99results about How to "Realize classification recognition" patented technology

The invention provides a detection system and detection method for low-altitude slow small targets. The detection method includes the following steps that: electromagnetic waves are emitted through adetection radar, electromagnetic waves reflected by a target are received, and the point trace and flight trace information of the target is obtained; unmanned aerial vehicle frequency band electromagnetic signals are received through spectrum detection, signal electromagnetic feature information is extracted, and is compared with unmanned aerial vehicle remote control and telemetry signal electromagnetic features in a database, whether the signals are unmanned aerial vehicle signals is judged, unmanned aerial vehicle direction finding information is obtained; the visible light video and infrared video information of the target is obtained through photoelectric detection; broadcast signals emitted from an onboard type ADS-B system are monitored in real time, the point trace and flight trace information and identity information of the cooperative unmanned aerial vehicle target are obtained; all the above information is integrated, so that the comprehensive information including point and flight trace, category, cooperative/non-cooperative identity information and tracking video information of the target can be obtained. Compared with the prior art, the detection system and detectionmethod can realize three-dimensional positioning, speed measurement, category identification and point trace generation of a low-altitude unmanned aerial vehicle target, a monitor the cooperative target in real time so as to obtain a low-altitude flight real-time trend.

The invention provides a weld surface defect identification method based on image texture. Miniature CCD camera photographing parameters are set according to the acquisition image standard; an acquired true color image is converted into a grey-scale map, and a gray scale co-occurrence matrix is created; 24 characteristic parameters in total, including energy, contrast, correlation, homogeneity, entropy and variance, are respectively extracted in the direction of 0 degree, 45 degrees, 90 degrees and 135 degrees, and normalization processing is performed on the extracted characteristic parameters; a BP neural network is trained by utilizing a training sample image, and the number of neurons of the neural network, a hidden layer transfer function, an output layer transfer function and a training algorithm transfer function are set; test sample characteristic parameters are outputted to the trained BP neural network to perform classification and identification; and the matching degree of the test sample and different types of surface welding quality training samples is calculated so that automatic classification and identification of the test sample surface welding quality can be completed.

The invention relates to an interference pattern recognition technology of a frequency hopping system, belonging to the field of broadband wireless communication. The technology is mainly used for analyzing various interference types of the frequency hopping system. On the premise of not increasing frequency hopping bandwidth and frequency hopping speed, the recognition for various interference patterns can be realized. First, pretreatment for removing frequency hopping points is carried out on the received signals; then, characteristic parameters such as time domain, frequency domain, time-frequency domain and the like of interference signals can be extracted for mode recognition, so that classified recognition of the different interference types can be realized. The technology has better recognition effect for the various common interference types as well as stronger robustness and stability, and provides reliable basis for further interference suppression. In addition, the method has low computational complexity and is easy to realize in real time by the system.

The invention discloses a multi-scale target detection model method based on metric learning, and the method comprises the steps: carrying out the initialization of a network through employing a pre-trained model through employing the idea of transfer learning, adding a loss function to carry out the fine adjustment of a weight parameter of the network, and improving the regression precision of aregion of interest in an image; adding a full-connection layer to vectorize feature information after a feature map is extracted from a region of interest in an image, and then performing distance measurement by using the feature information extracted by the full-connection layer to realize classification and identification of target region information. By adopting the method of the invention, theloss of the feature information can be reduced, and the classification and recognition accuracy of the detection target is improved.

The invention discloses an optical fiber intrusion signal recognition method based on a support vector machine, so as to solve the problem of how to accurately and effectively recognize the type of an optical fiber intrusion signal in a phase sensitive optical time domain reflection mechanism. The realization method comprises steps: time-frequency domain features of an optical fiber intrusion signal sample are firstly extracted to serve as a training sample; a kernel function is then used for mapping the sample to high-dimensional space; sequential minimization algorithm is then used for training a support vector machine classifier and mesh optimization and cross validation are used for acquiring the optimal classifier parameters; and finally, the feature vector of a to-be-classified optical fiber intrusion signal is inputted to the classifier, and a corresponding classification result is obtained according to output of the classifier. According to the method of combining the support vector machine and digital features of the optical fiber intrusion signal, the intrusion type can be accurately recognized.

The invention discloses a method for recognizing and classifying road barriers based on video. According to the method for recognizing and classifying the road barriers based on the video, according to the urban road monitoring video, a barrier feature extraction and recognition algorithm is studied, a hybrid Gaussian modeling method self-adaptive to background update is provided, and the road background is updated selectively according to static barrier targets obtained through detection; a moving target segmentation method based on concave-convex outline characteristics of the targets is provided, further accurate extraction and separation of the moving targets are achieved through the method, and a foundation is laid for outline-based segmentation of a blocked target; an algorithm for automatic detection of an ROI of a road is provided, and automatic extraction of the ROI of the road in a monitoring image is achieved. A road barrier classifying method involving self-adaptive clipping of the ROI is adopted, road barriers are recognized and classified to be illegally parked vehicles and scattered objects. By the adoption of the method for recognizing and classifying the road barriers based on the video, the barrier handling efficiency of the traffic department can be improved easily, and a foundation is laid for preventing road accidents.

