150 results about "Learning Recognition" patented technology

The invention discloses an image recognition method and an image recognition device. According to one concrete embodiment, the method comprises the following steps of: obtaining an image to be recognized containing an object to be recognized; sending the image to be recognized to a server, and receiving a confidence degree parameter and marker information of a target object corresponding to the object to be recognized, wherein the confidence degree parameter and the marker information are returned by a server and are obtained through recognition on the image to be recognized; when the confidence degree parameter is greater than a confidence threshold, using the marker information of the target object as a recognition result; when the confidence degree parameter is smaller than the confidence degree threshold, obtaining marking information associated with the image to be recognized from a third party platform; and using the marking information as a recognition result. The image recognition method and the image recognition device have the advantages that the effect that the server automatic recognition and the third party marking information are combined is achieved; the recognition accuracy is improved; the third part marking information is used for training a recognition model corresponding to a machine learning recognition mode used by the server; and the training effect of the recognition model is improved.

The invention aims to solve problems of poor effect, small quantity of training samples, large difference among samples of a traditional extracted feature resorting method due to large difficulty in melanoma skin lesion segmentation and provides a multi-scale deep learning recognition system based on expansion and convolution. The system includes performing data enhancement and normalization processing on training samples and training an extracted expansion and convolution based multi-scale feature learning neural network, performing multi-threshold segmentation based on an obtained predictionprobability map, and thus implementing segmentation of a melanoma skin lesion image. Finally, the segmentation accuracy is improved.

The invention discloses a power grid monitoring alarm event identification method based on convolution and a long-term and short-term memory network, and the method comprises the steps: generating aninformation vector through historical monitoring alarm information and time marks in a power grid monitoring system, extracting event samples from the collected historical monitoring alarm information, and constructing an alarm event sample library; secondly, establishing a deep learning recognition model based on the combination of a long-term and short-term memory network and a convolutional neural network, and training the model by utilizing an alarm event sample; and finally, identifying the monitoring alarm information by using the trained deep learning model, and outputting the event category with the maximum probability as an identification result. According to the method, the excellent performance of the long-term and short-term memory network in time sequence problem processing and the excellent performance of the convolutional neural network in short text local feature mining are combined, the combined model is established, rapid identification of the power grid alarm event can be realized, the screen monitoring pressure of monitoring service personnel is effectively reduced, and the working efficiency of daily monitoring and accident exception handling is improved.

The invention provides a webshell detection method, system and device based on a cloud platform, and a storage medium. The webshell detection method comprises obtaining a white list through keyword judgement, and correcting the white list by combining a machine learning prediction result and a human judgment result. According to the invention, by means of two-engine detection and judgement and cloud unified management, the webshell detection and processing logic is optimized, machine learning recognition detection is added, two engines of characteristic rule matching and machine learning prediction are employed to effectively prevent known and unknown webshell threats, at the same time, the management cost of operational staff is optimized, and the maintenance security cost is reduced.

The invention provides a method and device for analyzing the query requirement. The method comprises the steps that the model building process is carried out, wherein pattern matching trees corresponding to different types are built in advance according to various types of pattern dictionaries composed of query patterns of various types, and query, of various types, obtained according to the pattern matching trees serves as a sample to train a machine learning recognition model; the requirement recognition process is carried out, wherein query input by a user is classified according to the pattern matching trees and the machine learning recognition model. According to the method and device for analyzing the query requirement, the query can be classified by combining the pattern matching trees and the machine learning recognition model, the pattern matching trees and the machine learning recognition model can carry out mutual feedback, and can be continuously improved, and classification accuracy of the query is improved.

An infrared ship target detection and identification method in a complex sea surface environment comprises the following steps: 1, carrying out small target image screening on an acquired infrared image through a target identification and detection method, and carrying out region separation and pre-classification on a high-speed target and a ship; 2, extracting and establishing an infrared image feature set, screening an effective feature set, establishing a classic machine learning model based on a support vector machine, and achieving target ship recognition; 3, establishing a deep learningmodel based on a convolutional neural network for ship identification in the infrared image; and 4, performing decision fusion on a deep learning recognition result and a machine learning classification result to realize more accurate target recognition.

