28results about How to "Fast recognition" patented technology

This invention discloses a hybrid quantization method for image recognition models based on bit-flipping attacks, comprising: Step 1, performing initial quantization on the image recognition model to convert it into an initial quantized model; starting from the first layer of the initial quantized model, performing bit-flipping attacks layer by layer, and recording the sensitivity of each layer under attack; forming a sensitivity set of the initial quantized model from the sensitivities of all layers; constructing an optimization objective function and determining optimization constraints, and determining the optimal hybrid precision quantization strategy for the image recognition model based on the optimization objective function and the optimization constraints; Step 3, setting the bit width for storing the parameters of each layer of the image recognition model according to the optimal hybrid precision quantization strategy, thereby obtaining the optimal hybrid quantization model; Step 4, acquiring the image to be recognized and performing image recognition using the optimal hybrid quantization model.

The application provides a road surface state recognition method and device, and solves the problems of single road surface state recognition category, coarse result granularity and inability to distinguish specific road surface abnormal types caused by dependence on simple threshold judgment and shallow features in the prior art. The method comprises the following steps: acquiring a vertical vibration acceleration data segment of a vehicle in operation; extracting a characteristic value according to a preset optimal feature subset, wherein the subset comprises a frequency domain peak value and a spectrum energy; inputting the characteristic value into a classification model to obtain a recognition result, wherein the model is obtained by training based on data comprising at least three road surface state labels. Through fixed feature optimization, the application realizes fine recognition of multiple states such as well frame difference, expansion joint and deceleration strip, and significantly improves the calculation efficiency while ensuring the accuracy.

The invention discloses a modal shape and frequency identification method for a motor stator system with an inner gear ring, and the method comprises the steps: building a shell unit and beam unit mixed analysis model of the motor stator system provided with the inner gear ring, and obtaining a total stiffness matrix and a total mass matrix under a local rectangular coordinate system through a matrix displacement method, obtaining a structural stiffness matrix and a structural mass matrix under the overall cylindrical coordinate system by using coordinate transformation; and establishing a discrete free vibration model, inputting the structural stiffness matrix and the structural mass matrix into the model, and outputting modal shapes of different orders and corresponding inherent frequencies of the motor stator system, thereby completing identification. The method comprehensively considers the influence of the stator teeth and the tooth part of the inner gear ring, helps to rapidly identify the modal shape and inherent frequency of the stator system, provides guidance for subsequent vibration acquisition and optimization, and has the advantages of simple modeling, rapid identification and high precision.

The invention relates to the technical field of road identification, in particular to a straight road and intersection rapid identification method based on road width distribution characteristics, which comprises a window position selection method, a simple identification method of simple elements, an identification method of other elements and a comprehensive decision. According to the method, windows at the near, middle and far positions are recognized through a window position selection method, then elements are recognized through a simple element recognition method or other element recognition methods, and finally, comprehensive decision making is carried out according to recognition results of the near, middle and far windows, so that a final recognition result is obtained. According to the method, different shooting windows can be selected according to different vehicle speeds and camera positions through a set window position selection method, a simple recognition method of simple elements, a recognition method of other elements and a comprehensive decision, and then the width and the distribution characteristics of the road are calculated according to the inverse perspective images shot by the shooting windows; and finally, outputting the identified intersection type.

The application discloses a license plate quality judgment method based on semi-supervised learning, comprising: obtaining a license plate image to be judged, and obtaining a license plate character local image according to an existing license plate detection algorithm; inputting the license plate character local image into a pre-trained deep neural network model, wherein the deep neural network model comprises a feature extraction module and a license plate quality prediction module; using the feature extraction module to perform feature extraction on the license plate character local image to be judged, and obtaining a high-level feature map comprising the license plate character local image; using the license plate quality prediction module to process the high-level feature map, and obtaining a license plate quality index of the license plate character local image; the network model of the license plate quality judgment method has strong universality, is faster in recognition, has stronger robustness for low-quality license plate images with dirt, adhesion, character loss and inaccurate positioning, does not need additional labeling of license plate quality, and is more convenient in model training.

This application relates to the field of circuit board manufacturing technology, and discloses a method for manufacturing SMT carrier markings and an SMT carrier. The method for manufacturing SMT carrier markings includes: providing an SMT carrier with a marking area; performing an oxidation treatment on the SMT carrier to cover its surface with an oxide layer; using a first laser to engrave the oxide layer within the marking area, making the marking area white; and using a second laser to further engrave the oxide layer within the marking area to form a marking, which is black. The above-described method for manufacturing SMT carrier markings optimizes the structure of the prepared markings, creating a strong contrast between the marking and the background color of the marking area, making them easily identifiable; furthermore, the white background color of the marking area reflects radiation and lowers the surface temperature of the carrier, making it less prone to discoloration and improving recognition speed. The markings prepared by this application have strong anti-aging properties, increased strength and hardness, and are less prone to wear.

