212 results about "Feature description" patented technology

The invention discloses a robot navigation and positioning method based on machine vision. The robot navigation and positioning method comprises the following steps: acquiring and preprocessing an environment image sequence; performing feature point detection and feature description by using an improved DELF model to generate a visual feature point set; obtaining a visual pose estimation result; obtaining fusion pose information; constructing a local environment map by adopting a sparse point cloud reconstruction method; generating a global environment map; generating an optimal advancing path by using a path planning method; and a robot motion control instruction is generated, path adjustment and navigation correction are performed, and high-precision autonomous positioning and navigation of the robot in a complex environment are realized.

The invention discloses an intelligent system failure analysis method based on fault recognition, and relates to the technical field of intelligent system reliability analysis, and the analysis method comprises the specific steps: employing Modelica to construct a multi-physical domain digital twinborn model, combing historical data, and building a structured fault mode library; developing a fault injection algorithm to acquire data to form a standardized data set; preprocessing data to construct a causal link and a probability flow network; carding task scenes to construct a reliability evaluation model and training the reliability evaluation model; inputting actual parameters to trace faults, generating an optimization plan, and outputting an analysis report and a plan after verification; according to the method, the problem of insufficient fault correlation analysis caused by modeling scale splitting in a traditional method is solved by constructing a multi-physical domain digital twinborn model covering microcosmic, mesoscopic and macroscopic and establishing a structured fault mode library of a standardized feature description system containing three types of faults of hardware, software and environment interaction.

The invention relates to the technical field of thunder and lightning mode recognition, and discloses a thunder and lightning intelligent target early warning method and system for a local key target, and the method comprises the steps: collecting target feature description, estimating storm motion information based on lightning positioning data and radar echo images at continuous moments, and carrying out the recognition of a thunder and lightning mode. And generating a risk corridor in the upstream direction of each target area, and forming a target area around the target area in combination with the risk corridor. Unified spatio-temporal data is obtained, spatio-temporal feature extraction is carried out to obtain storm features, cross-modal correlation selection and weighting are carried out on the storm features with target feature description as a drive, and target features are constructed. And generating an early warning result in a predetermined time window according to the target feature. And when the thunder risk probability reaches an early warning threshold and the spatial risk distribution intersects with the buffer range of the target area, generating targeted early warning information. According to the invention, thunder and lightning targeted early warning is realized, the hit rate is improved, and false alarms are reduced.

The application relates to a power image model knowledge migration method and system based on predicted divergence confrontation, which comprises the following steps: performing feature description on an original power image through a visual model, performing hidden space sampling on the feature description of the original power image through a feature extractor, and generating an initial latent vector; inputting the initial latent vector into a generator to obtain a reconstructed power image, and generating an adversarial power image through gradient symbol method-based adversarial disturbance; inputting the adversarial power image into a target model and a substitute model respectively, and calculating a predicted divergence loss; training the substitute model to learn the knowledge of the target model through the predicted divergence loss; and when the performance of the substitute model on a verification set is stable and close to that of the target model, completing knowledge migration. Compared with the prior art, the application significantly reduces data labeling cost, improves the adaptability of knowledge migration in a resource-limited scene, and improves the learning efficiency of a substitute model for key features of a target model and the decision boundary exploration ability.

The application relates to the technical field of image target detection and discloses a limited data labeling-based image 3D target detection model training enhancement method, which obtains image 3D feature descriptions, target features of different feature layers are used to obtain different detection results through a student / teacher 3D target detection network, the target features of different feature layers are used to train a supervised image 3D target detection network through a self-distillation module and different detection results through a supervised learning module, a network training loss function is obtained based on the training of the supervised image 3D target detection network, the network training loss function is used for training the image 3D target detection network through back propagation, different training image 3D target detection networks are adopted, and then in the face of the need for robust monocular 3D target detection neural network model development, limited artificial labeling data is fully utilized to effectively train the network, and a monocular 3D target detection deep neural network model with stronger performance and more robust performance is ensured.

