694 results about "Feature extraction algorithm" patented technology

A method of automatically labeling of unlabeled text data can be practiced independent of human intervention, but that does not preclude manual intervention. The method can be used to extract relevant features of unlabeled text data for a keyword search. The method of automated labeling of unlabeled text data uses a document collection as a reference answer set. Members of the answer set are converted to vectors representing centroids of unknown groups of unlabeled text data. Unlabeled text data are clustered relative to the centroids by a nearest neighbor algorithm and the ID of the relevant answer is assigned to all documents in the cluster. At this point in the process, a supervised machine learning algorithm is trained on labeled data, and a classifier for assigning labels to new text data is output. Alternatively, a feature extraction algorithm may be run on classes generated by the step of clustering, and search features output which index the unlabeled text data.

A method for controlling digit movement of a prosthesis based on residual forelimb volitional pressures is provided. A filter is provided for discriminating from the distributed pressures to provide reliable digit control. The extracted commands can be used to control actuators in a hand prosthesis to restore biomimetic finger control to the user. Feature extraction algorithms can be based on signal amplitudes (peak amplitude or root mean square). Training signal validation can be based on cepstral encoding. The filter is trained under user supervision.

The invention relates to an electroencephalogram feature extracting method based on brain function network adjacent matrix decomposition. The current motion image electroencephalogram signal feature extraction algorithm mostly focuses on partially activating the qualitative and quantitative analysis of brain areas, and ignores the interrelation of the bran areas and the overall coordination. In light of a brain function network, and on the basis of complex brain network theory based on atlas analysis, the method comprises the steps of: firstly, establishing the brain function network through a multi-channel motion image electroencephalogram signal, secondly, carrying out singular value decomposition on the network adjacent matrix, thirdly, identifying a group of feature parameters based on the singular value obtained by the decomposition for showing the feature vector of the electroencephalogram signal, and fourthly, inputting the feature vector into a classifier of a supporting vector machine to complete the classification and identification of various motion image tasks. The method has a wide application prospect in the identification of a motion image task in the field of brain-machine interfaces.

The invention discloses an image stitching method based on unmanned aerial vehicle POS information and image SURF feature combination and relates to the digital image processing field, the GIS field,the survey field and other relevant fields. According to the method, first, geometric correction is performed on images; second, geographic coordinates of four corners of each image are calculated; based on the geographic coordinates of the first image, a position relation of homonymy matching target positions is obtained by extracting SURF features of adjacent image overlapping regions, and therefore the geographic coordinates of the following images are sequentially corrected; and last, an adaptive gradual-in-gradual-out fusion algorithm is adopted, a panoramic image with a good visual effect is obtained, and good stitching of the images is completed. Through the method, an image feature extraction algorithm and the geographic coordinates of the images are combined, and stitching efficiency and the visual effect are both improved greatly compared with traditional feature extraction and stitching algorithms; and the image obtained after stitching contains geographic information, so that the image has certain practical value.

The invention discloses a multiple dimensioned convolution neural network-based real time human body abnormal behavior identification method. A convolution neural network is used for replacing a conventional feature extraction algorithm, and the convolution neural network is improved so as to satisfy requirements for human body behavior classification; specifically, three dimensional convolution, three dimensional down-sampling, NIN, three dimensional pyramid structures are added; human body abnormal behavior feature extraction capability of the convolution neural network is enabled to be increased; training operation is performed in a specific video set, features with great classification capacity can be obtained, robustness and accuracy of a whole identification algorithm can be improved, GPU speed is increased so as to satisfy requirements for practical application, and therefore multi-channel videos can be monitored in real time.

The invention relates to a wearable human body feature information collecting and monitoring system which comprises at least one wearable wireless human body feature data collecting and transmitting module, at least one intelligent terminal and at least one open type cloud end processing platform. The wearable wireless human body feature data collecting and transmitting module is used for collecting the heartbeat rate and the respiratory rate of a human body. The intelligent terminal is used for being in communication with the wearable wireless human body feature data collecting and transmitting module and controlling the data collecting and storing of the wearable wireless human body feature data collecting and transmitting module. The open type cloud processing platform is used for being in communication with the intelligent terminal, processing data of the wearable wireless human body feature data collecting and transmitting module based on a data analyzing and feature extracting algorithm, automatically generating a report containing an analyzing result and then automatically informing a user. According to the wearable human body feature information collecting and monitoring system, due to the fact that massive human body features are collected and extracted, features of various diseases in different stages can be found effectively, and the user is prompted to pay attention to the own body changes through the comparison with the daily human body features.

