265 results about "Feature descriptor" patented technology

The invention provides a modified metal surface defect detection method and device and a medium, and the method comprises the steps: obtaining a multi-angle reflection image sequence of a modified metal surface under the irradiation of a multi-spectral light source, and generating a defect sensitive parameter set based on a preset modified metal material characteristic database and in combination with the spectral reflectivity distribution information of the multi-angle reflection image sequence; performing cooperative feature enhancement on the multi-angle reflection image sequence and the defect sensitive parameter set, and enhancing the feature contrast of a defect area and a normal area through weight distribution to obtain an enhanced defect feature set; joint anomaly detection of a spatial domain and a spectral domain is carried out on the enhanced defect feature set, a potential defect region of the modified metal surface is obtained through identification, and a defect region feature descriptor is generated; and performing defect morphological quantitative analysis according to the defect region feature descriptors, and determining the type, position and severity level of the modified metal surface defect. According to the invention, the comprehensiveness and reliability of a defect detection result can be improved.

The invention relates to the technical field of mineral processing visual detection, and discloses a visual analysis system for detecting the grade of a phosphorite flotation froth layer. The system comprises an image acquisition and decomposition module, a parallel feature extraction module, a dynamic feature fusion module, a foam evolution analysis module and a grade decision output module. The system performs multi-scale decomposition on a foam image, extracts physical and semantic features in parallel, constructs a dynamic fusion network based on bidirectional mapping to perform iterative interaction, and generates a multi-modal feature descriptor. Therefore, self-organizing growth of a foam evolution graph is driven, an evolution track of a key foam primitive is positioned and tracked, a foam grade state vector is formed, and finally, a regulation and control decision is output in combination with external control parameters. According to the system, deep fusion of physical and semantic features and deep analysis of the foam dynamic evolution process are achieved, the accuracy and predictability of foam grade state sensing are improved, and an effective means is provided for accurate control over the flotation process.

The embodiment of the invention discloses a building engineering crack detection method and system based on image recognition, and the method comprises the steps: obtaining a building surface image, carrying out the preprocessing of the image, obtaining a standardized image, and carrying out the multi-scale decomposition extraction and integration of various features, and forming a multi-dimensional feature descriptor set; after feature importance is evaluated, a compact feature vector is generated through dimension reduction, quantization coding and compression, and then a multi-level feature index mechanism for optimized compression is constructed. A query feature vector is extracted from a newly collected image, searching and screening are completed by means of an index mechanism and a tolerance threshold, and a crack matching result is obtained; and based on the result, positioning cracks, classifying types, measuring parameters and evaluating severity, and generating a crack state report. The crack trend is analyzed in combination with the historical data time sequence, a multi-stage early warning mechanism is designed, maintenance suggestions are provided, and a real-time monitoring and early warning system is formed. According to the embodiment of the invention, the technical problems of high storage pressure and low real-time detection efficiency in the prior art can be effectively solved.

The embodiment of the invention provides a three-dimensional space change detection method and device, equipment and a storage medium, and relates to the field of digital image processing. The method comprises the following steps: acquiring image sequences of the same area shot by the unmanned aerial vehicle in two periods; for each image sequence, performing point cloud reconstruction on the region based on the image sequence to obtain the group of point cloud data; point cloud registration is carried out according to the similarity between the feature descriptors of all the points in the two sets of point cloud data, a matching point pair set comprising a plurality of matching point pairs is obtained, and two points in each matching point pair belong to two pieces of point cloud data and are closest to each other; and for each matching point pair, geometric information of two points in the matching point pair in respective point cloud data is acquired, and if it is determined that the difference between the two pieces of geometric information meets a preset difference condition, it is determined that the corresponding position of the matching point pair in the region is changed. According to the embodiment of the invention, the problem of insufficient spatial change detection precision in a complex scene is solved.

