1389 results about "Histogram" patented technology

The invention relates to a hydraulic engineering concrete crack intelligent identification and quantitative analysis method and system based on machine vision. The method comprises the following steps: firstly, acquiring an original image sequence of the surface of a hydraulic engineering concrete structure, classifying according to illumination intensity, shooting angle and shooting distance, extracting crack edge features through a convolutional neural network, and fusing to obtain a crack feature set; correcting illumination through adaptive histogram equalization, correcting angles and distances through geometric transformation, and combining edge detection and scale invariant feature transformation to obtain standardized geometric parameters including crack length, maximum width and the like; if the parameter exceeds the engineering safety standard threshold value, tracking a crack track through an optical flow method to calculate increment, and inputting a neural network to output a damage trend; and finally, generating a three-color risk distribution diagram by using a finite element based on the trend, extracting high-risk data to calculate a real-time evaluation value, and dynamically adjusting the monitoring frequency to generate an optimization strategy. By adopting the method, the reliability and economy of engineering safety monitoring can be remarkably improved.

The invention discloses a mobile terminal streetscape image real-time segmentation method based on a lightweight neural network, and relates to the technical field of image segmentation. The method comprises the following steps: firstly, carrying out 320 * 320 adjustment, Z-score standardization, adaptive histogram equalization and 3 * 3 Gaussian filtering preprocessing on an input streetscape image; then, an improved MobileNetV3 backbone network is used, and a five-scale feature map is output in combination with DropBlock regularization through eight feature extraction stages including depth separable convolution and an SE attention module; multi-scale features are fused through a U-shaped structure, and a fusion feature map is generated through up-sampling, element-by-element addition of dimension reduction low-layer features and an attention gating module; and during reasoning, outputting a segmentation mask by using a convolutional layer, Softmax and a conditional random field, and finally performing knowledge distillation, weight pruning, 8-bit quantization and TensorRT optimization. According to the invention, high-precision real-time street view segmentation is realized, the robustness is high, and the method is suitable for different devices and scenes.

Systems and Methods are disclosed for detecting acute coronary syndrome (ACS) events, arrythmias, heart rate abnormalities, medication problems such as non-compliance or ineffective amount or type of medication, and demand / supply related cardiac ischemia. The system may have both implanted and external components that communicate with a Physicians's programmer, and smart-devices for monitoring and alerting to detected medically relevant events or states. At least one processor provides event detection using statistical threshold criteria calculated upon at least a portion of a patient's data / distributions and set for a patient or based upon what a doctor determines as abnormal for a patient. Cardiovascular condition is tracked using histogram, trend, and summary information related to heart rate and / or cardiac features such as S-T segment measures of heartbeats. Heartbeats with elevated rates, and below a “high” range, provide medically relevant detections including medication non-compliance. Novel methods of power management and patient monitoring are disclosed.

A method comprising: receiving an image block unit of a frame, the image block unit comprising samples in color channels comprising at least one chrominance channel and one luminance channel; reconstructing samples of said luminance channel of the image block unit; determining a reference area for predicting target samples of at least one color channel of the image block unit, wherein said reference area comprises one or more reference samples in a neighboring block in current color channel / frame, in the neighboring of a co-located block in reference color channel / frame; and / or inside the co-located block in reference color channel / frame; determining weights for predicting said target samples based on a ratio between a normalized luminance histogram in said reference area and a normalized luminance histogram co-locating said target samples; determining filter coefficients of a filter for said predicting based on the weights, the reference samples and a shape of the filter; and predicting samples of at least one color channel of the image block unit based on the samples of said luminance channel and the filter coefficients.

The invention provides a coal impact tendency evaluation method which comprises the following steps: sampling a coal rock material on an engineering site, and processing the sample into a coal rock test piece with a preset size and a preset shape; carrying out a uniaxial compression experiment on the coal rock test piece, and recording a motion image of an ejection body on the coal rock test piece in the loading process of the coal rock test piece; comparing cosine similarities among the plurality of images in the moving images, and determining the image with the maximum damage degree according to the cosine similarities; analyzing gray level distribution of the image with the maximum damage degree to obtain a gray level histogram; identifying particle information of the broken particles in the image with the maximum damage degree; representing the damage degree of the coal rock material by using the particle information of the crushed particles; and evaluating the coal impact tendency by using the damage degree of the coal rock material. According to the evaluation method for the coal impact tendency disclosed by the invention, a new coal impact tendency evaluation index is provided, the research on the damage process of the coal rock material is perfected, and basic parameters are provided for subsequent early warning of damage of the coal rock material.