The invention provides a brain-computer interface based doctor-patient interaction method, which adopts a brain-computer interface based doctor-patient interaction system comprising an electroencephalographic collection module and an electroencephalographic analysis and doctor-patient interaction module. The method comprises the following steps of: setting visual stimulation signals with different frequencies, displaying the signals on a display, and mapping into controls of four commands through different combined codes (00, 01, 10 and 11) of 0 and 1; collecting an electroencephalographic signal of a subject by an electrode in real time, sending into a computer after the signal is amplified; analyzing the electroencephalographic signal; and broadcasting a voice prompt corresponding to the identification result through a speaker, and displaying the result on the display. Therefore, medical staff can carry out corresponding rescue according to the display and voice. The brain state can be identified by the brain-computer interface technology, and external equipment is accurately driven in time, so that communication and control can be realized. Brain can be effectively excited to generate steady state visual evoked potential by visual stimulations in different frequencies, the electroencephalographic signal can be accurately identified under stimulations in different frequencies through signal calculation and processing functions of the computer, and real doctor-patient interaction can be realized in the result display and voice prompt mode.

The invention discloses a multi-angle indoor human action recognition method based on a 3D skeleton. The multi-angle indoor human action recognition method comprises the following steps that 1) videos of human motions at three angles of a front angle, a squint angle and a side angle are acquired; and the videos include training videos and test videos; 2) human skeleton 3D features in the videos are extracted through body feeling equipment; and the three-dimensional skeleton features include global motion features and arm and leg local motion features; 3) model training is performed; feature description is performed through the human skeleton 3D features in the training videos so that a training feature set is obtained; and the concrete process is listed as follows: online dictionary learning of the three-dimensional skeleton features is performed; and then dimension reduction is performed through sparse principal component analysis so that a feature set data set is formed; and 4) the feature set of the samples of the test videos is inputted, and recognition is performed through a linear support vector machine (LSVM). Classified recognition of the multi-angle motions is realized by the method so that the limitation of a single-angle recognition algorithm can be overcome and thus the method has more research value and actual application value.

The invention relates to an aerial target classification method based on wind field disturbance characteristics. The method includes steps of resolving wind field disturbance characteristics; accurately extracting the wind field disturbance characteristic parameters of trailing vortex core position, vortex core radius, vortex core spacing and vortex circulation by an algorithm for resolving wind field disturbance characteristics of an aerial target; inversing type identification characteristics; inversing to obtain type identification characteristic parameters of target track characteristics, physical characteristics and motion characteristics by an algorithm of inversing the type identification characteristics of the aerial target; identifying attribute of the aerial target; and outputting the type attribute of a plane target to be identified by inputting the type identification characteristic parameters of the plane target to be identified. Extraction of wind field disturbance characteristics of the aerial target is achieved, inversion of the wind field disturbance characteristics to target characteristics is achieved, and the aerial plane target is classified and identified on the basis of laser radar detection for specific airspace atmospheric wind fields.

The invention discloses a visible foreign matter and bubble classification recognition detection method for medical large volume injection (250ml), and the method comprises the following steps: 1), continuously obtaining a plurality of images of large volume injection in detection; 2), carrying out the image pre-processing: filtering the images based on Top-Hat morphological filter; 3), carrying out image segmentation: carrying out the segmenting of the filtered images through employing an interframe differential method based on the maximum information entropy; 4), carrying out extraction of defected edges: extracting the edges of the visible foreign matters and bubbles through employing an SUSAN algorithm; 5), carrying out an image feature extraction algorithm: extracting the shape, gray and movement feature parameters for describing foreign matters and bubbles through the research and analysis of the features of defects; 6), carrying out the defect classification and recognition: achieving the recognition and classification of foreign matters and bubbles through employing an IDS-ELM algorithm. The invention achieves the recognition and classification of foreign matters and bubbles, can precisely classify and recognize various types of defects, and enables the products with different types of defects to be removed and placed in different defective product regions.