This invention discloses a point cloud registration method and system based on statistical local feature description and matching, relating to the field of point cloud registration. The method includes: determining the local feature descriptor of each query point and the local feature descriptors of its corresponding neighboring points in 3D point cloud data; statistically weighting the local feature descriptor of each query point and the local feature descriptors of its corresponding neighboring points to obtain the statistical local feature descriptor of the corresponding query point; acquiring the statistical local feature descriptors of the source point cloud and the target point cloud; and improving the ICP algorithm based on the feature differences between point pairs in the source point cloud and the target point cloud and the average matching distance to perform point cloud registration. This invention can improve the accuracy and robustness of point cloud registration.

The invention discloses a low-illumination image enhancement method based on wavelet driving, and aims to solve the problems of insufficient local and global feature interaction, single frequency domain feature description view angle and lack of prior feature guidance in the existing low-illumination image enhancement technology. According to the method, a progressive frequency domain multi-view feature collaborative sensing network (WALLIE) is constructed, U-Net is taken as a basic framework, and the method comprises the following steps: decomposing input image features into multi-view frequency domain features through wavelet transform; a cooperative space-wavelet domain multi-view detail compensation (S2WD) module is used to realize spatial domain and wavelet domain feature cooperative compensation, and the local detail recovery capability is improved. According to the method, key indexes such as the PSNR and the SSIM on a plurality of public data sets are superior to those of an existing advanced method, the generalization ability is excellent, the advanced visual task performance such as downstream low-light semantic segmentation and target detection can be effectively improved, and the method has wide application prospects.

The invention discloses a multi-source time sequence image analysis method and system based on cell growth fusion evolution, and the method comprises the steps: S1, obtaining multi-source time sequence image data, and constructing a time sequence image sequence under a unified time axis; s2, the two-dimensional image at each moment is divided into m space patches, multi-dimensional response representation is constructed, and a time sequence feature description set is formed; s3, constructing a multi-layer confidence evolution model, introducing growth confidence, fusion confidence, uncertainty confidence, an evolution stability factor and a global evolution potential function, and performing joint modeling on a co-evolution relationship of a growth state and a fusion state in a time dimension; and S4, constructing a multi-target joint optimization function for fusion degree evaluation, carrying out weighted optimization on various losses, and outputting fusion degree estimation values, fusion stage judgment results and structure evolution description information corresponding to the multi-source time sequence images at the time nodes. The method has the advantages of high analysis stability, high applicability and the like.

The application discloses a crack feature extraction and quantification method and device based on a point cloud large language model, and relates to the technical field of civil engineering structure health monitoring. The method comprises the following steps: according to a two-dimensional image data set and a three-dimensional point cloud data set, performing cross-modal semantic analysis by using a crack feature analysis large model to obtain a crack semantic recognition result; performing skeleton line fitting by using a curve fitting method according to the crack semantic recognition result and the three-dimensional point cloud data set; performing crack feature quantification according to a crack local point cloud subset data set and a three-dimensional skeleton curve of the crack to obtain a crack structured feature; and performing time sequence comparison reasoning by using the crack feature analysis large model according to the crack structured feature and a feature description text based on a crack trend inquiry text to obtain a crack development trend prediction result. The application is a crack feature extraction and quantification method with high precision and strong explainability by combining three-dimensional point cloud data processing and a large language model.

The application relates to a threat feature uniform scanning engine scanning promotion method of application layer protection capability, and belongs to the field of cloud security network communication.The method comprises the following steps: establishing respective feature description information storage data structures of each protection capability of the application layer, and generating a feature library; marking all feature rule IDs of the protection capability, and establishing a comprehensive feature engine library; performing application layer service protection processing; the service traffic is scanned through the comprehensive feature engine library, and all hit features are returned; the feature rule ID is reversely processed into features of respective protection capabilities, feature description information storage data structures of the respective protection capabilities are searched, hit feature information is obtained, log information is sent, the execution action of the features is processed, and it is judged whether the service traffic is prohibited or released. The method has efficient detection processing, so that the next-generation firewall under the cloud has the abilities of rich protection functions and efficient service detection.