The invention relates to the technical field of unmanned aerial vehicle image processing, and discloses a high-definition image processing and transmission method based on an unmanned aerial vehicle. The method comprises the following steps: acquiring a multi-frame high-definition image sequence obtained in the flight process of an unmanned aerial vehicle; analyzing the image sequence frame by frame, extracting key feature points and generating feature descriptors; establishing a feature matching relationship between adjacent frames, and calculating an inter-frame motion vector; constructing a global motion model according to the motion vector and performing motion compensation; performing region segmentation on the compensated image sequence to divide the image sequence into a dynamic region and a static region; performing multi-scale noise reduction on the dynamic region, and performing texture enhancement on the static region; carrying out fusion reconstruction on the processed dynamic and static regions to generate an optimized image sequence; performing layered coding on the optimized sequence according to a transmission bandwidth condition, and generating base layer and enhancement layer data; dynamically adjusting the priority of two layers of data transmission according to the network state; and sending the data to a ground receiving end through a multi-path transmission protocol.

The invention discloses an urban canyon positioning method and system fusing GNSS, vision and low orbit satellites, GNSS signals and images near a visual sign are collected in an urban canyon to serve as mapping points, and sign angular point coordinates are recorded; and analyzing to obtain a GNSS latitude and longitude list corresponding to each mark, and extracting LoFTR feature descriptors and storing the LoFTR feature descriptors in a map database. During positioning, firstly performing rough matching with the current position through the KNN, and then realizing fine positioning in the candidate area through LoFTR feature matching and homography matrix calculation; and if the matching fails, calculating a joint confidence coefficient based on the satellite PDOP / number, and estimating the current position through dynamic weighting by fusing Kalman filtering with a visual odometer. According to the method, a two-stage positioning strategy is adopted, GNSS and visual information are preferentially used for high-precision matching positioning, a fusion positioning mechanism based on low-orbit satellite joint confidence is started after failure, and the matching success rate and the positioning precision under the complex visual condition are improved.

The invention relates to the technical field of data processing, in particular to an automatic driving vehicle data processing system based on identity recognition, comprising: a data acquisition module outputting an original data stream carrying a digital signature; the preprocessing module filters illegal data streams according to the digital signature, and outputs time-aligned laser radar data and visual data; the feature fusion module outputs a six-degree-of-freedom pose estimation result; the safety processing module generates safety feature descriptors according to the cosine similarity, the timestamp difference constraint range and the encryption seeds; the verification storage module stores the Merkel tree, the encrypted point cloud and the encrypted image to a candidate node group; and the calibration module calculates a correlation coefficient between the timestamp of the visual data and the timestamp of the laser radar data, and reconfigures a hardware trigger time sequence of the data acquisition module. According to the method, the problem of cross-modal data fusion credibility caused by sensor clock skew is solved, and the identity recognition accuracy is improved.

The invention discloses a high-precision satellite attitude correction method, system and equipment for a maximum initial positioning error, and belongs to the technical field of machine vision image matching. A proper reference region is selected to prepare a first reference image and a second reference image, and feature descriptors of the reference images are extracted offline; performing down-sampling on a remote sensing image acquired by a satellite remote sensing camera, and performing image matching on a real-time image after down-sampling and the first reference image; calculating the optimal window position of the ground calibration point of the reference area in the real-time image according to the matching result; cutting the real-time image according to the window position to obtain a high-resolution calibration point real-time image; performing image matching on the high-resolution calibration point real-time image and the second reference image; and calculating latitude and longitude coordinates of the center of the real-time image according to a matching result for satellite attitude correction. The position error of the target area is gradually reduced in a two-stage remote sensing image matching mode, the longitude and latitude coordinates of the center of the real-time image are accurately positioned, and high-precision satellite attitude correction is realized.

The invention discloses a screening method of insulating gas decomposition product sensing materials based on machine learning, and relates to the technical field of functional material design and gas sensing. The method provided by the invention comprises the following steps: selecting a first candidate sensing material according to an insulating gas and a decomposition product; doping elements or embedding atoms into a central cavity of the first candidate sensing material to obtain a second candidate sensing material, and constructing a candidate sensing material library; obtaining the most stable adsorption configuration and adsorption energy of gas molecules, constructing feature descriptors, forming an initial data set, and dividing a training set and a test set; constructing and training a model, and evaluating and screening out an optimal model by using a test set; predicting and screening the second candidate sensing material by using the optimal model to obtain a third candidate sensing material; and performing DFT calculation verification and performance index evaluation on the third candidate sensing material, and outputting a final candidate sensing material. The method provided by the invention is high in prediction accuracy and reliability, and a novel sensitive material can be found.