A mobile electron device is used for positioning and map building in an unknown environment. The mobile electron device comprises: a first sensor, a second sensor, a feature point extraction unit, a matching unit, and a positioning and map building unit, wherein the first sensor is used for acquiring a first image of the current scene with the current position and orientation; the second sensor is used for acquiring a second image of the current scene with the current position and orientation; the feature point extraction unit is used for extracting feature points of the first image and feature points of the second image by utilization of a feature extraction algorithm; the matching unit is used for matching the feature points of the current scene and feature points of a previous scene, and acquiring a transformation matrix from the previous scene to the current scene; and the positioning and map building unit is used for, based on the transformation matrix and the matched feature points, determining changes on position and orientation of the mobile electron device relative to the previous position and orientation so as to perform positioning, and is also used for combining the images of the current scene with a known map.

The invention discloses a real time panorama video splicing method based on ORB characteristics. The real time panorama video splicing method based on the ORB characteristics comprises the following steps: acquisition of multipath synchronized video data is started; pretreatment is carried out on images in various paths at a same moment, and color images are changed into gray scale images of 256 levels, and a de-noising processing is carried out on the images through employing a Gaussian filter; the ORB feature extraction algorithm is employed to carry out feature point extraction on the images in the various paths at the same moment, and ORB characteristic vectors of the feature points are calculated; through the adoption of the nearest neighborhood matching method and the RANSAC (random sample consensus) matching method to determine a homography matrix array between corresponding frames of the synchronized videos; frame scene splicing is carried out according to the homography matrix array; and finally spliced videos are output. The real time panorama video splicing method based on ORB characteristics and the apparatus are advantageous in that: the feature extraction speed and the coupling effect are improved in the image splicing process.

The invention discloses a pavement crack identification and feature extraction algorithm and system. The algorithm comprises the following steps: reading a pavement three-dimensional data matrix; performing de-noising processing; identifying a pavement crack to obtain a final crack image; extracting a crack seed point from the crack image and performing region growing to obtain a crack binary image; calibrating the crack with a plurality of identical squares, and framing all the squares with a rectangular frame in such a manner that the area of the rectangular frame is smallest; classifying the crack according to an included angle between the diagonal line of the rectangular frame and a horizontal direction: the crack of which the included angle is within [0, pi / 4] is a transverse crack, otherwise the crack is a longitudinal crack; calculating the length of the diagonal line of the rectangular frame as the length of the crack, and calculating the area of the rectangular frame as the area of a crack region. The crack can be identified and the type of a three-dimensional linear crack can be judged only by inputting the acquired three-dimensional data matrix in order to finish feature extraction; the algorithm and the system have the advantages of easiness in calculation, short running time and accurate detection result.

The invention relates to an inertia / vision integrated navigation method for a deep-space detection patrolling device. The method comprises the steps that a state model of inertia / vision integrated navigation of the patrolling device is established according to a serial inertial navigation mechanical arrangement in a fixed coordinate system of a planet; a binocular vision camera is used for shooting a surrounding environment of the patrolling device; a stereoimage sequence is obtained; pixel coordinates of characteristic points are obtained by an image characteristic extraction algorithm; a measurement model by taking the pixel coordinates of the characteristic points as measurement vectors is established; and a position, a speed and a state of the deep-space detection patrolling device are estimated by using Unscented kalman filter. The method can effectively correct a position error of inertial navigation and improve the navigation accuracy, is very applicable to autonomous navigation of the deep-space detection patrolling device, belongs to the technical field of space navigation, can provide high-precision navigational parameters for the deep-space detection patrolling device, and can provide references for the design of an autonomous navigation system of the deep-space detection patrolling device.