The application is a registration method based on super voxel segmentation and bidirectional nearest neighbor distance ratio matching, which overcomes the problem that the existing three-dimensional point cloud registration method cannot meet the higher requirements of target point cloud obtained by sensors such as laser radar on registration accuracy and registration speed. The application uses super voxel segmentation to accurately extract target feature points with stable structure, eliminates the interference of drift noise voxels, and uses bidirectional nearest neighbor distance ratio to improve registration accuracy. The application comprises the following steps: step one, using a super voxel segmentation algorithm to extract target feature points with stable structure; step two, using point cloud thickness layering to perform non-iterative threshold denoising while super voxel segmentation is performed; step three, using FPFH to perform feature description, and using a bidirectional nearest neighbor distance ratio method to perform initial registration on the point cloud; and step four, using a point cloud accurate registration method based on a double threshold to perform point cloud registration.

The present application relates to the technical field of unmanned aerial vehicle photoelectric load vibration control, and particularly relates to a multi-frequency band vibration compensation control method for an unmanned aerial vehicle airborne photoelectric platform based on a disturbance observer, comprising: obtaining unmanned aerial vehicle task phase information, synchronously collecting piezoelectric sensing signals and fiber grating sensing signals, performing time-frequency domain feature fusion and generating a composite feature description in combination with the task phase information; performing mode classification on the current disturbance based on the composite feature description and selecting a corresponding compensation algorithm, predicting the disturbance trend according to the task phase information and activating the compensation algorithm in an event-triggered manner by scheduling computing resources; obtaining an optical axis residual error and a deviation between an actual disturbance and a predicted disturbance, modifying time-frequency domain feature fusion parameters and mode classification discrimination boundaries based on the optical axis residual error, and modifying feature library parameters of the task phase information and a trigger threshold of event-triggered scheduling based on the deviation. The present application realizes adaptive and accurate suppression of wide-frequency band disturbances and improves long-term operation reliability.

The invention relates to the technical field of underwater robot control, and discloses an underwater robot control algorithm based on bionic motion. According to the algorithm, environment sensing data of the underwater robot is obtained, an initial motion instruction is deduced through analysis processing, and a basis is provided for initial actions of the robot. Posture monitoring data of the robot are obtained, and whether the initial motion instruction is corrected or not is decided according to the data. When it is judged that correction is needed, water flow visual data around the robot are collected, water flow mode features are extracted from the water flow visual data, feature description values are generated, and then the feature description values are compared with a historical feature library; if a matched feature description value exists in the historical feature library, correcting the initial motion instruction according to a comparison result; and if the historical feature library lacks the same feature description value, the algorithm calculates a similarity index between the feature description value and the historical feature, and corrects the initial motion instruction based on the index.

The invention discloses an archive data file synchronous transmission scheduling method and system, and the method comprises the steps: carrying out the multi-dimensional feature extraction and structure analysis of an archive data file to be synchronized, and generating a structured feature description; dynamically judging the transmission priority category of each file and estimating the network transmission overhead of the file based on the structural feature description; pre-allocating an initial bandwidth weight and a computing resource quota for each channel according to the transmission priority category and the network transmission overhead; in the data transmission process, the throughput performance and the node load of each virtual transmission channel are monitored in real time, and a cross-channel traffic load distribution strategy is dynamically adjusted according to the channel health degree; and based on the dynamically adjusted flow load distribution strategy, coordinating and executing multi-path parallel and sequential synchronous transmission of the archive data in the virtual transmission channel cluster. By utilizing the embodiment of the invention, refined hierarchical management and global resource optimization of archive data transmission can be realized, and the synchronization efficiency and the overall throughput performance of archives are improved.

The invention discloses a feature derivation method and computing equipment, the method comprises feature generation performed by multiple rounds of iteration, the feature generation of any round specifically comprises: determining an initial feature data set and an effect feedback cue word of the current round, the initial feature data set comprising a plurality of initial features, each initial feature comprising a feature description, utilizing a large language model, taking the initial feature data set as input data, generating a plurality of to-be-screened features according to the effect feedback prompt word, screening in the plurality of to-be-screened features according to the sample data set, determining a plurality of derivative features obtained in the current round, taking the plurality of derivative features as initial features, and extracting the initial features; determining an initial feature data set of the next round; and the large language model is utilized to determine the quality evaluation for the generation process according to the generation process of the current round, and the quality evaluation is used as the effect feedback prompt word of the next round, so that the feature generation process of the next round can be optimized according to the quality evaluation, and the quality of the generated new features is improved.