The invention relates to the technical field of power electronics, and discloses a surge voltage suppression system and a control method thereof. The system comprises a surge feature extraction unit, a suppression map generation unit, a strategy deduction and check unit and an instruction synthesis and distribution unit which are connected in sequence. According to the method, surge feature descriptors are extracted and fused with a historical case library and a real-time power grid topology, and a suppression atlas containing multi-path parameter combination is dynamically generated; and performing time sequence deduction and interaction influence calculation on each path in the map, iteratively correcting parameters to output a checked strategy sequence, and finally converting the strategy sequence into an instruction to control the action of the physical suppression equipment. According to the method, the whole process of the suppression strategy from historical experience learning, dynamic generation to global collaborative rehearsal optimization is realized, and the adaptability, the collaboration and the reliability of responding to surges in a complex power grid environment are improved.

The invention relates to the technical field of pathological section image splicing, and discloses a pathological section association-based pathological image splicing method and device. The method comprises the following steps: acquiring and preprocessing a serialized pathological sub-slice image; calculating a texture flow direction field representing microstructure arrangement of the tissue, and constructing a multi-scale feature descriptor; on the basis of the multi-scale features, the adjacency relation between the sub-slices is quantitatively evaluated by fusing gradient differences, texture distribution KL divergence and a multi-scale correlation degree function of frequency domain mutual information; constructing a global spliced graph model by taking the degree of association as an edge weight, establishing an energy function containing data fidelity and texture smoothness constraints, solving an optimal splicing parameter through iterative optimization, and controlling an iterative process according to convergence criteria of comprehensive splicing stability, residual matching potential and progress; and finally generating a seamless panoramic pathological image. According to the method, biological structure characteristics are utilized, robustness and matching accuracy under complex conditions are improved, and continuity and consistency of splicing results are ensured.

A method and system of feature matching for associating vehicle perception data for crowd sourcing map data. The system is configured to execute the method including the steps of receiving perception data from a plurality of vehicles; processing the perception data to identify first feature points observed by a first vehicle and second feature points observed by a second vehicle, respectively; creating feature descriptors using multilayer bitmaps for each of the first feature points and for each of the second feature points; determining pairs of matching first and second feature points by comparing histograms of the feature descriptors; associating pairs of matching first and second feature points; updating a map database with the associated pairs of matching first and second feature points; and operating an autonomous vehicle based on the updated map database.

The invention relates to the technical field of three-dimensional modeling, in particular to an augmented reality-oriented three-dimensional scene reconstruction method and system, and is used for solving the technical problem that an iterative nearest point algorithm cannot meet high real-time performance and high-precision reconstruction requirements required by augmented reality at the same time. The method comprises the following steps: acquiring a pose transformation relationship between multiple frames of point cloud data and adjacent frames of point cloud data; extracting a plurality of feature points in each frame of point cloud data, and determining a feature descriptor of each feature point; according to the feature descriptors of the feature points, calculating a matching weight between feature point pairs in two adjacent frames of point cloud data, and calculating a local matching error based on the matching weight and a pose transformation relationship; performing global matching optimization based on the local matching error of the plurality of feature point pairs and the distribution density of each feature point in the respective point cloud frame, and determining an optimal matching point pair set which minimizes the global matching error; and according to the optimal matching point pair set, fusing the multi-frame point cloud data, and reconstructing a three-dimensional scene model.

The invention relates to the technical field of planet remote sensing data processing and three-dimensional reconstruction, and discloses a Mars three-dimensional terrain reconstruction method based on feature enhancement and multi-view stereo matching, which comprises the following steps: firstly, obtaining and preprocessing Mars stereo image data; secondly, extracting texture features based on a gray-level co-occurrence matrix, and performing region segmentation by using local entropy to distinguish a high texture region from a low texture region; then, parameters such as the size of a matching block and a similarity threshold value are adaptively adjusted according to a segmentation result, and a multi-dimensional mixed feature descriptor is generated to perform homonymy point matching; a coupling threshold model of the solar incident angle and the terrain roughness is constructed, and parallax calculation of the shadow and the steep terrain is optimized; and finally, combining cross-scale parallax propagation and global bundle adjustment to reconstruct a three-dimensional terrain model. According to the method, the technical problems that the matching reliability of areas such as weak textures and shadows on the surface of Mars is insufficient and the model is discontinuous are solved, and the integrity, the self-adaptability and the global precision of the reconstruction model are improved.

