288 results about "Extraction algorithm" patented technology

The invention discloses a VLM model intelligent decision-making-based driving method and device and a storage medium, and relates to the technical field of data processing, and the method comprises the steps: extracting key visual features from continuous multi-frame driving scene images based on a preset visual feature extraction algorithm; inputting a user instruction and a time sequence visual Token corresponding to the key visual features into a preset VLM model for multi-modal alignment, and generating a target planning Token; inputting the target planning Token and the time sequence vision Token into a preset trajectory generation model to obtain a predicted trajectory; and converting the predicted trajectory into a control signal, and controlling the mobile device to complete a moving action based on the control signal. The problem of modal difference between a semantic space and an action space is solved.

The invention provides a multi-unmanned aerial vehicle adaptive migration method for a sudden no-fly zone, and relates to the technical field of unmanned aerial vehicles. And acquiring fusion environment data through cooperative sensing of multiple unmanned aerial vehicles, processing the data by adopting a semantic boundary extraction algorithm based on deep learning, and outputting a sudden no-fly zone map and unmanned aerial vehicle safety state codes. On the basis, the unmanned aerial vehicles are shunted according to the safety state codes and the sudden no-fly zone map, real-time obstacle avoidance tracks are generated for the non-intruding unmanned aerial vehicles, and cluster collaborative separation is executed for the intruding unmanned aerial vehicles. For unmanned aerial vehicles whose tasks are interrupted due to obstacle avoidance or separation, task decomposition is realized based on the sudden no-fly zone map and space-time isolation, and continuous execution of cluster tasks is realized through an energy consumption constraint auction mechanism, so that a plurality of unmanned aerial vehicles can better cope with a complex and changeable sudden no-fly zone environment.

The invention discloses a self-adaptive multi-level digital watermarking method based on coding process optimization. The method comprises the following steps: firstly, performing deep preprocessing on an input video stream through a pre-trained multi-modal deep learning model, extracting space complexity, time dynamics and content significance features, and generating a uniform feature vector; and constructing a dynamic decision model based on the feature vectors, adaptively determining the watermark intensity, the embedding position and the type, and realizing accurate matching of watermark parameters and video content characteristics. A multi-level watermark hierarchical embedding mechanism is adopted, a robust invisible watermark, a fragile invisible watermark and a dynamic visible watermark are embedded into a video, and the security is enhanced by combining chaotic encryption or Hash chain encryption. During copyright verification, watermark information of each layer is recovered through an adaptive extraction algorithm, and data verification and infringement traceability are completed in combination with a block chain evidence storage system. According to the method, a full-process copyright protection system of embedding, coding, extraction and verification is constructed.

The invention discloses a bridge safety early warning method and system, and the method comprises the steps: collecting bridge structure response and environment parameter data, and carrying out the preprocessing of the data, and obtaining a distributed multi-source sensing information matrix; constructing a distributed neuron cell automaton network, interacting state information through unit local communication, operating a lightweight feature extraction algorithm, and generating a multi-dimensional feature vector; carrying out distributed modeling by utilizing a pre-trained distributed LSTM-Transform hybrid model, and carrying out parallel calculation on a local prediction result by each unit to obtain a global health state evaluation result; a local gradient change anomaly detection mechanism is established, data anomaly is identified by comparing prediction results of adjacent units, and a potential risk area is positioned by combining a related algorithm; and constructing a hierarchical early warning threshold system, activating a corresponding early warning response by means of local decision logic, executing an alarm operation, generating recovery guidance information and monitoring uploaded data. According to the invention, two types of network models are combined, so that the system reliability and early warning accuracy can be remarkably improved.

The invention discloses an electronic document tracing method and device based on dynamic encryption and multi-modal watermarking, and relates to the technical field of information security and digital rights management. The method comprises the following steps: acquiring a PDF document uploaded by a user, encrypting the PDF document by adopting a randomly generated main encryption key, and acquiring and storing the encrypted PDF document; the content of the encrypted PDF document is analyzed, secret information is embedded into a picture in the document based on a two-dimensional discrete cosine transform algorithm, and picture watermark embedding is completed; a self-adaptive watermark embedding algorithm is adopted to embed secret information into a text in the document, and text watermark embedding is completed; related information of the checking operation is written into a block chain database; dynamically extracting watermark information from the divulged PDF document by adopting a watermark extraction algorithm; and querying a traceability database according to the extracted watermark information, and outputting a traceability result by comparing the extracted watermark information with information in a database storing various watermark information. According to the invention, the survival rate and the concealment of the watermark can be improved.