The invention discloses a component selection method and device, equipment, a storage medium and a program product, and the specific technical scheme comprises the steps: obtaining function description information of a component from a knowledge graph corresponding to the component, the function description information being used for representing the function of the component; calculating text similarity between the to-be-replaced components based on the function description information; and taking the components of which the text similarity is greater than a preset threshold value as candidate components. Therefore, the function of the component is determined through the knowledge graph, the text similarity between the component functions is calculated, the candidate component of the to-be-replaced component is finally obtained, the candidate component is used for replacing the to-be-replaced component, and the replacement efficiency of the component can be improved.

The application provides a process micro-fault detection method based on stationary subspace integrated probability slow features, and belongs to the technical field of industrial process fault detection. The technical scheme is as follows: the application is composed of an offline stage and an online stage. The application fuses the stationary information of SSA and the time sequence dynamics of SFA to form complementary feature description, finally realizes fine division of non-stationary multivariate data through integrated learning, realizes effective extraction of micro-fault probability information, and greatly improves the micro-fault detection rate.

The invention belongs to the field of network social group analysis, and provides a multi-expert collaborative network social group description method based on modal dynamic fusion. The method comprises the following steps: firstly, inputting multi-modal original data of a network social group, and respectively capturing multi-modal private feature representations through a modal representation adaptive extraction module; then, deep fusion of cross-modal information is realized by utilizing a bidirectional state space model through a modal complementary information fusion module, and unified group multi-modal fusion representation is generated; thirdly, a KAN multi-expert network architecture is introduced through a multi-expert collaborative prediction module, different experts are adapted to diversified group characteristic modes such as mainstream, small crowds and temporary groups, and dynamic weights are calculated through a gating network to achieve accurate fusion of expert output; and finally, in combination with long-tail boundary perception loss function optimization model training, outputting a multi-dimensional feature description result of the group. According to the method, more comprehensive and accurate group feature representation can be obtained, and the depicting performance of the network social group in a complex scene is improved.

The application discloses a kind of pre-processing method and system for rock fissure infrared image analysis, it is related to infrared image processing technical field, the method includes: the noise level estimation is carried out to the original infrared image of rock fissure obtained, generates noise feature description information;Based on noise feature description information, the original infrared image is carried out adaptive denoising processing, generates denoising image;Based on denoising image, background separation and high-frequency component extraction are carried out, and generate high-frequency feature image;Based on high-frequency feature image, carry out detail enhancement processing, and generate detail enhancement image;Based on detail enhancement image, carry out adaptive brightness correction processing, and generate pre-processing output image.Adaptive noise suppression and edge preservation to rock fissure infrared image are realized, fissure high-frequency detail feature and trend continuity are strengthened, and the contrast of low brightness area is improved, and high-quality data basis is established for downstream detection and segmentation task.

This invention discloses a code generation method and system for a specific programming language in a low-resource domain, belonging to the field of code generation technology. The method includes: generating an intermediate representation and domain feature description of historical code, and constructing a vector library based on the historical code; retrieving the vector library based on the domain feature description of the target task to obtain several historical tasks; and generating code for the target task based on the domain feature description of the target task and the retrieved historical tasks. This invention can accurately generate code for a target domain using a low-resource domain.

The invention relates to the technical field of big data visualization, and provides a chained quality monitoring information visualization method and system based on meteorological observation engineering. The method comprises the following steps: acquiring project quality monitoring data of each key quality control node in meteorological observation project implementation, and constructing a whole-process stage quality monitoring model comprising a node quality feature description unit, an inter-node quality influence transmission path unit and a quality control threshold interval unit according to the project quality monitoring data; performing association analysis on related units of the model in combination with quality control permission information of each participant in the matrix type organization collaborative architecture to generate a chain type collaborative interaction strategy; and finally, converting the engineering quality monitoring data into a multi-dimensional dynamic view containing a quality control node association relationship view, a quality influence transmission path view and a quality index change trend view according to the strategy, and performing abnormal mode identification on the quality index change trend view. And generating graded early warning information containing the early warning grade and the early warning range.