The invention provides a microscopic automatic focusing method and system based on image gray histogram characteristics, and the method comprises the steps: collecting an image sequence under different focal lengths, carrying out the fuzzy processing, extracting a gray histogram of each frame of image, and calculating the peak position and full width at half maximum of the histogram as the evaluation characteristics of the image definition; calculating the variance of each feature and automatically allocating a weight according to the relative response degree; and finally, comprehensively evaluating the image definition through a weighted definition scoring function, and selecting the focal length corresponding to the image with the optimal score as the optimal focusing position. The method is based on the global features of the gray histogram, is high in anti-noise capability, is adaptive to different imaging scenes through weight adaptive adjustment, is low in calculation complexity, supports real-time focusing, is especially suitable for high-noise fluorescence microscopic imaging scenes, is high in system portability, is low in operation threshold, and effectively improves the accuracy and stability of microscopic automatic focusing.

The invention belongs to the technical field of image processing, and particularly relates to a monitoring image defogging enhancement method in a high-dust environment of a mineral separation site, which comprises the following steps of: performing smooth denoising on an original image by using weighted guided filtering, and obtaining global atmospheric light through a quadtree subdivision method; constructing a same-color heterogeneous discrimination index combining a spectrum similarity factor and a texture confidence factor for distinguishing dust and ore with similar colors; calculating a pixel-level dynamic defogging coefficient based on the discriminant index, and obtaining an adaptive transmissivity in combination with dark channel prior; and finally, restoring the image by using an atmospheric scattering model and carrying out contrast-limited adaptive histogram equalization processing. According to the method, the problem of misjudgment caused by the fact that the colors of the ore and the dust on the ore dressing site are similar is effectively solved, powerful defogging of the dust area and detail reservation of the ore area are achieved, and the definition and the contrast ratio of the monitoring image are improved.

The invention relates to the technical field of deep learning and computer vision, and discloses a histogram interactive attention-based pavement crack segmentation network processing method and system, so as to enhance the edge detail fidelity and improve the crack segmentation precision. The method comprises the steps of image preprocessing, up-sampling, down-sampling, feature fusion and image reconstruction processing. Wherein global feature modeling in the intensity sub-boxes and among the sub-boxes is realized by constructing a histogram interactive attention module (HIA); a double-branch detail enhancement feedforward module (DDEF) is introduced to enhance spatial detail and high-frequency edge information expression; meanwhile, a Fourier jump enhancement module (FFSM) is adopted to jointly refine jump connection features in a spatial domain and a frequency domain. Through the synergistic effect of the modules, the network can realize continuous recovery and structural consistency modeling of a crack boundary in a complex pavement environment, so that the accuracy and the stability of a segmentation result are remarkably improved.

The invention belongs to the technical field of image enhancement, and particularly relates to an image enhancement method for VCSEL (Vertical Cavity Surface Emitting Laser) epitaxial wafer detection, which comprises the following steps of: analyzing local and global gray information and structure tensor characteristics of pixel points, and calculating a defect saliency value of each pixel point to quantify the possibility of the pixel point as a defect; interference of a stripe structure in an epitaxial wafer image is weakened; a weighted histogram is constructed based on the defect saliency value, so that the defect pixel occupies a higher weight in the histogram; and determining an optimal segmentation threshold value by searching an energy median point of the weighted histogram, and finally performing BBHE enhancement on the image by using the segmentation threshold value. According to the method, the contrast ratio of the tiny defects can be remarkably improved, a high-quality image is provided for subsequent epitaxial wafer detection, and the detection accuracy is improved.

The invention discloses a single-point supervised directional target detection method based on adaptive sample allocation and a global-local context enhancement module, which comprises the following steps: inputting an original remote sensing image into a backbone network, extracting to obtain a multi-scale feature map set, inputting a feature map with the highest resolution into a global-local context module to obtain an enhanced feature, and extracting the enhanced feature from the global-local context module; inputting the enhanced features into a projection layer to generate a category probability graph; based on the category probability graph, training and optimizing the category probability graph to obtain a probability histogram set by adopting a sample distribution and optimization strategy combining adaptive positive sample selection, online difficult case mining and a focus loss dynamic weighting mechanism based on prediction confidence; and based on the full training set category probability histogram set, carrying out adaptive pseudo tag threshold calculation, and finally, realizing end-to-end joint training of pseudo tags and enhanced features through a prediction quality guide weighting strategy. According to the method, the target contour precision and the detection robustness are remarkably improved, and high-performance remote sensing image directional target detection is realized under a weak supervision condition.