The invention discloses a remote-sensing image multi-scale object-oriented classification method based on joint sparsity representation. The method includes the steps of firstly, mining a spatial characteristic of a remote-sensing image and constructing an augmentation characteristic through the combination of the spatial characteristic and a spectral characteristic, secondly, constructing a redundant dictionary by means of a training pixel sample and the augmentation characteristic, carrying out joint sparsity representation on initially segmented patches by the adoption of the redundant dictionary, thirdly, carrying out homogeneity analysis and reconstruction effect analysis on the patches on the basis of the joint sparsity representation, and finally, judging whether patch segmentation is reasonable according to analysis results of the homogeneity analysis and the reconstruction effect analysis, and carrying out classification identification on the patches meeting the requirements of the homogeneity level and the reconstruction effect. The method achieves organic combination of the segmentation process and the classification process so as to obtain proper ground object patches for classification, achieves classification identification of the remote-sensing image on the object level, can obtain a classification result meeting the requirement of visual interpretation, greatly improves interpretation precision of the remote-sensing image, and has great application value.

The invention discloses a defective goods eliminating and sorting device based on a medical infusion visible impurity detecting system. The device comprises a cutter, a swing plate, a driving device, a clamping cylinder and clamping fingers, wherein one end of the swing plate is mounted on a main shaft of the cutter; the driving device is arranged on the swing plate; the driving device comprises a motor, a motor mounting base, a cylinder mounting base and a lead screw; the clamping cylinder is mounted below the cylinder mounting base; the clamping fingers are arranged on the clamping cylinder; and the cutter and the motor are both controlled by a PLC control unit of a medical infusion visible impurity classifying, identifying and detecting system. The device is combined with an IDS-ELM algorithm to realize the identification and the classification of visual impurities and bubbles, can precisely classify and identify various defects, and can discard the defects with different types to different defective goods areas. The cutter is skillfully applied to solve the sorting problem which causes trouble during infusion for a long time in the prior art; and the device is simple in structure and easy to realize.

The invention discloses an intelligent radar radiation source signal classification method based on a long-short time memory model. The method mainly solves the problem that the prior art is low in identification rate and slow in identification speed. The realization scheme is as follows: 1) generating a radar radiation source signal data set, and performing data preprocessing on the same; 2) acquiring a training sample set, a testing sample set and a verification sample set from the preprocessed data set; 3) constructing a seven-layer long-short time memory unit network and setting parametersof a network model; 4) adjusting the hyper-parameter of the network model and training the long-short time memory unit network by utilizing the training sample set and the testing sample set; and 5)inputting the verification sample set into the trained long-short time memory unit network model, thereby acquiring a radar radiation source signal classification result. Through the classification method disclosed by the invention, the automatic feature extraction and accurate signal classification can be performed on the one-dimensional signal; and the classification result is excellent, the time complexity is low, the stability is good, and the method can be used for the radar radiation source signal identification under the complex electromagnetic environment.

The invention relates to the technical field of network information, in particular to a private encryption protocol message classification method based on sparse representation and a convolutional neural network, which comprises the following steps: obtaining and preprocessing network traffic data to obtain a data set file and a label file; importing the data set file into a sparse auto-encoder for unsupervised feature learning to obtain input data with smaller dimension; and training the two-dimensional convolutional neural network by using the training set after sparse representation and thetraining set label, performing convolution and pooling, and minimizing errors to obtain a classifier. According to the classification method disclosed by the invention, the classification characteristics of the private encryption protocol message are automatically learned from the original network flow, and classification identification is realized; the method does not depend on the IP address and port number information of the header of the network traffic data packet, and the generalization ability of the classification model is high; sparse representation is used for learning local features of private encryption protocol messages, a two-dimensional convolutional neural network is used for learning global features of the private encryption protocol messages, and the recognition precision of the classifier is improved.

The invention provides a modulating signal recognition method based on a complexity feature under low signal to noise ratio. The method comprises the following steps: extracting multi-fractal dimension features of different communication modulating signals, protruding features of different probability points in a time signal sequence so as to extract the features of different communication modulating signal types; furthermore, grouping pre-processed discrete-time signal sequences so as to simplify the computation of the multi-fractal dimension on the one hand, and translate the long time signal sequence into small sequence sections to observe and compute on the other hand, thereby performing small-range feature expression on the signal, and extracting the feature of the signal in a more refined manner; moreover, performing grey correlation processing on the extracted multi-fractal dimension feature of the unknown communication modulating signal and the computed multi-fractal dimension feature of the known communication modulating signal in a database, selecting the modulating type of the signal with large correlation degree as the modulating type of the unknown communication modulating signal, and then realizing the classification recognition of the modulating type.

The invention provides a power grid fault classification method based on a classification model. Character string matching fault identification based on atomic energy entropy is carried out; transientcharacteristic information characteristics of the characteristic quantity are analyzed, the atomic energy entropy is selected as the fault characteristic quantity, so that the defect of poor adaptability of a primary function to signal characteristics in wavelet analysis can be overcome; meanwhile, the fault characteristics can be completely described; a power distribution network power transmission line fault classification model is established based on a coarse neural network; ten fault types of the high-voltage transmission line are classified and recognized, based on the difference of equivalent instantaneous excitation inductance under various operation conditions, maximum overlapping discrete wavelet transform analysis is added to extract fault feature vectors to serve as training and testing values of a decision tree, and rapid fault recognition and classification can be effectively achieved.