The invention discloses an assembly robot part deep learning recognition method which comprises the following steps: firstly, obtaining an image of a to-be-recognized workpiece by using an industrialcamera, then, recognizing the image by using a YOLOv3 network, and outputting part category and position information; wherein the YOLOv3 network comprises five residual network blocks, and is characterized in that a CFENet module is introduced after each residual network block, and the CFENet module is integrated into the Darknet-53 feature extraction network for image feature extraction.. The method has the advantages that the workpiece under the normal pose can be recognized, the good detection effect is achieved on the parts under the complex conditions of camera overexposure, workpiece mutual shielding and the like, and the recognition accuracy is high.

Systems and methods for recognition learning of objects and their characteristics. The invention is a comprehensive system comprising methods and means of data access, nature assignment, and presentation. The user is able to automatically, selectively, and continually access a plurality of enhanced data sources comprising sex offender, terrorist, criminal, missing children, missing pet and other data. Following access to these data, the user may assign respective natures to object data that will affect presentation. The invention's means of automated presentation and entertaining rehearsal of object data results in recognition of objects wherein social utility is achieved. One example would be a child being able to recognize and avoid a threatening person such as a sex offender or criminal. Another example would be recognition of a missing child based on the methods of the invention. The systems operate with any means of information processing such as notebook computers, personal digital assistants, cell phones, and watches. Through the systems and methods of the invention, the objective of automated, learned recognition of socially relevant objects of interest is realized.

The invention provides a plant disease recognition and early warning method and device, and relates to the technical field of agricultural plant protection. According to the method and device, a colored image to be recognized is acquired, wherein the image to be recognized comprises corresponding spectrum-texture features; the acquired image to be recognized is input into a trained deep learning recognition model for recognition and classification, and the disease variety and the severity of diseases are obtained through the deep learning recognition model according to the spectrum-texture features of the image to be recognized; a corresponding early warning hint is given out according to the severity. According to the scheme, manpower and material resources for recognizing the plant diseases can be reduced, and the accuracy and recognition speed of plant disease recognition can be increased; in addition, according to the scheme, the severity of the plant diseases can be obtained, thecorresponding early warning hint can be given out according to the severity, and related personnel are helped to prevent and control the plant diseases in time.

The invention discloses a road disease detection method based on deep learning. The method comprises the following steps: (S1) acquiring an image of a road; (2) inputting the image of the road into adeep learning recognition model to obtain a disease detection recognition result; (S3) correcting the disease detection and identification result; and (S4) adding GPS coordinates, a road name and thetype of the road disease to the image with the identified road disease. According to the method, automatic detection of diseases can be realized, detection personnel only need to be in the maintenancevehicle to obtain pavement information, manual intervention is not needed in the whole process, and the working intensity of the personnel is greatly reduced. When the method is implemented, only thehigh-definition camera needs to be installed above the roof of the maintenance vehicle, the industrial personal computer, the router and other devices are all placed below a driving position or in atrunk, the attractiveness of the vehicle is not affected, and the vehicle is convenient to transform.

The invention discloses an AMT transmission gear-shifting meshing point position self-learning method. An AMT transmission comprises a synchronizer and does not comprise a clutch. The method includes the steps that after meshing is completed, whether the AMT transmission enters a meshing point for self cleaning is carried out or not is judged according to a set self-learning triggering condition, meshing point self cleaning is started according to the self-learning recognition condition, whether the current shifting fork position is a qualified meshing point or not is judged, and meshing point position modification is carried out according to a certain step length if the current shifting fork position is not the qualified meshing point. According to the mechanical automatic transmission gear-shifting meshing point self-learning method applied to an electric car, the automatic transmission is provided with the synchronizer structure, in the gear-shifting process, meshing point deviations caused by mechanical manufacturing are eliminated through meshing point position self learning, and smoothness of the gear shifting process is guaranteed.