The present invention relates to the field of privacy computation. Disclosed are a secure multi-party computation method and system based on a decision tree and privacy protection. In a data processing stage, the method uses a gradient-based one-sided sampling algorithm, so as to exclude most small-gradient samples, thus ensuring balance in accuracy while reducing the data volume; then, the method uses a histogram algorithm to construct a decision tree, thus reducing the memory consumption and increasing the computational speed; and in addition, the method also uses a differential privacy technology to perform encryption and perturbation processing on parameters of local models, so as to ensure that individual privacy information is not leaked. The secure multi-party computation method based on a decision tree and privacy protection of the prevent invention can implement prediction tasks such as classification efficiently and reliably while protecting privacy.

The invention discloses an ultra-low illumination-oriented SPAD active imaging data compression method, which comprises the following steps of: performing wavelet decomposition on histogram data of a single pixel in a time-frequency domain based on wavelet transform to obtain low-frequency data of each pixel after wavelet decomposition; by taking each pixel as a center, performing non-maximum suppression and data enhancement on the low-frequency data after the wavelet decomposition of the current pixel by using adjacent pixels to obtain processed compressed data; customizing different Gaussian kernel parameters for each pixel according to the possibility that each pixel is located at the boundary, and performing Gaussian filtering on the processed compressed data of each pixel to obtain filtered data; and performing depth estimation on the filtered data to obtain a final depth image. According to the method, the depth reconstruction performance of the laser pulse can be improved by utilizing the multi-resolution characteristic of wavelet transform, the space-time correlation of signal photons and the smoothness of Gaussian filtering.

The invention relates to the technical field of computer vision, and discloses a jade defect intelligent detection method and system based on machine vision and deep learning, and the method comprises the following steps: S1, based on a high-resolution industrial camera and a laser three-dimensional scanner, adopting a multi-mode synchronous collection strategy, and rotating a jade sample through a precise motion control system, a jade surface high-resolution two-dimensional color image and high-precision three-dimensional point cloud data are respectively obtained, and a jade multi-mode original data set is generated. A high-resolution two-dimensional color image and high-precision three-dimensional point cloud data are integrated through a multi-modal synchronous acquisition strategy, and multi-dimensional feature expression under unified coordinates is constructed, so that the limitation of a single data source is effectively overcome; an image registration algorithm and a feature pyramid network are combined with a point cloud network to perform multi-modal feature fusion, complementarity of color texture and geometric morphology information is enhanced, and image quality is optimized based on adaptive histogram equalization and non-local mean filtering.

The invention discloses a flotation froth image feature extraction method based on adaptive multi-scale weighted fusion, and belongs to the technical field of intelligent monitoring and image processing in the mineral flotation process, and the method comprises the steps: fusing a color histogram, texture and edge features, and depth features extracted by VGGNet, GoogLeNet and ResNet; and mapping and fusion of multi-modal features are realized by using a self-adaptive multi-scale weighted feature fusion network. And dynamic weighting and fusion are carried out on different characteristic modes through a nonlinear neural network model, and unified characteristic representation is generated to be used for accurate representation of the flotation froth state. The method has the advantages of high recognition precision, high robustness and flexible deployment, online monitoring, optimal control and index prediction of the flotation process can be effectively supported, and reliable technical support is provided for an industrial field.

The invention relates to a solid-state laser radar ranging method and system based on neural network processing, and belongs to the technical field of laser radar ranging, and the method comprises the steps: obtaining a laser echo signal when laser radar detection is carried out on a surrounding target object, and obtaining a photon counting histogram based on the laser echo signal; preprocessing the photon counting histogram, and inputting the preprocessed photon counting histogram into a pre-trained neural network model to obtain a peak confidence map containing an echo peak value and a peak time offset map; on the basis of the peak confidence map and the peak time offset map, obtaining one or more echo peak values of which the confidence exceeds a predetermined threshold and time position information of the one or more echo peak values of which the confidence exceeds the predetermined threshold; and based on the time position information, the distance information with one or more target objects is determined, so that the time position information measurement accuracy of one or more echo peak values is improved, and the laser radar ranging accuracy is improved.

