoltaic panel regular texture region feature point extraction capability is improved, uniform distribution is realized through non-maximum suppression, and the problems of sparse feature points and non-uniform distribution in a traditional algorithm are solved. And a SuperGlue graph neural network is introduced, and a similarity matrix with a direction weight is constructed in combination with a self-crossover attention mechanism, so that the accuracy of the edge feature points of the photovoltaic panel is improved.

The invention relates to the technical field of highway tunnel lining disease evaluation, and discloses a highway tunnel lining disease dynamic clustering evaluation system, which comprises a data acquisition module used for acquiring multi-dimensional disease data of a tunnel lining and preprocessing the data so as to construct a disease dynamic matrix. And the data analysis module is connected with the data acquisition module, establishes a fuzzy similar matrix among the disease space units based on the disease dynamic matrix, generates a closed similar matrix in combination with a transitive closure algorithm, and further analyzes the closed similar matrix through a dynamic clustering algorithm to obtain a disease clustering result. And the data output module is connected with the data analysis module, constructs a comprehensive evaluation model according to the clustering result and the historical maintenance data, and finally determines and outputs the risk score and the maintenance priority of the tunnel lining disease. According to the invention, tunnel disease identification precision and maintenance resource configuration efficiency are improved, and safe operation of the tunnel is guaranteed.

The invention discloses a dialogue control method and system based on an artificial intelligence dialogue large model, and the method comprises the steps: receiving the dialogue content of a user, extracting features, and constructing an initial dialogue feature set; calling a dialogue dual processing model to generate intermediate processing features; determining a core feature vector through a topic tracking mechanism, updating a topic feature library, and calculating a similarity matrix; starting off-question detection based on the similarity matrix, and marking potential off-question content; processing off-question content in combination with dialogue control parameter intervention, and generating a subject regression prompt; and generating and outputting a target response, and storing dialogue related data. The system comprises a dialogue content processing unit, a feature construction unit, a dual processing unit, a theme tracking and off-question detection unit, a control intervention unit and a response generation output unit. According to the method and the system, through multi-mechanism cooperation, dialogue coherence and theme consistency are improved, and accurate control of the dialogue is realized.

The invention discloses a loopback detection method based on semantic segmentation and image blocking, and the method comprises the following steps: S1, generating a semantic tag graph based on a current frame image and a corresponding depth image, and extracting a depth mutation edge point set; s2, mapping the mutation edge point set to a semantic tag graph, performing boundary segmentation, and generating a structure segmentation semantic graph; s3, combining moment constrained maximum entropy modeling and information entropy distribution estimation, using a DBSCAN method to cluster semantic regions, and constructing a semantic instance set; s4, extracting static category features, constructing a static semantic relevancy matrix, and generating a static semantic vector; s5, inputting the static semantic vector into a cross attention network for alignment, constructing a semantic similarity matrix, and extracting an optimal matching path; and S6, calculating a similarity score, and outputting a loopback detection result. According to the method, high-robustness loopback detection of an image level can be realized, and the matching precision and the judgment reliability in a dynamic environment are effectively improved.

The invention discloses a serialization-based point cloud over-segmentation method, device and equipment and a medium, and relates to the technical field of point cloud over-segmentation. The point cloud over-segmentation method comprises the following steps: obtaining point cloud data set preprocessing; serializing the preprocessed point cloud data into a Hilbert curve; dividing the serialized point cloud data into a plurality of initial segments based on spatial continuity, and obtaining a multi-scale segment set; the feature similarity between adjacent segments is calculated, and a similarity matrix is obtained and established to provide a basis for subsequent super-point clustering; carrying out super-point clustering on the initial segments based on a self-adaptive updating algorithm; updating super-point features through a cross attention mechanism, establishing dynamic association between points and super-points, and obtaining a new super-point structure; and according to the updated super-point features, constructing a super-point graph, performing enhancement processing through a graph convolutional network, fusing features from the backbone point cloud network and corresponding multi-level super-point features, then transmitting the fused features to a segmentation head for segmentation, and obtaining semantic segmentation output.