The invention relates to the technical field of biological recognition, and provides a body posture recognition method based on LSTM. The body posture recognition method comprises the steps: obtainingan action video of a to-be-recognized main body; extracting action feature information in the acquired action video of the to-be-identified main body through OpenPose, wherein the action feature information at least comprises skeleton key point information; and recognizing an action specification degree corresponding to the action feature information according to the action feature information and a pre-trained and generated body posture recognition model, wherein the body posture recognition model is a target neural network model generated according to a preset standard action, and the target neural network model is generated by training according to standard action feature information arranged according to a time sequence. According to the body posture recognition method, video actionsdo not need to be cut into isolated features to be recognized, and learning recognition is carried out through cooperation with the neural network model, and the body posture recognition process is rapid and accurate, and the user experience is improved.

The invention discloses a distributed optical fiber sensor vibration signal classification method and an identification classification system. The method comprises the following steps: firstly, usingan optical fiber sensing system for obtaining vibration signals acting on an optical cable; preprocessing the optical fiber vibration signal; calculating the short-time energy and the short-time zero-crossing rate of the optical fiber vibration signal; setting double thresholds of short-time energy and a short-time zero-crossing rate, if the double thresholds exceed the thresholds, extracting an effective data segment, and judging the effective data segment as a disturbance event; drawing a spectrogram on the time-frequency domain of the optical fiber vibration signal; extracting a Mel frequency cepstrum coefficient of the optical fiber vibration signal; establishing a deep learning recognition model based on the Mel frequency cepstrum coefficient and the time-frequency domain spectrogramof the disturbance event signal; and matching with a deep learning recognition model based on two characteristics of a spectrogram and a Mel frequency cepstrum coefficient in a vibration signal time-frequency domain, and judging the type of the optical fiber vibration signal. According to the invention, feature extraction and accurate identification and classification of the optical fiber sensingsignals are realized, and the technical problem of low identification and classification accuracy of the optical fiber sensing intrusion signals is solved.

The invention discloses a radar radiation source deep learning recognition method based on a non-fingerprint signal eliminator, an original radar radiation source signal comprises a fingerprint feature part and a non-fingerprint feature part, and the recognition accuracy of a radar radiation source can be improved to a great extent through extraction of fingerprint features and suppression and elimination of non-fingerprint features. According to the invention, the deep learning network is used as the radiation source signal feature extractor, and the proposed non-fingerprint signal eliminatoris combined to extract the fingerprint information of the radiation source signal, eliminate and suppress the non-fingerprint signal, and improve the radar radiation source recognition effect.

The invention discloses an equipment natural vibration mode self-learning recognition method based on online vibration data, which is characterized by comprising the steps of carrying out primary denoising on equipment vibration signals x(n) by using a wavelet packet transform algorithm so as to acquire denoised signals f1<^>(n); carrying out secondary denoising on the signals f1<^>(n) by using singular value decomposition so as to acquire secondarily denoised vibration signals f2<^>(n); carrying out spectral analysis on the denoised vibration signals f2<^>(n) by using windowed discrete Fourier algorithm, and calculating to acquire equipment vibration spectrum; training by using a self-learning algorithm to acquire the equipment vibration spectrum, and finally acquiring the natural characteristic frequency and the amplitude of equipment. The equipment natural vibration mode self-learning recognition method based on the online vibration data has the beneficial effect that the characteristic frequency of components can be recognized accurately when a research object of a complicated system.