The invention provides an abnormal capital collection account group identification, clustering and label labeling method based on capital flow data, and belongs to the technical field of financial risk prevention and control and data mining. A time sequence LSTM-structured self-encoding fusion AI feature extraction module is customized; the method comprises the following steps: capturing time sequence features such as periodic transfer and large-amount concentrated transfer through an LSTM attention layer, extracting structured features such as cross-regional association through an auto-encoder with a risk penalty term, and performing weighted fusion to obtain 12-dimensional AI features; and then semi-supervised K-means is used to identify a suspicious account group, spectral clustering is used to divide a case cluster, a random forest is used to label a'capitator / investor 'label, and the precision is ensured through three-layer verification. The method solves the problems of incomplete artificial feature coverage and poor universal AI adaptability in the prior art, and is suitable for abnormal capital investigation of financial supervision departments.

This application provides a method, apparatus, device, and storage medium for identifying a target object, relating to the field of video processing technology. The method includes: when identifying a target object, first acquiring video frame images of the target object captured by a vision device, and simultaneously detecting the position of the target object by a radar device; and based on the position of the target object, determining the target image region where the target object is located from the video frame images; based on the target image region, extracting a portion of the video frame images containing the target object; and then identifying the target object in the portion of the video frame images to obtain the identification result of the target object. Using the technical solution of this application, the accuracy of the identification result can be improved without being limited by computing power.

The application discloses a substation equipment identification method and system based on an improved point pair feature matching algorithm, and comprises the following steps: obtaining original point cloud data of equipment in a substation, performing mean value processing on point clouds in a grid by using a grid downsampling method to obtain new point cloud data; wherein the size of the grid is automatically adjusted according to the number of the original point cloud data; performing key point screening on the new point cloud data, randomly selecting a reference point from the extracted key points, and respectively forming point pair features of the reference point and other points in the scene; matching the point pair features with pre-stored equipment model point pair features in a hash table to obtain matched point pair features, and identifying substation equipment based on the matched point pair features. By optimizing and screening the key points, the application reduces invalid points and redundant points, thereby improving the calculation speed and matching accuracy, and finally accelerating the substation equipment identification speed.

The invention discloses a bird and unmanned aerial vehicle identification method based on compensated multi-frame target slice and flight path fusion, and the method specifically comprises the steps: firstly, obtaining a multi-frame R-D domain slice data set of a rotor unmanned aerial vehicle and a bird through the staring distance compensation and Doppler compensation operation; secondly, a bidirectional GRU network model fusing track information is adopted, multi-frame target R-D domain slice data are input, and target feature extraction and integration are completed through double-layer convolution and a full connection layer; then, target track features are added into the integrated target feature set, and the target track features and the integrated target feature set are input into a bidirectional GRU network together; and finally, completing the output of a target identification result label through a full connection layer, and completing an identification task. According to the method, the track information and the multi-frame target information are combined, the track features and the motion features of the target are fully utilized, the target recognition accuracy is improved, and the method is simple, high in recognition speed and high in real-time performance.

The application discloses a face attribute recognition model training method and device, and a face attribute recognition method and device. The face attribute recognition model training method comprises the following steps: obtaining a to-be-trained face image, and marking multiple face attribute information in the to-be-trained face image; performing feature extraction on the to-be-trained face image by using a bottom feature extraction network in a face attribute recognition model to obtain a bottom feature extraction result; performing feature extraction and recognition on the bottom feature extraction result by using multiple face attribute recognition networks in the face attribute recognition model respectively to obtain multiple face attribute recognition results; determining a loss value of each face attribute recognition result based on a preset loss function of each face attribute recognition network respectively, and updating parameters of the face attribute recognition model by using the loss value of each face attribute recognition result to obtain a trained face attribute recognition model. The face attribute recognition model can recognize multiple face attribute information simultaneously.