The invention provides a target positioning method and system based on a distributed networking radar system, and belongs to the technical field of radars, and the method comprises the steps: controlling a plurality of radar nodes to transmit stepping linear frequency modulation continuous wave pulse signals, and collecting target echoes; carrying out band-pass filtering, pulse compression and fast time synchronization processing on the echoes, and generating interference-free data through a moving target extraction algorithm; realizing cross-node clock synchronization based on a moving target behavior cognition model and a template matching algorithm; constructing a multi-node distance estimation equation set, generating an accurate distance value through calibration broadband synthesis and FFT, and determining a target three-dimensional coordinate; and generating a motion track and speed prediction through Kalman filtering based on the target position sequence. The system comprises a signal emission and acquisition module, a signal processing module, a clock synchronization module, a distance estimation and coordinate fusion module and a trajectory prediction module. The method supports multi-target tracking, improves positioning precision and time synchronization precision, has few errors, and is compatible with any node deployment form.

The invention relates to a lightweight acoustic feature extraction method for end-side audio and video quality inspection. The method comprises the following steps: end-side equipment separates audio and video stream data through a double-time-sequence anchor point alignment method, and acquires independent audio data; performing grading preprocessing on the independent audio data to obtain noise-reduced independent audio data; on the basis of the segmented audio, a time domain and frequency domain collaborative extraction algorithm is adopted, multi-dimensional time domain features and frequency features sensitive to audio quality difference are obtained, and a lightweight acoustic feature vector is obtained through an incremental principal component analysis method; based on the lightweight acoustic feature vector, utilizing a dual-threshold matching judgment mode to obtain similarity between features; and on the basis of the feature similarity, through feature dimension deviation verification, obtaining a quality inspection result containing a standard grade and feature anomaly positioning. Lightweight acoustic feature extraction of end side audio and video quality inspection is realized.

The invention discloses a watermark embedding and watermark tracing method and device for digital information, equipment and a medium, and relates to the technical field of digital watermarking, and the method comprises the steps: generating an initial watermark based on the Redis atom incremental operation when carrying out the watermark embedding of a plurality of initial sensitive digital information, and segmenting the initial watermark to obtain a plurality of initial sub-watermarks; randomly selecting an initial sub-watermark as a target watermark, and respectively embedding the target watermark into each piece of initial sensitive digital information to obtain each piece of target sensitive digital information containing the watermark; and when watermark tracing is carried out on a plurality of pieces of target sensitive digital information, the to-be-confirmed watermarks corresponding to the target sensitive digital information are analyzed based on a watermark extraction algorithm, the target watermarks are determined from the to-be-confirmed watermarks, and corresponding watermark tracing results are generated. According to the method, protection of sensitive digital information is realized through a watermark embedding and tracing scheme which is high in concealment and robustness and is adaptive to multiple types of short data.

The invention discloses a BIM (Building Information Modeling)-based pile support foundation pit whole-process safety management system, relates to the technical field of foundation pit safety management, and aims to solve the technical problem of frequent false alarm caused by insufficient construction interference identification capability of a fixed threshold alarm mechanism. And a decision intervention module. According to the method, the multi-source information atlas is constructed, and construction activities, equipment states, environmental factors and monitoring data are dynamically associated. When the data is abnormal, the root cause correlation analysis unit is started, and the intensity and duration characteristics of the abnormal signal are calculated by using a time sequence characteristic extraction algorithm and are compared with the historical pattern library. The mechanism can intelligently judge whether data abnormity is derived from construction interference such as substantive soil body / structure instability or heavy equipment operation, the alarm accuracy is remarkably improved, and the problem of frequent false alarm caused by insufficient construction interference identification capability of a fixed threshold value alarm mechanism is solved.