The invention provides a real-time processing method for endoscopic image blood vessel enhancement and microcirculation evaluation, which comprises the following steps of: processing an original endoscopic image, and locally enhancing a brightness channel through a contrast-limited adaptive histogram equalization technology; self-adaptive frequency domain-space domain decomposition is realized based on local texture complexity analysis; performing multi-scale Hessian matrix blood vessel detection on the low-frequency component; applying a directional Gabor filter bank to the high-frequency component; spatial-temporal feature fusion is realized through multi-resolution pyramid optical flow calculation; synchronously completing blood vessel probability prediction, blood vessel diameter estimation and blood flow direction prediction by using a lightweight multi-task deep learning network; the blood flow velocity is analyzed and calculated based on a speckle mode, the perfusion density is subjected to accelerated statistics through an integrogram, and vascular morphological parameters are extracted by adopting an improved skeleton algorithm. According to the invention, an enhanced blood vessel visualization effect and a real-time microcirculation quantitative evaluation function can be provided, and the overall improvement of the endoscope image processing quality and efficiency is realized.

A system and method for 3D anomaly detection and tracking are provided that uses multimodal fusion, reduced training data, recursive segmentation and histogram statistic distance. The anomaly may be a defect or a configuration error.

The invention relates to a balanced color perception enhancement method for rail transit target detection, and belongs to the technical field of rail transit and computer vision. According to the method, brightness channel adaptive histogram equalization enhancement CLAHE-LC, a multi-segment tone channel mask mechanism MSHCM-M and a three-stage hybrid mechanism non-maximum suppression TSLSH-NMS technology are combined, so that the problems of complex illumination, multi-target shielding, color interference and the like in a rail transit scene are solved, and the accuracy and recall rate of target detection are improved. The method can be seamlessly integrated into an existing deep learning target detection framework, is compatible with multi-label, multi-category and multi-scale features, and remarkably improves the precision and recall rate of target detection in a rail transit scene. Experimental verification shows that the real-time performance is guaranteed, meanwhile, the detection performance is obviously improved compared with a traditional scheme, and the method is suitable for being applied to an actual rail transit safety monitoring and intelligent maintenance system.

The invention provides an automatic equipment anomaly monitoring method and system based on image processing, and relates to the technical field of automatic equipment anomaly monitoring, and the method comprises the steps: converting an equipment operation video into a multi-dimensional physical field feature: in a motion field dimension, based on optical flow field analysis, extracting a full-period displacement statistical histogram and a space thermodynamic diagram, the motion instability characteristic of the mechanical transmission system is accurately quantified; in a vibration field dimension, an energy spectrum and a vibration thermodynamic diagram are generated innovatively through time-frequency transformation of a displacement signal of a frequency spectrum monitoring point, and frequency domain feature visualization of a hidden vibration fault is achieved; in the dimension of a structure field, edge gradient analysis and texture feature extraction technologies are fused, a time sequence structure thermodynamic diagram sequence is constructed to capture a progressive damage evolution rule, a three-field abnormal index dynamic weighting fusion mechanism overcomes the limitation of traditional single-point monitoring, connected domain analysis of a fused thermodynamic diagram is combined with an LBP texture and morphological feature decision tree, and the defect of the prior art is overcome. Automatic equipment abnormity monitoring based on image processing is realized.

The invention discloses an SLAM front-end optimization method based on brightness grading and gradient constraint, and relates to the technical field of computer vision and vision SLAM. In order to solve the defects that adaptive grading enhancement and parameterization control based on illumination types are lacked in the prior art, and feature point quality, spatial distribution uniformity and front-end real-time performance are difficult to consider in a high-frequency input scene, the invention provides a comprehensive brightness grading and feature optimization scheme. The method comprises the following steps: firstly, calculating the average brightness of an input image, dividing the image into a dark light type, a normal type and an overexposure type, and respectively adopting gamma correction, contrast limited adaptive histogram equalization and inversion enhancement strategies for different types to realize illumination adaptive enhancement; then screening high-quality feature points with significant local curvature changes; and updating the detection area. According to the method, the feature stability, the matching precision and the real-time performance of the SLAM front end in a complex indoor environment are remarkably improved, and the method is suitable for a self-localization and mapping system.