The invention provides a retina image registration method and system based on grid supervision and key point learning, and belongs to the technical field of medical image processing. The method comprises the following steps: acquiring an RGB fundus image pair; encoding the RGB eye fundus image pair through a shared feature extraction branch to obtain a down-sampling feature map; processing the down-sampling feature graph through a key point extraction branch and a descriptor decoder branch respectively, and extracting a key point probability graph and a feature descriptor; predicting the offset of the key points in the local window based on the paired feature descriptors through a key point offset prediction module; and carrying out position calibration on the key points in the key point probability graph based on the offset, and outputting a registered image. According to the method, self-supervised training is realized by using dense gridding marking points, and manual marking is not needed; through lightweight network design and a key point offset calibration mechanism, high registration precision is maintained while the parameter quantity is significantly reduced, and the method is suitable for clinical real-time deployment and retinal disease analysis.

The invention provides an image splicing method and device based on a pod, electronic equipment and a storage medium, and relates to the technical field of image processing. The method comprises the following steps: sending a control instruction to a stabilizing and pointing unit; when return information sent by the stabilizing and pointing unit is received, acquiring N paths of video data of the pod, and determining a plurality of to-be-spliced images and a reference image from the video data; performing feature processing on the reference image and each to-be-spliced image by using the FPGA to generate a first feature descriptor and a second feature descriptor, and performing feature matching on the reference image and each to-be-spliced image according to the first feature descriptor and the second feature descriptor to obtain a feature matching point pair set; according to the attitude angle information, the SOC chip is utilized to perform estimation transformation on each feature matching point pair in the feature matching point pair set, and multi-path splicing processing is performed on the reference image and each to-be-spliced image based on the transformation matrix set to obtain the target spliced image, so that the image splicing accuracy is improved.

The invention provides a fault prediction and self-protection method and system for heave compensation of a diving bell, and the method comprises the following steps: obtaining the displacement of a hydraulic cylinder of a heave compensation device and the pressure data of an energy accumulator, constructing a hysteresis loop, converting the hysteresis loop into a feature field image, and analyzing and calculating a topological feature descriptor through geometric moment. By performing inverse frequency domain analysis on the pressure data, a friction texture signal is separated and an energy entropy value of the friction texture signal is calculated. Mapping the topological feature descriptor to a control variable space defined by a nonlinear cuspidal point mutation model, and calculating a geometric distance from a current state point to the boundary of a model bifurcation set; and solving an instability prediction index by combining the distance and the energy entropy. And when the index exceeds the limit, the system triggers variable damping control and gradually locks the hydraulic cylinder to realize self-protection. According to the method, the technical problem that the sudden instability failure of the system caused by multi-factor coupling is difficult to warn in advance by a traditional method is solved.

The application discloses a kind of motion target fusion processing method, equipment and computer readable storage medium, wherein, the method comprises: in the process of shooting, the mask of the motion target area of current frame is acquired, and the minimum circumscribed rectangle of the mask is calculated;The feature descriptor of the edge of the minimum circumscribed rectangle is calculated;According to the feature descriptor, feature matching is carried out to obtain the highest similarity of the feature descriptor of motion target in the motion target area in the current frame with the coordinates;In the current frame, the motion target is fused into the highest similarity position according to the coordinates, to determine the freeze target of the motion target.An efficient motion target fusion processing scheme is realized, which effectively improves the extraction accuracy and fusion effect of the motion target, and enhances the user's shooting experience.