The invention discloses a multi-model collaborative two-dimensional Gaussian splash three-dimensional reconstruction method, which comprises the following steps: acquiring a multi-view RGB image, outputting camera parameters and sparse point clouds by using a VGGT model, and generating a depth image and a mask image through a monocular depth estimation network and an image segmentation network; carrying out densification processing by utilizing a visual shell algorithm; initializing a two-dimensional Gaussian primitive by using the dense point cloud; carrying out training on the initialized two-dimensional Gaussian primitives; re-rendering depth maps of all visual angles, and generating a three-dimensional grid through a truncated symbol distance function and a contour surface extraction algorithm; performing grid optimization based on local bilateral filtering on the three-dimensional grid to obtain a smooth grid; and carrying out scale correction and smoothing the grid to obtain a three-dimensional model with a real scale. According to the method, an omnibearing three-dimensional model with high-fidelity geometry, real scale and new view rendering capability is generated through a multi-model collaborative initialization and optimization strategy.

The invention discloses an optical detection system and method, and belongs to the technical field of optical non-contact measurement. According to the invention, a low-coherence wide-spectrum light source is adopted, a dispersion element and an array detector are combined to separate a wide spectrum in space so as to construct a plurality of parallel independent wavelength measurement channels, and meanwhile, an envelope extraction algorithm based on empirical mode decomposition, Hilbert transform and window length adaptive Savitzky-Golay filtering is integrated; real-time quality evaluation and dynamic weighted fusion are carried out on each channel signal, vibration measurement continuity and stability in a severe speckle environment are remarkably improved, high-dimension wavelength diversity is realized at low cost, real-time quality evaluation and weighted fusion are carried out through a self-adaptive envelope extraction algorithm, and the accuracy of vibration measurement is improved. And finally, a vibration measurement value of which the signal-to-noise ratio and the measurement precision are remarkably improved is output.

The invention discloses a method for accessing electric power terminal equipment to an electric power trusted wireless local area network, which relates to the technical field of electric power communication and comprises the following steps: acquiring known connected electric power terminal equipment of an electric power business platform, and constructing a metadata portrait according to historical operation preference parameters; through a fingerprint extraction algorithm, collecting equipment operation metadata in real time, and constructing a power terminal request instruction sequence; obtaining a cloud object model and the request instruction sequence to perform association matching screening, and packaging into an adaptation protocol standard data frame of the power terminal request instruction; establishing a multi-protocol monitoring interface power access model, and verifying whether the real-time power terminal request instruction has request abnormity or not; if it is judged that connection establishment is allowed, a dynamic communication behavior monitoring model is established according to the metadata portrait of the known power terminal equipment and the adaptation protocol standard data frame of the power terminal request instruction, and whether the power terminal request instruction is legal or not is analyzed. The method has the beneficial effect that the stability of the power system is enhanced.

The application discloses a fisheye image correction method and system based on a geometric model and an effective area extraction algorithm, and relates to the field of image processing.The method comprises the following steps: scanning pixel points from the center of a picture outward, judging the edge condition of a fisheye image where the current row and column are located according to the scanning result, adaptively adjusting the scanning step length, and further shortening the scanning time; and the ratio of the width and height of the fisheye image to the whole picture can be used to distinguish the light spot and the fisheye image, so that the accurate fisheye image radius can be obtained in real time in different devices and various measured pictures, and the image correction based on the geometric model is carried out by using the parameter, so that the accuracy of the extracted fisheye image is ensured, and the fisheye effective area can be extracted in different devices and various measured pictures.

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

The invention discloses a power plant wastewater treatment intelligent control method based on digital twinning, and particularly relates to the field of prediction.The power plant wastewater treatment intelligent control method comprises the steps that S1, a whole-process digital twinning body mapping a power plant wastewater treatment physical system is constructed, and multi-dimensional core parameters are collected in real time; s3, training a BP neural network model optimized by a genetic algorithm to realize prediction of water quality parameters; s4, solving a multi-objective optimization problem through an NSGA-III algorithm, and generating an optimal process parameter formula to support a decision; s6, dynamically iterating the digital twin and the prediction model based on the actual operation data, and continuously improving the prediction and optimization reliability; according to the method, through construction of the whole-process digital twinborn body and the improved LSTM time sequence feature extraction algorithm, accurate prediction of the water quality change of the power plant wastewater is realized, the problem of prediction lag in the prior art is solved, sufficient data support is provided for process adaptation, the risk that effluent exceeds the standard is effectively reduced, and stable operation of a treatment system is guaranteed.