The invention relates to the technical field of image segmentation, in particular to an anorectal focus automatic segmentation method based on deep learning, which comprises the following steps: acquiring an anorectal image pixel map, extracting contrast and direction offset to mark candidate focus points, screening overlapped marks to generate a focus activation mark map, and establishing a response map to generate a boundary response distribution map. And training the network to output a classification graph, and extracting a truncation path to complete image segmentation. According to the invention, through extracting the contrast value and the gradient amplitude of the local gray level co-occurrence matrix, accurate capturing of the spatial difference of the lesion area under a complex background is realized, through constructing a response map and direction consistency comparison mechanism and combining multi-dimensional features such as a direction gradient histogram and a structure tensor, the area discrimination capability and the edge classification precision are improved, and the accuracy of edge classification is improved. The texture stability is judged by means of anisotropic standard deviation, a fuzzy edge mask is set, truncation paths are screened in combination with a main direction vector included angle deviation trend, and continuity and stability of a boundary convergence position are ensured.

The invention discloses a sparse visual angle three-dimensional reconstruction method and system based on voxel grid constraint, and belongs to the technical field of visual three-dimensional reconstruction. Constructing a three-dimensional voxel grid of a self-adaptive scene scale based on point cloud distribution, and dividing the point cloud to the corresponding voxel grid; fusing the geometric features of the fast point feature histogram and the confidence score in the grid, generating a geometric confidence comprehensive measure, and screening key points to initialize Gaussian primitives; designing a voxel grid constrained gradient clipping strategy, limiting Gaussian primitive error diffusion through a distance attenuation coefficient, and adaptively optimizing grid distribution in combination with a dynamic grid deletion and addition mechanism; and finally, carrying out iterative training by using a 3D Gaussian splash radiation field loss function to realize high-fidelity static scene reconstruction under a sparse view angle. According to the method, scene geometric priori is introduced, an optimization strategy based on voxel grid constraint is designed to effectively control excessive diffusion or drift of Gaussian primitives, generation of artifacts is reduced, and meanwhile, the situation that robustness is reduced due to the influence of priori quality is avoided.

The invention relates to the technical field of machine vision and wind power detection, and discloses a surface defect identification method and system for a rotor core of a wind driven generator, and the method comprises the steps: collecting an original image of the surface of the rotor core, and carrying out the gray mapping, and obtaining a single-channel gray image; carrying out frequency domain periodic texture suppression and local histogram equalization processing with contrast limitation on the single-channel grayscale image to obtain an enhanced feature image; constructing a background fitting model for the enhanced feature image and performing differential operation to obtain a background differential image; performing adaptive segmentation and morphological refining based on a texture direction according to the background difference image to obtain a refined defect connected domain; local gray level distribution is extracted, sub-pixel-level geometric feature calculation is carried out, and defect geometric attribute data are obtained; and spatial clustering and grading evaluation are carried out according to the data, and a final defect distribution map is determined. According to the method, periodic texture interference can be effectively suppressed, and sub-pixel-level precise positioning and intelligent grading of the tiny defects are realized.

The invention relates to a progressive memory delay monitoring and diagnosis method oriented to a virtual memory subsystem. The method comprises the following steps: mounting a BPF program to a plurality of key event points on a kernel; when the key event point is triggered, acquiring state information of a corresponding event; packaging the state information into an event, and writing the event into a perf ring buffer associated with the outputmap; and the user mode program asynchronously reads the packaged event from the perf ring buffer, performs classification and aggregation according to the process and the event type, generates a delay terminal log and a histogram, and outputs a diagnosis result. Through systematic architecture design, the state keeping capability and the kernel context access capability of the eBPF are combined with an event model of a virtual memory subsystem, and a set of closed-loop monitoring system oriented to a diagnosis target is constructed. High-precision and high-reliability memory operation delay measurement is realized.

The invention discloses an industrial Internet of Things federated learning method based on a federated increment decision tree, and the method comprises the steps: a cloud server deploys and initializes a federated learning global model and the federated increment decision tree, and sets a statistical histogram bucket boundary set of each feature value; in each federated learning iteration, the cloud server broadcasts a federated learning global model parameter, each industrial device adopts a local data set to train and count to obtain a local gradient histogram parameter, the edge server performs local aggregation on the local gradient histogram parameter and the federated learning model parameter, and the edge server performs local aggregation on the local gradient histogram parameter and the federated learning global model parameter; and the cloud server globally aggregates the local aggregation parameters of the gradient histogram and then incrementally trains the federated increment decision tree, simultaneously aggregates the parameters of a federated learning global model, and adaptively calculates an aggregation weight based on a second-order gradient value during local aggregation and global aggregation of the parameters of the federated learning model. The federal learning overhead can be effectively reduced, and the performance of the federal learning model is improved.