This invention discloses a method for reusing obstacle avoidance trajectory parameters based on local environmental geometric and motion constraint features. The method includes the following steps: acquiring a local point cloud to construct a local map; calculating the geometric residuals from points to planes based on the local map to obtain residual statistical features; calculating the plane normal distribution based on the local map to obtain structural orientation features; and constructing an information matrix based on the Jacobian matrix optimized by the direct method pose to obtain motion constraint features; combining the residual statistical features, structural orientation features, and motion constraint features into a local environmental feature descriptor; binding the local environmental feature descriptor with trajectory generation parameters of historical obstacle avoidance trajectories and storing it in a database; calculating the similarity between the current local environmental feature descriptor and the historical feature descriptors in the database, and selecting the corresponding historical trajectory generation parameters based on similarity matching; and using the historical trajectory generation parameters as the initial parameters for the current obstacle avoidance trajectory planning.

The invention provides an intersection multi-source three-dimensional image processing method and device and computer equipment, and belongs to the technical field of image processing. The method comprises the following steps: acquiring an original image of each camera of a target intersection, and generating a priori depth map of each camera based on a three-dimensional map; performing feature extraction on each original image by an image processing model to obtain each image feature, splicing each image feature and each depth map, and inputting the spliced image features and each depth map into a feature encoder to obtain each aerial view feature map; performing feature separation on each aerial view feature map to obtain each path of similarity description sub-feature and each aerial view feature; on the basis of each path of similarity descriptor feature, determining a feature descriptor corresponding to the to-be-detected target; and predicting the target attribute of the to-be-detected target according to the feature descriptor and each aerial view feature. According to the invention, the detection precision of the intersection multi-source three-dimensional image can be improved, the prior advantages of intersection multi-source three-dimensional image detection can be brought into full play, and the safety and reliability of a roadside sensing system can be improved.

The application discloses a kind of integrated circuit surface defect classification methods based on differential manifold, including: microscopic imaging acquisition system, image pre-processing system and defect image classification system, first, the integrated circuit image after pre-processing is mapped to Riemann manifold space;Second, the feature descriptor of image is constructed, the characteristic vector is obtained, that is, covariance matrix can be constructed and corrected as the positive definite symmetric matrix of Lie group manifold;Then, further map the matrix to low-dimensional Lie group space, design Lie group kernel function;Finally, realize the accurate classification of integrated circuit surface diversity defect image.The application solves the problem of rapid classification of integrated circuit surface diversity defect image, improves the reliability in the process of integrated circuit manufacturing, and guarantees the quality of integrated circuit downstream application industry.

The invention discloses a method and system for monitoring the start-stop state of a conveying belt based on a traditional visual algorithm, and belongs to the technical field of traditional visual algorithms, and the method comprises the steps: installing an industrial camera at a proper position above the conveying belt; after receiving the original image acquired by the industrial camera, the monitoring host carries out graying processing on the image and carries out de-noising processing on the grayscale image; performing histogram equalization processing on the denoised image to enhance the contrast of the image; selecting a plurality of feature points in the belt area, and recording coordinates and feature descriptors of each feature point; for every two adjacent frames of images, matching the extracted feature points by using a feature point matching algorithm, and calculating an optical flow vector of each pixel point between the adjacent frames by using a feature descriptor-based matching algorithm and an optical flow method; and outputting a state and giving an alarm. Manual intervention is greatly reduced, accuracy is high, cost is low, adaptability is high, and implementation and maintenance are easy.

The application discloses a kind of based on machine learning prediction amide bond molecular synthesis conversion rate method, belong to chemical technical field.The application is based on high-throughput experiment collection reaction conversion rate data, establishes data set, and utilizes quantum chemistry calculation means to extract the feature descriptor of reaction related molecule, and the model after training finds that random forest model only needs a small amount of data to reach good prediction accuracy.Through the prediction of unknown aromatic amine molecular conversion rate in amidation reaction, it is found that a small amount of relevant reaction data can be added in the training set of random forest model to significantly improve the prediction performance of unfamiliar data.This has important guiding significance for exploring the synthesis of amide bond and the future application of machine learning to the field of chemical synthesis.

The invention provides an interface control processing method, electronic equipment, a storage medium and a program product, and relates to the technical field of image processing. The method comprises the steps of obtaining window image data of a target application; based on an image template library, matching the window image data to determine whether an interface control exists in the target application, the image template library being used for storing feature descriptors of the interface control; if the interface control exists, determining position coordinates of the interface control according to the window image data; and according to the control type and the position coordinate of the interface control, executing an interaction operation matched with the control type. According to the method and the device, the recognition accuracy of the lightweight interface control is improved.