The invention is applicable to the technical field of computer aided design, and provides an automatic tolerance labeling method based on a multi-modal large model, which comprises the following operation steps: receiving an STP file of an assembly as input, analyzing the structure of the assembly, calculating geometric information of parts and feature surfaces, distributing a unique identifier and establishing an assembly tree; and through shell part identification and a centroid-based iterative RANSAC profile extraction algorithm, generating a multi-view overall image and a profile image. According to the automatic tolerance labeling method based on the multi-modal large model, the manual labeling time is greatly shortened by automatically extracting geometric information and generating part description and tolerance labeling, and through guidance of a rule base and an international standard, it is guaranteed that tolerance labeling is standard and unified, the risk of human errors is reduced, and the accuracy of the manual labeling is improved. The multi-modal large model is combined with geometric data and natural language description, comprehensive understanding of part functions and assembly relations is achieved, the reasonability of tolerance marking is improved, and the method is suitable for multiple parts.

The invention discloses a path planning method and system for autonomous operation of an underground carry-scraper. The method comprises the following steps: acquiring map data information of an autonomous operation area of a target underground carry-scraper; using a skeleton extraction algorithm to obtain skeleton information of the autonomous operation area; acquiring the shortest reachable path from the set starting point to the set ending point; extracting a curved center point on the shortest reachable path, and sequentially connecting the set starting point, each curved center point and the set ending point to obtain an initial broken line path; interpolating and cutting the initial broken line path to obtain a collision-free broken line from a set starting point to a set ending point; vehicle turning-around judgment is carried out, and intersection detection is carried out; and constructing a V-shaped turn-around path at the intersection and replacing a corner, and finishing path planning of autonomous operation of the underground carry-scraper from a set starting point to a set ending point. According to the method, the path planning for the autonomous operation of the underground carry-scraper is realized, the curvature is continuous, the reliability is higher, and the feasibility and the accuracy are better.

Motion picture film is scanned by high-resolution, continuous sprocketless scanner, producing a first sequence of digital images each representing a plurality of motion picture frames and perforations (perfs) of the film input. The first sequence of images is processed using a processor running an analysis and extraction algorithm, producing a second sequence of images each including a single, edge-stabilized frame of the motion picture.

The invention discloses a groove welding seam processing method, and belongs to the field of machine identification welding seam, the method comprises the following steps: using a line structured light camera to collect a point cloud, converting the point cloud into a three-dimensional coordinate of an actual position, and combining and outputting to obtain a complete point cloud; random downsampling and clustering are carried out on original point cloud data to remove discrete points, and the point cloud direction is unified by rotating the point cloud direction according to the center point of a point cloud bounding box; segmenting the preprocessed point cloud data by using a deep learning algorithm, and dividing the point cloud data into three classes; an edge contour extraction algorithm is adopted for the point clouds on the left side and the right side to obtain a key contour point set, the key contour point set is segmented after moving average filtering, arc fitting is conducted on the segmented point clouds, the left boundary and the right boundary of the weld joint are obtained, the method is suitable for complex and changeable application scenes, and powerful support is provided for welding quality control.

The invention discloses a shielded cable electricity testing method based on ground loop impedance potential lifting, and relates to the technical field of measurement. The electroscope comprises an electricity testing step, an induction electrode on the cambered surface of the electroscope is used for obtaining induced electricity, a broadband direct sampling framework is formed on the physical level, and broadband direct sampling of millivolt-level power frequency displacement current characteristics is achieved. Based on a variational mode decomposition-Duffing oscillator fusion weak signal extraction algorithm, a variational mode decomposition algorithm is utilized to adaptively strip broadband noise and higher harmonics, a reconstructed signal is subjected to scale normalization and then introduced into a critical chaotic state Duffing oscillator algorithm, phase transition of a phase track from a chaotic attractor to a large-scale limit cycle is monitored, and a weak signal is extracted. A non-linear dynamic weak signal extraction algorithm is formed on the algorithm level, and weak power frequency features masked by strong noise are accurately recognized; a broadband direct acquisition framework and a weak signal extraction algorithm form a dual detection framework, and a quasi-induction electric signal is detected and judged in the live part verification process of the shielded cable.