The application provides a commodity risk identification method and device and electronic equipment. The method comprises: obtaining commodity information of a plurality of to-be-identified commodities; inputting the commodity information of each to-be-identified commodity into each prediction model in a plurality of pre-trained prediction models to obtain a prediction risk probability output by each prediction model for each to-be-identified commodity; sorting a plurality of prediction risk probabilities corresponding to a plurality of to-be-identified commodities on the same prediction model to determine an initial sorting result of each to-be-identified commodity, and repeating the step until a plurality of initial sorting results corresponding to a plurality of prediction models for each to-be-identified commodity are obtained; calculating a risk sorting result of each to-be-identified commodity according to the plurality of initial sorting results of each to-be-identified commodity; and identifying the risk of a plurality of to-be-identified commodities and associated stores according to the risk sorting result. The above method can improve the identification coverage, speed and accuracy of commodity risk.

The present application belongs to a novel chromatographic stationary phase, and particularly relates to a preparation method of a cyclodextrin chromatographic stationary phase. The present application adopts tetrafluoro-p-phenylenedinitrile as a representative spacer, is easy to be crosslinked with cyclodextrin to generate a crosslinked cyclodextrin compound with a high specific surface area, and solves the problem of low cyclodextrin bonding amount of the traditional cyclodextrin chromatographic stationary phase. The chromatographic stationary phase can provide pi-pi interaction, hydrogen bond interaction and steric hindrance effect for the separation and analysis of structural analogs, and the inclusion effect of the cyclodextrin hydrophobic cavity is mutually coordinated and cooperated, the selectivity for the structural analogs is improved, and a unique separation effect is generated. As shown in the experimental example 1 of the present application, the peak orders of the compounds with similar structures on the present chromatographic column and the C18 column are different, and the separation selectivity is different. For the compounds which are difficult to separate, a new selectable chromatographic stationary phase is provided, and the separation efficiency of the C18 chromatographic stationary phase is improved.

The application discloses an EPS platform positioning system and method. The EPS platform positioning system comprises an EPS open terminal, a positioning module, a control center module and a map module. A signal transmitting module is used for transmitting an open WIFI signal to enable a user to be positioned to be connected. An environment recognizing module is used for recognizing the surrounding environment by using a camera and searching for the user to be positioned connected to the open WIFI signal. The positioning module is used for determining the position range of the user to be positioned according to the real-time position of the EPS open terminal connected to the user to be positioned, calling the map data of the position range of the user to be positioned, matching the map data with the information recognized by the environment recognizing module and determining the position information of the user to be positioned. The application shortens the rescue time, has high positioning precision, strong real-time performance, improves the environment recognizing speed and improves the positioning accuracy.

A human motion recognition method based on a hierarchical spatiotemporal Transformer network comprises the following steps: collecting motion videos, processing motion videos, constructing a motion dataset, extracting training and testing branches, constructing a hierarchical spatiotemporal Transformer network, training the hierarchical spatiotemporal Transformer network, and testing the hierarchical spatiotemporal Transformer network. Based on the constructed motion dataset, this invention employs a hierarchical spatiotemporal Transformer network to extract high-order spatial feature information of joints at different levels, long- and short-term temporal dependencies, and correlation information between different joints across frames. Compared with existing skeleton motion recognition methods, this invention has advantages such as high recognition accuracy and fast recognition speed, and can be applied to the field of computer-aided human motion recognition technology.

The present application relates to a kind of future and present fusion customer intelligent credit and credit risk management system and method, system includes multi-dimensional customer data acquisition and fusion module;AI dynamic credit scoring model: construct basic credit score, and introduce dynamic adjustment factor to the real-time correction of customer credit score, the comprehensive rating of the customer credit score after correction, early warning according to rating;Future and present comprehensive credit limit real-time management module;Intelligent overdue early warning and collection management module;Customer credit file and blacklist management module: for each customer signal file whole life cycle management, when customer appears set condition, execute blacklist linkage mechanism.The present application is by LightGBM dynamic risk adjustment layer, commodity price fluctuation rate, customer account change trend, associated enterprise risk and so on dynamic factor are included in credit score, risk customer is identified in advance about 30-45 days, and key time window is won for enterprise to take measures in advance.

This utility model provides a combined bottle cap positioning and pressing device, including: a frame, a rotating base, several capping mechanisms, and an image recognition component. The rotating base is rotatably mounted on the frame. The capping mechanism includes a first positioning seat, a second positioning seat, a lifting drive component, and a capping drive component. The first and second positioning seats are mounted on the rotating base. The image recognition component includes a lower image acquisition module and an upper image acquisition module. This combined bottle cap positioning and pressing device of the utility model simultaneously identifies the angles of the first and second caps through the image recognition component, which speeds up the identification process. After identification, assembly can be performed quickly, allowing identification and assembly to be completed on the same device. This reduces intermediate transport paths and avoids inaccurate angles of the first and second caps after subsequent rotation due to shaking or misalignment during transport. It features a high degree of automation, improves production efficiency, and reduces labor costs.