The invention provides a machine learning recognition method based on deep learning. Multiple times of different learning training can be performed on a machine learning model f1 with a certain quantity of known classes of multimedia data samples in different contrast sample input arranging orders, the machine learning model f1 obtained by learning is used for recognizing multimedia data classes and selects a convolutional neural network model or a fully connected neural network model, dependence on a huge number of training samples is reduced substantially, the multimedia data classes not subjected to learning training can be expanded conveniently for class recognition, the problem that practical applicability and generality are limited due to dependence of the conventional machine learning method for multimedia data classification on the huge number of training samples and failure in direct classification recognition on the classes not subjected to learning training is solved well, and the machine learning recognition method can be applied to more specific multimedia data classification occasions more extensively and more effectively.

The invention relates to the field of handwriting character recognition, and particularly relates to a Chinese and Japanese handwritten character recognition method. By using the four recognition methods, such as offline recognition, online recognition, language context recognition and depth learning recognition, handwritten characters are recognized to form a variety of candidate sets. All candidate sets are sorted according to a set weight ratio, and a recognition result is output. The recognition accuracy of Chinese and Japanese handwritten characters is improved. Depth learning and existing text recognition methods are combined together to complement each other and take advantages from each other. An experimental result shows that the recognition accuracy of Chinese and Japanese handwritten characters is greatly improved, and the method has a better beneficial effect compared with the prior art.

The invention discloses a machine learning recognition and process parameter optimization method for abrasive belt abrasion. The method comprises the following steps: S1, making a training set and a test set required by convolutional neural network training; S2, training a machine learning classification model based on a neural network; S3, an abrasive particle abrasion image on the surface of theabrasive belt is obtained; S4, identifying and distinguishing a wear region, an unworn region and a blocked region in the abrasive belt wear image through a machine learning classification model; S5,calculating the area and the area rate of each area; and S6, judging whether the process parameters are reasonable or not according to the area ratio of each part, and optimizing the existing parameters by adopting a basic particle swarm optimization algorithm. According to the method, the abrasion condition is identified through the model obtained through machine learning, and the process parameter optimization direction is predicted. The abrasive belt abrasion measuring and calculating process is simplified, intelligent image detection of the abrasive belt abrasion degree is achieved, the abrasive belt abrasion condition can be accurately, rapidly and conveniently measured, and good measuring precision is achieved.

The invention discloses a nested entity data identification method and device, and electronice equipment, and relates to the technical field of data identification. The method comprises the steps of permutating and combining seed entity vocabularies of different entity categories to generate a short text data set; defining at least one entity category label for a short text in the short text dataset, and index information of starting and ending of a sub-text, corresponding to each entity category label, in the short text; training a deep learning recognition model by using the defined short text data set as a training set; and recognizing the nested entity data by using the recognition model which is trained to reach the standard. According to the method, the entity labeling information is defined for the statement according to the start and end indexes and the entity category label, so that the multi-nested entity content labeling is simpler to implement, the labeling process and workload of nested entity recognition are optimized, the time cost and the labor cost are saved. And thus, the identification efficiency and accuracy of the nested entity data can be improved.

The invention provides a bearing fault identification method and device, computer equipment and a storage medium. The method comprises the following steps: obtaining a vibration signal of a bearing tobe tested during the working process; converting the vibration signal into an acceleration envelope frequency spectrogram of the bearing to be tested; and inputting the acceleration envelope frequency spectrogram into a pre-trained convolutional neural network to extract image features of the acceleration envelope frequency spectrogram, and identifying fault probability of the bearing to be tested based on the image features. By inputting the envelope frequency spectrogram into the multi-layer convolutional neural network, the method can judge the fault without experience of workers or without carrying out operation of feature extraction feature selection and the like; and through combination of acceleration envelope fault diagnosis and a deep learning recognition method, accuracy and feasibility of fault diagnosis are improved.