The application relates to the technical field of data processing, in particular to an accelerometer vibration rectification error analysis method, which comprises the following steps: receiving multiple original current signals of accelerometers, processing collected data to output a characteristic point coordinate sequence with confidence; dynamically scheduling an optimization algorithm combination in a knowledge base according to the confidence and frequency band distribution characteristics to calculate a local statistical data set; inputting the data set into a mechanical dynamics model to perform trajectory simulation, simultaneously inputting a control model to reconstruct a signal to generate an error correction coefficient matrix by comparing the differences of double-path positioning drifts; adopting the matrix to reconstruct a vibration rectification error quantitative value to output a knowledge base update instruction and a parameter reset instruction; importing the quantitative value to calculate a theoretical angle deviation and verify convergence, and triggering neural network retraining when the deviation exceeds a threshold value. Through a non-smooth feature extraction algorithm and a closed-loop feedback mechanism, the signal distortion problem caused by the smoothing operation in the prior art is solved.

A method for extracting key contour points of Chinese characters based on key skeleton points and stroke width belongs to the field of computer and Chinese character vectorization, which comprises the following steps: inputting a single Chinese character picture, performing pretreatment to generate a binary image; extracting the skeleton of the Chinese character by using a skeleton extraction algorithm; further extracting key skeleton points of the Chinese character; obtaining the stroke width at each key skeleton point of the Chinese character; obtaining the contour points of the Chinese character by using a contour extraction algorithm; sequentially traversing all contour points of the Chinese character, calculating the distance from the current contour point to the nearest key skeleton point, calculating the distance threshold of the current nearest key skeleton point, judging whether the current contour point is a key contour point by comparing the distance from the current contour point to the nearest key skeleton point with the distance threshold, and recording the current contour point as a key contour point if the distance is less than the distance threshold. The application effectively improves the quality of Chinese character vectorization, maintains all detail features, effectively removes noise contour points and redundant contour points, and reduces the storage space size of the vectorized Chinese characters.

The present application belongs to the technical field of household photovoltaic, and particularly relates to a household photovoltaic fault diagnosis method and system, which comprises: collecting the volt-ampere characteristic curve data of a photovoltaic system and synchronous environmental irradiance; extracting at least one volt-ampere characteristic curve parameter, including short-circuit current, open-circuit voltage, maximum power or fill factor; according to the current irradiance, based on a preset dynamic health baseline, judging whether the system has a fault through a first preset rule. The first preset rule realizes fault diagnosis by obtaining a parameter expected health value and comparing it with an actual extracted value. The present application adopts a dynamic health baseline to adapt to environmental changes, combines an efficient parameter extraction algorithm with an intelligent fault judgment mechanism, significantly improves the diagnosis accuracy, reduces false positives and false negatives, and is suitable for low-cost, automated operation and maintenance of household photovoltaic systems.

The application discloses a lightweight indoor 3D laser ground extraction algorithm method, system and medium. The method comprises the following steps: acquiring indoor 3D laser radar data; obtaining coordinate values based on a laser radar coordinate system according to the indoor 3D laser radar data; sending the coordinate values to a preset plane equation to obtain a ground plane normal vector; sending the ground plane normal vector to a first preset coordinate system to establish a constraint to obtain a comprehensive error value; and accumulating squares of values in the comprehensive error value to obtain a comprehensive error sum of squares; extracting a minimum value in the comprehensive error sum of squares, and setting a coordinate point corresponding to the minimum value as an optimal current pose. The application is suitable for various lasers, and by giving a preset threshold, whether it is 16 lines, 32 lines or 128 lines, the algorithm accuracy and robustness are ensured, and the accuracy of recognized ground points is improved.

The application discloses a kind of nematode body bending behavior identification and counting method and related equipment, the method comprises: obtaining nematode video, the original gray image is obtained by frame processing to nematode video, and binary image is obtained according to the original gray image of nematode;According to the original gray image of nematode, the numerical coordinate regression algorithm based on convolutional neural network obtains head coordinate and tail coordinate;According to the binary image of the nematode, the feature point extraction algorithm based on curvature calculates the peak point on the center line of nematode;The maximum distance of the line between the pharynx and tail of nematode is calculated, and the number of nematode body bending is calculated according to the change of maximum distance of each frame.The numerical coordinate regression algorithm based on convolutional neural network is used for the coordinate identification of nematode head and tail, the accuracy is improved, and the number of nematode body bending is calculated more simply and quickly by extracting feature points.