The application provides a kind of image splicing method and system based on depth point line feature, comprising: collecting the image of scene to be spliced, obtains feature extraction network;The image of scene to be spliced is input into feature extraction network, and point feature, line feature and point line feature description are extracted;Get the geometric coplanar feature of the image of scene to be spliced, use geometric coplanar feature and point line feature description to carry out feature matching, obtain image splicing model;Get the grid deformation constraint condition of the image of scene to be spliced, carry out image pixel fusion to grid deformation constraint condition and image splicing model, output splicing image result.The application uses the depth point line feature of the image to be spliced, carries out feature extraction and matching to the image with certain overlap degree, then projects these images onto the selected reference plane based on the feature registration relationship, so that the common parts of different images are accurately aligned, with the characteristics of high alignment accuracy and high restoration degree.

This invention discloses a method for estimating the pose of stacked workpieces based on the combination of 2D images and 3D point clouds. Using depth images as the processing object, the method performs 2D image processing, including background separation, hole filling, noise removal, and connected component analysis, to remove useless background information. The 2D pixel coordinates of each stacked workpiece are extracted, and point cloud data for each workpiece is generated using mapping relationships. The quality of the generated data is improved through downsampling, filtering, and surface smoothing. The model point cloud is obtained by multi-angle stitching. Key points are extracted from both the model and workpiece point clouds to establish FPFH feature descriptions. An improved Kuhn-Munkres algorithm is used to pair the feature points of both, and clustering is performed based on geometric constraints. The precise pose of the workpiece is obtained using SVD and ICP nearest point iteration. This method solves the problems of poor accuracy in 2D image detection and slow 3D point cloud registration, enabling stacked workpiece recognition to achieve good accuracy and real-time performance without requiring a large amount of training data.

The present application relates to the technical fields of supply chain logistics and big data analysis, and discloses a supply chain logistics demand prediction method and system based on a neural network, which comprises the following steps: obtaining supply chain node data, constructing a weighted directed graph to generate a network topology map; extracting upstream and downstream influence features to obtain a node dynamic interaction vector; calculating the influence weight of interaction frequency on demand fluctuation, marking a key path and determining a traffic pressure distribution value; fusing multi-dimensional data to generate a comprehensive feature description vector; correcting a demand time series to correct prediction bias; extracting seasonal rules and node correlations to generate demand prediction adjustment parameters; and using the adjustment parameters to correct new data, time series extrapolation to obtain high-precision demand prediction results. The present application realizes accurate prediction of the whole process of supply chain logistics demand and effectively solves the problem of low prediction accuracy in the prior art.

The invention relates to the technical field of kernel code generation, and discloses an automatic reconstruction kernel code generation method, which comprises the following steps: collecting example code data of a kernel part, and carrying out structured processing on example codes to obtain a logic transfer diagram of the example codes; constructing an example code feature extraction model to perform embedded representation on the logic transfer diagram of the example code, and converting an embedded representation result into structural features of the example code; semantic features of the function description text are extracted; and receiving the kernel demand description text by using the automatic reconstruction model of the depth kernel code obtained by optimization solution, and reconstructing the kernel code to generate the kernel code meeting the kernel demand description text. According to the method, feature extraction is carried out on structural features representing a logic sequence and a logic structure of an example code and semantic features representing semantic information of a function description text, and the example code similar to semantic of a kernel demand description text is selected to carry out code structure reconstruction and kernel code generation.

The invention relates to the technical field of e-commerce and cloud computing, and discloses an e-commerce order management method based on cloud data analysis. The method comprises the following steps: clustering cloud order time sequence data to form a plurality of differentiated data clusters; performing multi-level feature extraction on each cluster to generate a cluster-level feature description vector; analyzing the vector by using a trained time sequence mode matching model, and identifying a potential processing delay mode; according to the identified mode, actively triggering a resource pre-allocation strategy and executing a corresponding instruction; and finally, completing order transfer state judgment in combination with the deployed system operation data and the delay mode. According to the method, the features are extracted from the order group and the historical delay mode is matched, so that the processing delay risk is predicted in advance and the resources are dynamically pre-adjusted, and the processing efficiency and the operation stability of the order system are improved.