The application relates to a self-adaptive welding method for a stainless steel furniture module in continuous conveying. The self-adaptive welding method for the stainless steel furniture module in continuous conveying comprises the following steps: S2, continuous conveying and profile acquisition; continuous follow-up conveying is performed, and profile acquisition is performed through laser to obtain a laser signal of a keel in a continuous conveying state; S3, relative position determination; the laser signal of the keel obtained in step S2 is solved through an algorithm to obtain the relative position of the keel and a panel; S6, welding fixation; the relative position of the keel and the panel obtained in step S3 is used for welding the keel and the panel. The self-adaptive welding method for the stainless steel furniture module in continuous conveying has the advantages of automation and high efficiency.

The invention provides a voiceprint recognition system and method based on a distributed consistency protocol, and belongs to the technical field of biological feature recognition, and the system comprises a message queue module which is used for receiving, storing and buffering a voiceprint recognition or registration request from the outside; the distributed processing cluster module is composed of a plurality of service nodes, and each service node runs a voiceprint recognition engine and a state machine copy of a distributed consistency protocol and is used for processing a request from the message queue module and maintaining the consistency of a cluster state through the distributed consistency protocol; and the distributed database module is used for persistently storing voiceprint feature data and identifying historical records. The maintainability, the expandability and the state consistency of the system are improved.

The invention discloses a rapid all-sky star map identification method suitable for a star sensor, and relates to the field of photoelectric imaging, the method comprises the following steps: constructing a star map matching model based on a navigation star catalogue, a star diagonal distance table and a star diagonal distance index table, extracting and calibrating star points after obtaining a star map, obtaining a plurality of star vectors, and sorting according to star magnitude; selecting a star vector combination corresponding to the minimum star number threshold value for matching in a manner of gradually expanding a search boundary; in the matching process, calculating a star diagonal distance between star vectors, discretizing the star diagonal distance, quickly retrieving navigation star pair information in an error range by utilizing a star diagonal distance index table, recording a matching relationship in combination with a bit array so as to determine candidate navigation star IDs of the star vectors, and performing overall consistency verification on candidate combinations so as to complete star map recognition; when matching fails, invalid combinations are automatically skipped, and searching is continued. According to the method, while the matching accuracy is ensured, the calculation complexity is effectively reduced, and the star map recognition speed and the false star resistance are improved.

The invention discloses a composite watermark low-coupling embedding and efficient identification method based on wavelet transform. The method comprises the following steps: carrying out wavelet transform on a carrier image and dividing sub-band priorities based on visual sensitivity; carrying out differential preprocessing on the copyright watermark, the authentication watermark and the content description watermark; embedding the watermarks into the corresponding sub-bands according to the priorities and the security requirements, and performing noise suppression; generating a watermark-containing image through wavelet inverse transformation; and extracting a verification watermark by adopting a hierarchical cascade identification architecture, and dynamically adjusting a matching threshold value according to an attack type. According to the method, the robustness of the watermark when facing noise, compression and geometric attacks is improved, the watermark embedding capacity is increased, and the recognition accuracy and efficiency are remarkably improved.

The application discloses a subway leakage water detection method based on point cloud information, utilizes three-dimensional laser point cloud data to segment a tunnel, segments tunnel ring numbers into multiple single-ring point cloud models, uniformly changes point cloud scalar values in each single ring, corrects the range of gray values, replaces the gray values with point cloud intensity values, sets a threshold value, counts all ring piece leakage water areas and proportions, screens ring pieces with larger water seepage areas, and achieves the effect of rapidly detecting a leakage water area. The application directly utilizes intensity values of point cloud information to distinguish the leakage water area. An image is composed of pixel points, generally is converted into an image through gridding of point cloud data, different materials have different light reflection intensities, the intensity values can be directly utilized as gray values for leakage water distinguishing, thus the calculation process can be simplified, the recognition speed is improved, and the application has strong popularization value.