The invention relates to a finger vein machine learning recognition method and device based on terrain concave-convex characteristics. The method comprises the steps of performing size normalization processing on a registered finger vein image and a verified finger vein image; performing image enhancement processing on the registered finger vein image and the verified finger vein image; obtainingterrain concave-convex features from the registered finger vein image and the verified finger vein image based on a digital elevation model, and extracting registration features and verification features; translation and rotation calibration correction are carried out on the registration features and the verification features, and sliding window similarity calculation is carried out on an overlapping region of the vein features after calibration correction; optimizing the feature extraction technical parameters and the recognition technical parameters of the finger vein based on the calibration correction parameters and the sliding window similarity parameters; the device comprises a normalization processing module, an image enhancement module, a feature extraction module, a parameter calculation module and an optimization module. According to the method, the technical capability of finger vein recognition and the adaptability to different image qualities are improved.

The invention provides a positioning method, a controller and a removing device for removing the top center of cotton, wherein, the method comprises the following steps: obtaining color image data anddepth image data of a cotton plant to be removed from the top center; inputting the color image data into a pre-trained depth learning recognition network model to recognize the cotton top center andobtain pixel coordinates of the cotton top center in the color image, generating the depth learning recognition network model by pre-training according to a plurality of top-core cotton plant samplesto be removed, determining three-dimensional spatial position information of the cotton top center according to pixel coordinates of the cotton top center in the color image and the depth image data;wherein the three-dimensional spatial position information is used for positioning the positioning mechanism of cotton top center removal. The technical scheme improves the precision and efficiency of cotton top center removal.

The invention discloses a sweeping control method and device, a sweeping robot and a computer readable storage medium, and the method comprises the following steps that the image information of a to-be-swept area is obtained, the deep learning recognition of the image information is carried out, and the room region information in the image information is obtained; according to the room area information, a cleanable area in the to-be-cleaned area is determined, and the cleanable area is a passable room area capable of being entered by a sweeping robot in the to-be-cleaned area; and a target area is determined according to the sweeping motion trail, the sweeping robot is controlled to conduct supplementary sweeping on the target area, and the target area is an area which is not swept and exists in the cleanable area. The sweeping robot can conduct supplementary sweeping on the target area which is not swept in the sweeping process, and the sweeping efficiency and the sweeping effect of the sweeping robot are improved.

The invention discloses an automatic extraction and intelligent scoring system for handwritten Chinese characters or components and strokes. The automatic extraction and intelligent scoring system comprises an automatic extraction module for calligraphy practicing book images, an automatic extraction module for single Chinese characters, a Chinese character recognition module and a Chinese character scoring module. The automatic extraction module for calligraphy practicing books carries out preprocessing, image capture, correction, error reporting and format judgment on uploaded handwritten Chinese character photos, and the accuracy and the automation effect of image extraction are improved through the complete process. The automatic extraction module for single Chinese characters comprises a special processing method for pencil character extraction, so that a pencil character image can be effectively extracted; the established deep learning recognition model can quickly recognize handwritten Chinese characters, including space recognition; and Chinese character scoring adopts a structure+content comprehensive scoring method, the structure is the length and width values of Chinesecharacters, and the content is determined through cosine similarity. According to the method, the handwritten character pictures which are randomly uploaded can be automatically extracted and scored,and the method is suitable for low-age students who begin to learn Chinese characters and is greatly helpful for daily calligraphy practicing of calligraphy enthusiasts.

The invention discloses a sleep apnea monitoring system based on a millimeter-wave radar. The sleep apnea monitoring system comprises a millimeter-wave radar module, a signal processing module, an MCUprocessor, a deep learning recognition system and a rear-end display system. The system is characterized in that the millimeter-wave radar module is connected with the MUC processor, transmits linearfrequency modulation continuous wave signals in the working process according to the Doppler radar detection principle, receives reflected echo signals and transmits the signals to the signal processing module; the signal processing module is connected with the MCU, performs signal processing such as demodulation and amplification on echo signals to obtain vital sign monitoring data of a user inthe sleeping process, and transmits the signals to the deep learning recognition system; and the deep learning recognition system filters the received signals, and filters noise and interference through a band-pass filter which is generated by cascading a low-pass digital filter and a high-pass digital filter and has the pass band response of 2-50Hz according to the frequency of the signals.