The invention discloses a mechanical equipment residual service life prediction method and system. The method comprises the steps that a time convolution network serves as a feature extraction algorithm, a long-term and short-term memory network serves as a regression prediction algorithm, a deep neural network life prediction model is constructed, and the deep neural network life prediction modelis trained; according to the model of the tested equipment and the data acquisition time sequence, constructing the acquired real-time operation data of the tested equipment into a service life prediction data set with time sequence characteristics; and carrying out prediction processing on the life prediction data set by using the deep neural network life prediction model to obtain the residualservice life of the tested equipment. A state monitoring signal output by a sensor for monitoring mechanical equipment has the characteristics of a time sequence; a time convolution network and a longshort-term memory network are combined, a deep neural network life prediction model is established for RUL prediction of mechanical equipment, the problems of over-fitting and gradient disappearanceexisting in a common deep neural network model are solved, and the prediction accuracy is improved.

The invention relates to a pedestrian detecting method based on improved HOG feature and PCA (Principal Component Analysis). The method comprises the following steps: extracting a sample feature by using a feature extracting algorithm for HOG feature cascaded PCA in a concentrated area of gradient information of a pedestrian in a sample image; training a SVM classifier by utilizing the feature extracted from the sample; and utilizing a feature extracting method to extract a feature vector for a detected sample and utilizing the trained and obtained SVM classifier to detect the pedestrian. The invention provides a pedestrian detecting method based on improved HOG feature and PCA, with the capabilities of efficiently reducing the training speed and reducing the false detecting rate and missing report rate.

The method discloses a robust feature tracking and stereoscopic vision positioning technology based on image processing and machine vision. The technology can integrate inertial information and visual information and achieve reliable stereoscopic vision positioning under camera waggling conditions and outdoor light conditions. Images are collected through a binocular video camera in real time, and rotation information of the camera is collected with an inertial measurement unit. Feature points in the images are extracted with a feature extraction algorithm, and the feature points of the left image and the feature points of the right images are matched stereoscopically. The inertial information is combined and the inertia and the KLT algorithm are integrated to track the feature points, so that the reliability of the feature tracking is promoted. Three-dimensional information of the feature points is restored according to the double vision geometric principle. Motion parameters of the camera are obtained through position information of the feature points with the Gaussian and Newton iteration method. The accuracy of visual positioning is further promoted with the RANSIC algorithm. The whole process is iterated continuously, and thus real-time calculation of the posture and the position of the camera is achieved.

The invention relates to a wearable heart sound and electrocardio feature information collecting and monitoring system. The wearable heart sound and electrocardio feature information collecting and monitoring system comprises an electrocardio and heart sound signal collecting device, a terminal device and a remote data center. The electrocardio and heart sound signal collecting device is used for collecting electrocardio and heart sound signals and carrying out data transmission through a wireless network. The terminal device is used for receiving the electrocardio and heart sound signals collected by the electrocardio and heart sound signal collecting device and transmitted through the wireless network, converting the electrocardio and heart sound signals into waveforms for storage and uploading the waveforms to the remote data center. The remote data center is used for receiving the electrocardio and heart sound signals and the waveforms of the electrocardio and heart sound signals sent by the terminal device for filing, analyzing data through a data analyzing and feature extracting algorithm, then generating a report and sending the report to a user. According to the wearable heart sound and electrocardio feature information collecting and monitoring system, by means of massive human body feature collection and feature extraction, features of various heart diseases in all stages can be effectively found, the features are compared with daily heart features of the user, and the user is reminded to pay attention to the changes of the heart.

The invention provides an electroencephalogram signal characteristic extracting method. Network average route lengths and clustering coefficients are calculated through wavelet reconstruction, windowing horizontal visibility map complex network conversion and complex network analysis. The average route lengths and clustering coefficients composed of electroencephalogram signals are calculated to achieve characteristic analysis of electroencephalogram signals and chaotic time sequence signals of the electroencephalogram signals of different rhythms. The electroencephalogram signal characteristic extracting method has the advantages that one-dimensional chaotic time sequences are converted into complex networks, according to analysis of network characteristic parameters, fractal characters of the electroencephalogram signals are revealed, the complex non-linearity signals of the electroencephalogram signals are depicted from a brand new angle. The electroencephalogram signal characteristics can be applied to automatic diagnosis of mental disease and a characteristic identifying module of a brain-machine port system. The electroencephalogram signal characteristic extracting method can effectively distinguish the electroencephalogram signals of an epilepsia attach stage and an epilepsia non-attach stage, the equation p<0.1 is met after Mann-Whitney detection, and the electroencephalogram signal characteristic extracting method can be applied to epilepsia electroencephalogram automatic identification.