This invention discloses an AI-interactive AR (Augmented Reality) cultural wall system for cultural tourism, belonging to the field of AR system technology. The surface of the cultural wall is equipped with a pre-embedded feature set, including visual markers, differentiated texture features, 3D depth variation areas, and specifically designed interactive hotspots. These features provide the system with stable visual benchmarks, semantic distinction criteria, depth cues, and precise positioning anchors. The system includes: a feature point acquisition module, used to collect the feature set on the cultural wall through a multimodal feature extraction algorithm to construct a 3D spatial map with semantic annotations; a user positioning and pose estimation module; an AI content generation engine; and a spatial adaptation and rendering module. Through the cultural wall feature set and the multi-module collaborative system, precise spatial integration of virtual content and the physical wall surface is achieved, along with intelligent dynamic content generation based on user behavior, effectively enhancing the immersion, interactivity, and personalization of the AR cultural experience.

The invention discloses a muzzle arc energy and mass center spatio-temporal evolution analysis method and system, and relates to the technical field of muzzle arc analysis, and the method comprises the steps: building an electromagnetic track launch muzzle arc simulation experiment platform, and obtaining a muzzle arc image through a high-speed photographing system; an arc contour image is obtained based on a local adaptive threshold segmentation and morphological gradient extraction algorithm, and the integrity and robustness of arc contour extraction are improved; based on a robust centroid tracking algorithm of multi-connected domain energy fusion and time sequence consistency constraint, a final arc centroid is obtained, and stable positioning and tracking of the arc energy centroid in the full evolution stage are achieved; and quantitatively analyzing the time-domain evolution characteristics of the highlighted area of the muzzle arc and the law of the spatial motion trail of the mass center of the highlighted area based on the muzzle arc images at different muzzle speed. An effective means is provided for quantitative characterization of a bore port arc, and an experimental basis and a theoretical reference are provided for optimization of a bore port structure, anti-ablation design, regulation and control of emission working conditions and the like of an electromagnetic emission system.

The invention relates to the technical field of computer vision, in particular to an end-to-end single target tracking method combining underwater frequency domain scene prompt and light parameter fine tuning. The method comprises the following steps: acquiring a tracking template graph and a search area image through a to-be-tracked video; generating a scene feature map by using a frequency domain enhancement algorithm; recoding the underwater target features by using a feature embedding extraction algorithm; taking the token mark sequence and the scene feature token mark sequence as the input of an underwater scene prompter, respectively calculating the attention scores of the two token mark branches by using a target perception attention algorithm and an auxiliary modal importance evaluation algorithm, and generating new scene features; introducing a cross-layer weight fusion mechanism, performing further feature extraction on the underwater scene in combination with shallow and deep features, and outputting target frame information through a prediction head network; in a network model training process, the performance of executing a specific underwater task is improved by adjusting a partial multi-head attention mechanism algorithm of a trunk vision Transform layer encoder.

The application provides a case text information-based accident logic graph analysis method and system. The method comprises the following steps: based on a template extraction algorithm, performing scene event extraction on a pre-acquired accident report sample to construct an initial accident logic graph K1 of a mine outburst accident; based on a reinforcement learning algorithm, optimizing the accident logic relationship in the initial accident logic graph K1 to obtain an application accident logic graph of the mine outburst accident; and based on a preset layout algorithm, performing layout display on the graph data of the application accident logic graph to generate an accident main graph solution of the mine outburst accident. In this way, the accident graph solution is automatically generated from the accident report sample, the automatically generated accident graph solution can help professional personnel quickly understand the accident causes, improve the reading speed of the accident report, reduce the analysis threshold of non-professional personnel, and improve the efficiency of accident guidance analysis.

This invention relates to a passive multi-station multi-target cross-localization method for removing false points, comprising: determining the direction finding line corresponding to each direction finding station based on the angle of arrival information of each target obtained by each direction finding station; obtaining the intersection set of the direction finding lines of all direction finding stations using bi-station cross-localization; determining high-density points in the intersection set using a high-density point extraction algorithm; clustering the high-density points in the high-density point set into clusters based on the cutoff radius corresponding to all high-density points in the high-density point set; calculating the azimuth angle of the cluster center of each cluster relative to each direction finding station; filtering and merging the clusters in the cluster set based on the calculation results to determine the final cluster set; and obtaining the target localization result based on the average position of each cluster in the final cluster set. This invention is simple and effective, and can complete the target localization using only the angle information measured by multiple stations and its own position.