The invention discloses an optical orbital angular momentum machine learning recognition method based on a turbulence effect, and solves the problems that a huge training sample is needed when an existing machine learning method is used for recognizing an orbital angular momentum mode, and synchronous recognition of various turbulence environments is difficult to realize. The method comprises thesteps of obtaining a training sample under a numerical simulation condition; training a support vector machine multi-classification model with more optimized parameters by using a genetic algorithm and the feature vectors; performing orbital angular momentum recognition by using the trained support vector machine multi-classification model; and grouping joint identification is carried out on the images with a large identification range. According to the invention, a physical mechanism and machine learning are combined for optical orbital angular momentum identification. According to the method, the orbital angular momentum mode number of the vortex beam in various atmospheric turbulence environments can be effectively recognized, the accuracy is far higher than that of a traditional optical detection method, compared with a machine learning method based on an image algorithm, training samples can be effectively reduced, and the learning difficulty is lower. The optical orbital angularmomentum machine learning recognition method is used for free space optical communication.

There are provided a characteristic obtaining unit configured to obtain a subject characteristic including a characteristic of a subject, an image processing unit configured to generate a duplicate subject image by performing an image process to an image of the subject according to the subject characteristic obtained by the characteristic obtaining unit, and a learning unit configured to learn a matching dictionary by using the duplicate subject image generated by the image processing unit. Thus, it is possible to reduce the number of subject images necessary for the learning.

The invention relates to an infant quilt kicking behavior recognition method and device, computer equipment and a storage medium, and the method comprises the following steps: obtaining a real-time image of an infant so as to obtain a to-be-recognized image; recognizing the to-be-recognized image by adopting a deep learning recognition model to obtain a recognition result; outputting the recognition result to the terminal to prompt the terminal, wherein the deep learning recognition model is obtained by training a deep learning convolutional neural network by taking a plurality of infant quiltkicking behavior images and infant non-quilt-kicking behavior images as a sample set. According to the invention, the deep learning recognition model adopts a three-layer network for obtaining a candidate box of the whole body area of the baby; the classification network divides the candidate box into a plurality of local areas and maps the local areas to the score feature map to obtain a relatedfeature map, and the probability of each category is calculated according to the related feature map; the deep learning recognition model is adopted to recognize the images to obtain the categories,the accuracy of the whole infant quilt kicking behavior recognition process is improved, and the recognition complexity is reduced.

The invention belongs to the technical field of wine information management, and provides a wine label identification method and device, a wine product information management method and device, equipment and a storage medium The wine label identification method comprises the following steps of: obtaining a wine product image; performing OCR recognition on the wine product image to obtain characters contained in the wine product image; performing deep learning recognition on the wine product image according to a preset deep learning recognition mode to obtain image features contained in the wine product image; screening out a target wine label matched with the characters and image features from a preset wine label database according to the characters and image features, and taking the target wine label as a wine label corresponding to the wine product image. By means of deep learning, the defect of wine label character recognition in a complex environment where wine is located by OCR character recognition is overcome; the defect of wine label image recognition by deep learning is overcome by fully utilizing OCR character recognition; the accuracy and efficiency of wine label recognition are improved;and the automation efficiency of wine product information management is improved.

The invention discloses a construction method and device and a recognition method and a recognition device for a deep learning recognition model of an inclined license plate. The construction method comprises the steps: determining a license plate coordinate from a collected license plate image, and calculating an affine parameter; constructing a deep learning network framework for identifying theinclined license plate; and training a positioning network by using the collected data set, and training a license plate character recognition network through the trained parameter model and the trained license plate data set. The invention provides a recognition network framework based on a deep learning method for inclined license plate recognition, and the technical effect of greatly improvingthe recognition precision of the inclined license plate can be achieved.