The invention discloses a rapid facial expression recognition method based on an ELM self-encoding algorithm. The rapid facial expression recognition method comprises the steps of: step 1, training an Adaboost-based face region detection classifier and performing face detection; step 2, preprocessing a detected face region, including clipping, size normalization and histogram equalization processing; step 3, adopting an ELM-AE algorithm based on an auto-encoder and an extreme learning machine as a feature extraction algorithm, and performing feature extraction on a facial expression image after preprocessing; step 4, and constructing a facial expression classifier based on the extreme learning machine, inputting vectors of feature extraction into the facial expression classifier, and regarding an output result as a mood of a face. The rapid facial expression recognition method can extract primary information and perform dimensionality reduction more quickly and efficiently. When in facial expression recognition classification, the ELM just needs to adjust a parameter of a neuron, the recognition is short in operating time and high in accuracy rate, and the rapid facial expression recognition method is efficient and rapid in learning speed.

The invention discloses an upper limb rehabilitation system based on biological signals. The upper limb rehabilitation system based on the biological signals is characterized in that a brain electric cap is attached to the surface of a brain, brain electric signals can be sensed through the brain electric cap, features of the brain electric signals are extracted and classified sequentially through a CSP feature extraction algorithm and a classifier with an adaptive LDA classification algorithm after amplification, filtering and noise reduction are performed on the brain electric signals, and then the brain electric signals are translated into drive instructions used to control drive motor equipment and pneumatic tendon auxiliary equipment, a system central processing unit fuses and sorts the drive instructions and motion state data collected by a motion state collector, and then feeds the drive instructions and the motion state data, which are fused and sorted, back to the drive motor equipment, the pneumatic tendon auxiliary equipment and a mobile terminal, and therefore a mechanical arm is driven to perform rehabilitation exercise. Work personnel remotely monitor motion and rehabilitation states of patients through the mobile terminal, and the patients also can invoke rehabilitation games in a display platform through a voice input and output device so as to perform auxiliary rehabilitation training. Accordingly, the upper limb rehabilitation system based on the biological signals improves training effectiveness of the patients, and simultaneously guarantees safety and stability in the training.

The invention discloses a rapid image registration implementation method based on space sparsity and SIFT feature extraction and mainly aims at solving the problem of feature mismatching due to frequently extracted feature points unstable in smooth region and texture region in a classical SIFT feature extraction algorithm. The rapid image registration implementation method based on the space sparsity and the SIFT feature extraction comprises the steps of 1) extracting the sparse regions of a reference image and an image to be matched, respectively, 2) extracting SIFT feature points from the sparse regions of the reference image and the image to be matched, 3) performing rough matching on the set of SIFT feature points extracted from the reference image and the image to be matched, 4) filtering out mismatch in the rough matching result by use of a random consistency estimation algorithm, and 5) realizing registration on two SAR images by use of an affine transformation coefficient obtained through affine transformation by utilizing the final matched point pairs of the reference image and the image to be matched. The rapid image registration implementation method based on the space sparsity and the SIFT feature extraction is capable of improving the refrigeration efficiency under the premise of guaranteeing the accuracy of the classical SIFT feature refrigeration algorithm and can be applied to the refrigeration processing of the SAR images.