The application discloses a kind of mobile phone control role interaction method, comprising: generating virtual person;Mobile phone completes arbitrary action;Collect mobile action data;Action data is handled according to preset processing rule;The result after processing is adapted according to preset judging rule and is judged, determine what kind of prescribed action is completed by mobile phone;According to the adaptation result of prescribed action, the preset interactive action of virtual person is output.The application innovates mobile phone control interaction mode, identifies hand dance action type and action amplitude by various mobile phone inherent sensors such as gyroscope and gravity sensor, and uses it as the basis for interacting with virtual person in mobile game.At the same time, the data processing method for identifying user hand action is low in power consumption, and the mobile action interaction experience is excellent, which realizes the fusion and promotion of playability and interactivity.

The utility model provides a fillet weld identification auxiliary device. The fillet weld identification auxiliary device comprises two positioning components which are respectively arranged on two fillet weld welding surfaces of a structure to be welded, each positioning assembly comprises a fillet weld starting point marking piece, a fillet weld terminal point marking piece and an auxiliary point marking piece which are arranged in a non-collinear manner; according to the device, the initial starting point and the initial ending point of the fillet weld are marked through the two positioning assemblies, a small collection range is provided for subsequent point cloud data collection with the initial starting point and the initial ending point as the center, and point cloud collection equipment does not need to collect overall point cloud data of a to-be-welded structure; besides, the three marking pieces in the positioning assembly are not collinear, the plane where the fillet weld welding face is located can be determined, the processing amount of the point cloud data of the fillet weld welding face is reduced, and therefore the overall data volume of the point cloud data during fillet weld recognition is greatly reduced, the data processing calculation amount in the subsequent analysis process is further reduced, and the accuracy of fillet weld recognition is improved. The identification speed is improved, installation is simple and convenient, and practical engineering operation is facilitated.

The present application relates to a kind of K-Shell-based FPGA irradiation sensitive signal automatic identification method, system and storage medium, overcome the limitation and inflexibility of prior art only a small amount, limited coverage irradiation sensitive signal is carried out circuit function monitoring, this method is with the aid of K-Shell network model, the top module network of FPGA engineering code file is constructed, all nodes of top module network are screened by calculating the node comprehensive degree or Ks value of node, to obtain key submodule, then according to K-Shell network model, establish key submodule network, with key submodule network related parameters as evaluation index, based on evaluation method, select the irradiation sensitive signal to be determined;Finally, based on specific screening method, determine the final irradiation sensitive signal from it.This method compared with the irradiation sensitive signal selected by human, wide coverage and algorithm identification speed is fast, efficiency is high, can be applicable to different FPGA engineering.

The application discloses a kind of based on Group-Res2Block network's intelligent synthetic speech speaker confirmation method, comprising:1 obtains intelligent synthetic speech data set;2 establishes the feature extraction network model based on Group-Res2Block;3 trains the feature extraction network model based on Group-Res2Block established;4 realizes prediction using the model established, to reach the purpose of confirming the speaker subject of intelligent synthetic speech.This application can maximize to obtain the common effective voiceprint features of natural human voice and intelligent synthetic speech, so as to effectively realize intelligent synthetic speech speaker confirmation, and can improve confirmation accuracy.

The invention discloses a signal modulation pattern blind identification method and system under non-cooperative communication, and the method comprises the steps: receiving a signal through an antenna, carrying out the front-end processing of a receiver, outputting a broadband intermediate frequency signal, detecting or selecting a target signal on a broadband frequency spectrum, carrying out the narrowband preprocessing according to the bandwidth and center frequency of the target signal, and carrying out the blind identification of the signal modulation pattern. Narrowband IQ data is obtained; calculating a high-order spectrum for the narrowband IQ data, and measuring a signal carrier frequency parameter based on a spectral line search method or a gravity center method of the high-order spectrum; calculating a delay spectrum or a square spectrum for the narrowband IQ data, and measuring a signal code rate parameter based on the delay spectrum or the square spectrum; extracting signal characteristic parameters used for distinguishing multiple modulation styles; and designing a multi-feature combined modulation classifier based on the signal feature parameters, and identifying various signal modulation styles through the modulation classifier. The method is comprehensive in modulation pattern recognition, low in calculation complexity, easy for engineering implementation and capable of meeting the engineering real-time application requirement.

This application discloses a human pose recognition method and apparatus. The method includes: performing human detection on a target person image frame by combining a pre-set tracker and a human detection network to obtain a target human bounding box; the pre-set tracker is a network that tracks the human body based on IOU and threshold; predicting human key points in the target human bounding box using a pre-set key point prediction network, which includes a baseline pose estimation network and an interference part classification network; and recognizing the human pose in the target person image frame based on the human key points to obtain a human pose recognition result. This application can solve the technical problem that existing technologies cannot simultaneously achieve high recognition accuracy and fast recognition speed.