The invention relates to the technical field of computer graphics and deep learning, in particular to a simulation bait automatic coloring method and system based on a 3D model. The method comprises the following steps: acquiring an uncolored 3D model and a reference image, and generating standard data through analysis verification, curvature grid division and image compliance detection; performing color conversion, texture enhancement and multi-scale downsampling on the compliant image to construct an image pyramid; performing multi-level feature extraction and adversarial training optimization based on a pre-trained convolutional neural network and a generative adversarial network, and generating an enhanced color texture map; performing UV expansion, color mapping and normal mapping fusion in combination with the model topology, and constructing an intermediate model with physical rendering attributes; and batch color consistency verification is realized through color histogram comparison, adaptive threshold segmentation and iteration parameter adjustment, and a standard model group is generated. According to the invention, efficient and highly realistic automatic coloring of the simulated bait is realized, and color consistency and rendering quality in batch production are guaranteed.

Linear unretimed interfaces lack clock and data recovery and retiming circuits implemented on a sophisticated digital signal processor. When the interface is not a retimed interface, it can be especially useful to extract some information about the link to allow for debugging and optimization of system deployment. An efficient solution can be implemented in unretimed interfaces to compute metrics such as the impulse response and signal histogram. Having the metrics allows for the solution to check and monitor the quality and the status of the link. Based on the link quality and status information, it is possible to address non-idealities and optimize performance of the unretimed interfaces and the link.

The invention discloses a target vehicle attitude recognition method based on 3D point cloud feature extraction, and the method comprises the steps: carrying out the preprocessing of the point cloud data of a target vehicle obtained through a laser radar, and the preprocessing comprises Gaussian statistical filtering, voxelization sampling, and vehicle tire point cloud extraction based on a region growing algorithm; based on the point cloud data, key points of tire steering features are extracted, and a Harris corner detection method is used for detecting the key points; calculating a fast point feature histogram (FPFH) descriptor based on the extracted key points, and performing coarse registration by adopting an SAC-IA algorithm to obtain a coarse registration matrix; and using a point-to-surface nearest point iteration ICP algorithm to perform fine registration on the point cloud after coarse registration to obtain attitude information of the target vehicle. According to the method, the three-dimensional attitude angle of the target vehicle is accurately estimated by preprocessing the laser radar point cloud, extracting tire key points and carrying out SAC-IA coarse registration and ICP fine registration. Therefore, vehicle attitude recognition with high shielding robustness and high precision is realized, and the safety of automatic driving is enhanced.

The invention discloses an unmanned aerial vehicle small target detection method based on Mama feature fusion, and relates to the technical field of computer vision, and the method comprises the steps: multi-modal data collection, synchronous collection of images and point cloud data through three types of sensors, and coverage of multi-scene and environment conditions; image preprocessing adopts improved bilateral filtering, adaptive histogram equalization and point cloud downsampling to unify a coordinate system; in the multi-scale feature extraction, five-level scale features are output through an improved CSPDarknet network, and the five-level scale features are enhanced through an SENet attention module; in the Mama feature fusion, multi-modal and multi-scale features are processed through a three-stage unit; generating a candidate frame by a dynamic anchor frame, screening according to IOU, and improving YOLOHead to realize classification and positioning; the dynamic optimization of the detection result finely adjusts parameters through on-line distillation. According to the method, the precision, the real-time performance and the anti-interference capability of small target detection of the unmanned aerial vehicle are improved, and reliable technical support is provided for low-altitude security, exploration and other scenes.

The invention discloses a three-dimensional molecule generation method fusing conditional flow matching and Van der Waals radius constraint, which comprises the following steps: processing a molecule training data set, and extracting a total number of atoms and a training element component histogram; based on the optimal transmission path interpolation, combining the sampling time step and the standard Gaussian noise to construct a noise coordinate and a target condition velocity field; the noise coordinates are input into a continuous flow matching prediction model, node features are extracted through affine transformation modulation, a prediction velocity field is obtained, soft atom type distribution is generated, and the expected Van der Waals radius of each atom type is calculated; calculating flow matching loss through a prediction velocity field and a target condition velocity field, calculating a geometric collision penalty term in combination with an expected Van der Waals radius and a noise coordinate, and constructing a total loss function training model parameter; and defining an ordinary differential equation by using the trained parameters for solving, and outputting a three-dimensional molecular structure file. According to the method, atom space overlapping is inhibited, and the physical rationality and chemical effectiveness of generated molecules are improved.