The invention discloses a detection and matching mechanism for recognition of handwritten letters using WiFi signals. The invention mainly makes use of the characteristics of wireless signal multipath propagation, and acquires CSI (Channel State Information) data that carry information reflecting external environmental characteristics, including hand movements when letters are written. The CSI data after de-noising (via Butterworth filtering and principal component analysis filtering) show that in the stage of writing letters, the CSI data fluctuate significantly, and the CSI data are relatively stable in the stage without handwriting, and on the basis of the characteristics, the invention provides a waveform feature extraction algorithm to detect and extract waveforms only including information on 26 handwritten letters. Finally, the invention provides a solution combining feature matching and context error correction, takes the waveform as a matching feature, carries out machine learning classification for the extracted 26 feature waveforms, matches the waveforms by using a DTW (dynamic time warping), and improves the recognition accuracy of the system through the context error correction algorithm.

The invention discloses an improved adaptive genetic algorithm based neural network image classification method. The method comprises the following steps: extracting texture features of sample images by adopting a gray-level co-occurrence matrix based feature extraction algorithm to obtain texture features of a training sample and a test sample; by taking the texture feature of the training sample as an input of an RBF neural network, training the RBF neural network with a genetic optimization based neural network learning method to generate a trained RBF neural network; and inputting the texture feature of the test sample into the trained RBF neural network, and performing an image classification test. For the deficiency that a k-means clustering algorithm and other algorithms are sensitive to initial value selection, the method can better avoid the ''prematurity'' convergence of a genetic algorithm, simplify the network structure of a neural network classifier and improve the generalization capability of the network and the correct classification efficiency of the images.

A method for grading subjective questions includes sentence preprocessing, feature extraction, feature fusion, similarity calculation and comprehensive scoring. Wherein, the sentence preprocessing isused for clause, word segmentation, keyword detection, part-of-speech tagging and sentence emotion analysis of the target paragraph. The feature extraction algorithm is used for extracting a word vector, a sentence vector, a word structure and a syntactic structure. The feature fusion is used for fusing a target paragraph containing M sentences into a contrast template containing N templates (N (M): the similarity calculation is used for calculating word similarity and sentence similarity. The comprehensive score is used for constructing a weight model according to the word similarity, sentence similarity, word structure similarity, syntactic structure similarity, keyword score and affective score between the student answers and the comparative template, and then grading the student answers. The invention adapts to the scoring requirements of subjective questions of various disciplines, and good scoring effect can be obtained through training of a small number of samples.

The invention discloses an infrared thermal image defect feature recognition method based on dynamic multi-objective optimization. The transient thermal response of a pixel point is selected by changing the step length of a thermal image sequence; FCM is used for classification. obtaining the category of the transient thermal response of each pixel point; the pixel value similarity of each type ofpixel points and the same type of pixel points is considered; the difference with different types of pixel points is realized; constructing a corresponding multi-objective function; meanwhile, afterthe environment is changed each time; Prediction mechanism, a guiding direction is provided for population evolution; a multi-objective optimization algorithm is helped to quickly respond to the new change; The method has the advantages that the dimension reduction result of the thermal image sequence is acquired by the aid of the feature extraction algorithm, the defect features of the infrared thermal image are extracted by the aid of the pulse coupling neural network, accordingly, accurate selection of transient heat representations (temperature points) can be realized, the accuracy of defect feature extraction can be guaranteed, and computing consumption of the transient heat representations for acquiring various types of information in dynamic environments can be reduced.

Systems and methods for detecting gases, airborne compounds, and other particulates, are provided. The system detects materials of interest, including but not limited to, volatile organic compounds, aerosols, particulates, and biological and other pathogens in a three dimensional volume over an area of interest. The system detects the concentration of analytes of interest in the presence of atmospheric contaminants. Data points form a three-dimensional “point cloud” to which particle swarm optimization and feature extraction algorithms are applied, providing leak detection, mapping of chemical plumes, and short-term and long-term flux measurements, among other functions.

The invention discloses a robot barrier identification method based on a gradient histogram and a support vector machine. The method comprises two parts of a characteristic extraction stage and a target identification stage, for the characteristic extraction stage, a characteristic extraction algorithm of a power transmission line barrier of a principal component gradient histogram is proposed, the characteristic that typical barriers have different structures and space layouts is utilized, statistics characteristics of common online barriers are calculated, characteristic extraction is carried out by utilizing an HOG algorithm, characteristic points irrelevant to illumination and scale change can be acquired, interference can be effectively removed, moreover, dimension reduction operation for acquired characteristic vectors can be realized by utilizing main component analysis to acquire the principal component gradient histogram, irrelevant characteristics can be effectively reduced, operand is reduced, least characteristics are utilized to establish a characteristic set of the corresponding barriers, and excellent support is provided for next target identification; for the target identification stage, the linearity support vector machine is utilized for identification, and the excellent identification effect is acquired.