This invention discloses a rotation- and scale-invariant feature extraction method and system based on dual-coordinate system collaboration, belonging to the field of image processing technology. Addressing the poor matching performance of existing image feature extraction methods under rotation and scale changes, this invention employs the following scheme: The input image undergoes a logarithmic polar coordinate transformation to generate a polar coordinate image; a dual-branch network is constructed, where the feature extraction branch extracts keypoint location information in a Cartesian coordinate system, and the feature description branch extracts a polar coordinate feature map in a logarithmic polar coordinate system; the keypoint locations in the Cartesian coordinate system are mapped to their corresponding positions in the polar coordinate feature map using a coordinate mapping function, and interpolation sampling is performed to generate feature descriptors that integrate precise location information and rotation / scale invariance information; based on the feature descriptors, matching point pairs are calculated, and the spatial transformation relationship between images is determined. This invention achieves highly robust feature extraction under rotation and scale changes while maintaining high real-time processing efficiency.

Provided are systems and methods and computer-implemented systems for generating synthetic data and optimizing performance of large language models, by integrating semi-automatic synthetic data generation, prompt optimization and continual learning into a unified system, without full fine-tuning of the model, to improve performance of the large language model. The cold start problem of the large language model is solved through a multi-stage process, including: generating a diversified parameter set and natural language sentences; constructing synthetic user queries using feature descriptions, parameter sentences and constraint instructions; bootstrapping with a small amount of samples with the help of a compiler, and iteratively evaluating instruction-example combinations to optimize the large language model module; and combining real-time data and synthetic data to complete continual validation and optimization.

The invention discloses a multi-view fusion cloud native malicious software detection method and system based on high-speed introspection of a virtual machine, and the method comprises the steps: 1), deploying a high-speed introspection module in a monitoring layer of the virtual machine, capturing an API call sequence of an internal process of a target virtual machine from the outside in a safe and low-invasion manner, and structuring the API call sequence into a runtime log; (2) the API calling sequence is regarded as a sentence, and a Word2Vec model is used for training to generate a structure embedding vector of the API; then, constructing a directed heterogeneous graph containing a file, a thread and API calling for each sample, taking the structure embedded vector as an initial feature of an API node, encoding the graph by using a graph attention network, and extracting a structure context feature vector; 3) extracting an API official function description text by utilizing the pre-training language model to generate a semantic embedding vector; constructing a directed heterogeneous graph for each sample, replacing the initial features of the API nodes in the graph with the semantic embedding vector, coding the graph by using the graph attention network again, and extracting a functional semantic feature vector; 4) firstly performing function classification on the APIs, and performing dimensionality reduction on the complete API calling sequence to obtain a limited function state sequence; constructing a Markov transition probability matrix for the state sequence of each sample, and selectively stacking a multi-order transition matrix to form a multi-channel feature tensor; inputting the feature tensor into a convolutional neural network for coding, and extracting a macroscopic behavior evolution feature vector; 5) splicing the structure context feature vector, the function semantic feature vector and the behavior evolution feature vector to form a final comprehensive feature vector; and inputting the comprehensive feature vector into a multi-layer perceptron classifier, and training the classifier in an end-to-end manner to enable an output sample of the classifier to be a classification result of malicious software or benign software.

The invention relates to the technical field of model construction, and provides a method and system for reconstructing and displaying a billion-pixel high-definition image, and the method comprises the steps: obtaining image data containing visible light, infrared light and depth information through a multi-modal image collection device, carrying out the feature point extraction and preprocessing of the image data, and obtaining a feature point of the image data; obtaining feature description information for correction; generating corrected image data according to the feature description information; obtaining a brightness balanced image based on the corrected image data; according to the brightness balanced image, a seam optimization and detail enhancement processing model is adopted to repair a splicing seam and enhance local details, and an optimized image is obtained; based on the optimized image, constructing a dynamic block rendering strategy, calculating priority information of each block, and generating a block rendering control instruction; and according to a block rendering control instruction, carrying out real-time block rendering and interactive optimization on the optimized image, and then carrying out adaptive display. According to the invention, the image reconstruction and display effect can be improved.