Systems and techniques are provided for processing sensor data. For example, a process can include obtaining input data and processing, using a non-machine learning based feature detector feature detector, the input data to determine one or more feature points in the input data. The process can further include determining, using a machine learning system, a respective feature descriptor for each respective feature point of the one or more feature points.

The invention relates to the field of point cloud registration, and discloses an ICP registration method fusing KNN vector features and quaternion optimization, and the method comprises the steps: obtaining a local plane normal vector of a source point cloud, calculating a covariance matrix based on the source point cloud, carrying out the feature decomposition of the covariance matrix, obtaining a feature vector, and calculating the minimum feature vector of the source point cloud; calculating a feature descriptor of the point cloud corresponding to the minimum feature vector in the source point cloud based on the local plane normal vector of the source point cloud, and selecting the point cloud of which the feature descriptor is greater than a preset threshold as a feature point; constructing a neighborhood for the feature points and carrying out feature point matching to obtain matched feature points; obtaining a first precise point pair based on the matched feature points, and performing weight configuration on the first precise point pair to complete preliminary registration; and performing iterative registration on the first precise point pair after weight configuration based on quaternion optimization. According to the method, registration is optimized by fusing KNN vector features and quaternion, and accurate registration of point clouds can be realized.

The invention provides an insulator detection method and system based on cooperative work of an unmanned aerial vehicle and a robot. The method comprises the following steps: preprocessing an original laser radar point cloud to obtain preprocessed laser radar point cloud data; segmenting the preprocessed laser radar point cloud data and removing earth plane point cloud; clustering the residual laser radar point cloud data to obtain a plurality of point cloud clusters; calculating a feature descriptor of each point cloud cluster; calculating the similarity between the feature descriptors of the to-be-identified cluster and descriptors in a template library, and determining an identification result of the to-be-identified cluster based on the similarity; and obtaining a target whose identification result is an insulator chain, carrying a detection robot by the unmanned aerial vehicle, and placing the detection robot on the corresponding insulator chain so as to complete detection of the insulator chain. According to the method, the feature descriptor of each point cloud cluster is calculated and is matched with the template library, so that the insulator string and other components can be accurately distinguished, and the detection robot can accurately put and quickly recover the insulator string.

This invention discloses an automatic mural splicing control system and method based on image recognition, belonging to the field of image recognition technology. The method includes: performing image recognition decomposition and structured encoding on the mural design drawing to obtain an assembly code map, a target wall reference rendering, and a 3D preview model; acquiring multi-view images of the wall work area and completing camera calibration, determining the transformation relationship between the camera coordinate system and the wall coordinate system to form a wall mapping matrix, and using the wall mapping matrix to map the assembly code map into a target placement coordinate table; acquiring pixel block images of the material area, extracting the appearance contour and appearance feature descriptors, and performing nearest neighbor matching with the assembly code map to obtain the corresponding entries in the target placement coordinate table. By combining the assembly code map with the target placement coordinate table, this invention can accurately generate the grasping and placement posture sequence of the robotic arm, achieving automated assembly and significantly improving the accuracy and efficiency of the assembly process.

This invention belongs to the field of national defense and security technology, and specifically relates to a method for single-angle rotation sampling and feature extraction for target detection in remote sensing images. It efficiently extracts features from the input feature map, generating refined feature descriptors for predicting weights and angles. A weight-angle dual-path prediction head, based on the refined features, predicts a set of optimal rotation angles and corresponding combined weights in parallel. A multi-kernel dynamic combination, serving as the output and application stage of the routing function, performs a one-time rotation and weighted fusion on m basic convolutional kernels based on the predicted sum, generating the final dynamic rotation kernel used for convolution, thus achieving multi-angle feature extraction equivalent to a single operation. This invention constructs an efficient routing prediction network that can directly and accurately predict the optimal convolutional kernel rotation angle and the combined weights of a set of basic convolutional kernels for the current target region based on the contextual information of the input features.