The invention provides a business quality analysis method and system under a micro-service architecture, and the method comprises the steps: obtaining all micro-service call record data in real time,carrying out the analysis of the call record data, and obtaining the topological relation between a business and all micro-services; obtaining an occurrence probability index of each micro-service inthe business based on the topological relation, and obtaining the weight of each micro-service in the business based on the probability index; and calculating a performance index of the micro-service,and obtaining a real-time health degree index of the service based on the weight of the micro-service in the service and the performance index of the micro-service. According to the influence dependence degree of each micro-service on the business link and the performance index of each micro-service, the efficient and accurate real-time monitoring and analysis capability of the quality of the business process is realized, the cluster aggregation identification is realized through track slicing and vector conversion and through a feature extraction algorithm and a secondary similarity algorithm, and thus a large amount of manpower is prevented from being invested to carry out the combing process.

The present invention relates to a recommender system and method comprising a first extractor (S200) for applying a first feature extraction algorithm to extract first features characterizing a content of a data input (e.g. webpage) processed by a first application (e.g. Internet browser) running on the system, and a second extractor (S100) for applying a second feature extraction algorithm to extract second features characterizing a content of a database of a second application (e.g. personal TV or movie access) running on the system. Additionally, a comparator (S300) is provided for comparing the first and second features to identify matching items used for the recommendation.

The invention discloses a double-camera recognition method for a human face of a living body, and the method comprises the steps: obtaining a black-and-white image and an infrared image through a black-and-white camera of the configured two cameras; obtaining a color image through a color camera; extracting two-dimensional state organ feature points through combining the human face parts in the black-and-white image and the infrared image with a feature extraction algorithm; forming a three-dimensional state organ feature points, recognizing the features of the human face of the living body through a human face recognition algorithm, and judging whether the human face image is the image of the human face of the living body or not. The invention also discloses a double-camera recognition device for the human face of the living body. The color image and the near infrared image are collected through the two cameras, and a background removing image is obtained through background difference calculation, i.e., the human face part. The feature points of an organ are extracted through the feature extraction algorithm, and the features of the human face of the living body are recognized through the human face feature recognition algorithm, so as to judge whether the human face image is the image of the human face of the living body or not. The method is high in reliability, is convenient and practical, and is low in implementation cost.

The invention provides an information processing feature extracting method. The method comprises the following steps of: establishing a feature item set which comprises the original feature items; calculating feature items in each class and the weight of the class; sorting the calculated weights of the feature items from high to low in each class and extracting the first K feature items; and combining the feature items extracted from the classes and unifying a feature space. An information processing feature extracting algorithm is provided, and a method for extracting spatial features by using word frequency is implemented. In the process of selecting a feature extracting algorithm, the time and space complexity and the feature extracting effect of the algorithm are comprehensively considered, and the feature extracting algorithm which is simple and practicable is designed and implemented.

The invention discloses a stereoscopic vision positioning method capable of realizing accurate positioning in a dynamic environment. An environment image is acquired in real time by a binocular camera. Characteristic points in the image are extracted by using a characteristic extraction algorithm. Four images acquired by the binocular camera at two adjacent moments are stereoscopically matched. The three-dimensional information of the characteristic points is recovered by using a double-vision geometric principle. Matching accuracy is improved in a loop-back matching mode. The characteristic points on a dynamic object are removed by using a field flow method. A field flow method effect is improved by considering error causing factors and calculating a covariance matrix. Motion parameters of a robot are obtained by using a Gaussian Newton iteration method according to the position information of the characteristic points. Vision positioning accuracy is further improved by using an RANSIC algorithm. The whole process is continuously iterative, and the gesture and position of the robot are calculated in real time.