The invention relates to a visible light unmanned aerial vehicle remote sensing image forest tree species classification method based on multi-feature optimization, and the method comprises the steps:obtaining a visible light remote sensing image of a needed forest region through unmanned aerial vehicle aerial photography, and generating a digital surface model and a digital orthographic image map through preprocessing; establishing a visible light unmanned aerial vehicle image tree species classification system; extracting multiple types of features from the digital surface model and the digital orthographic image map to construct a feature space; selecting an optimal classification feature subset by using a recursion elimination random forest algorithm; using a random forest algorithm on the feature subset to realize tree species classification, and extracting a tree species distribution diagram; and performing precision evaluation on the classification result. The method is beneficial to popularization and application of the unmanned aerial vehicle visible light remote sensing image in forest type and tree species identification in a forest region.

The invention discloses a garbage carrying and sorting robot which comprises an electric chassis truck and further comprises a tray, a mechanical arm, a first depth camera and an industrial camera that are arranged on the electric chassis truck, a second depth camera is arranged at the tail end of the mechanical arm, and an industrial personal computer is arranged in the electric chassis truck; and the mechanical arm, a driving motor of the electric chassis vehicle, the industrial camera and the depth camera are respectively connected with the industrial personal computer. The invention further discloses a control method of the garbage carrying and sorting robot. A garbage can may be carried, the manual workload can be effectively reduced, garbage in tasks can be classified, the garbage is carried according to a classification result, erosion of household garbage to land is effectively reduced, and recycling of the garbage turned into wealth and conversion of material energy are achieved.

The invention discloses a Chinese paper cutting identification method based on space constraint characteristic selection and combination thereof, comprising the following steps of: (1) extracting an initial feature of a sample by adopting a method combining space pyramid matching and context dependent histogram to form a candidate feature of the sample; (2) processing the candidate feature by utilizing a feature selection and combination technology based on AdaBoost to obtain a distinctive feature; (3) characterizing the class through the center feature vector of all combination features in each class, the distinctive feature and a distance calculation formula for defining the center feature vector and the distinctive feature; and (4) calculating the distance between the distinctive feature of a testing sample and the center feature of each class to obtain a classification and identification result of paper cutting works. The invention effectively combines the two ways of space pyramid matching and context dependent histogram, overcomes the limitation thereof on expressing the shape of an image, extracts and forms distinctive paper cutting image shape features, and realizes the classification and the identification of paper cutting works on the basis.

The invention discloses a CNV brain electrical lie detection method based on brain network analysis. The method comprises four parts of electroencephalogram signal acquisition, electroencephalogram signal preprocessing, electroencephalogram signal feature extraction and electroencephalogram signal display. The method uses a brain network analysis method to extract features of CNV brain waves, and makes up for the deficiency of existing lie detection technologies on cognitive difference of the whole brain region of electroencephalogram signals. Meanwhile, the electroencephalogram signals of multiple subjects in an experimental group and a control group are separately collected by software, the collected electroencephalogram signals are stored in a preset storage unit, pretreatment and feature extraction are conducted on electroencephalogram data of the same subject in different states, and analysis and comparison are conducted on the pretreatment and characteristics so as to provide reasonable lie detection results.

The invention discloses a rock identification classification method and a system. The rock identification classification method comprises steps of picking up laser intensity value corresponding to the every point having spatial geometry information from three-dimension point cloud data of outcrop, performing correction on laser intensity of each point of the outcrop to obtain a corrected laser intensity value, and performing classification and identification according to the corrected laser intensity. The invention obtains the laser intensity value data of the outcrop. As the laser intensity value reflects the outcrop ability of reflecting the laser, and, meanwhile, the different lithological outcrops have different reflection abilities for laser, the classification identification of the outcrop rock can be realized through the laser intensity value. Furthermore, through the correction on the obtained laser intensity value, the accuracy of the classification identification of the outcrop rock can be improved.

The invention discloses a non-human multi-target real-time track extraction method in a traffic video scene. The non-human multi-target real-time track extraction method comprises the steps: carryingout the target detection of an inputted traffic video image through combining with a background difference method, and obtaining all traffic entity targets of each frame in the video image; realizingclassification and identification of motor vehicles, non-motor vehicles and pedestrians by utilizing the length-width ratio and the two-dimensional area characteristics of the traffic entity targets;designing a non-human multi-target real-time trajectory matching tracking algorithm, matching the trajectories of the traffic entity targets detected in each frame with the trajectories of the previously detected traffic entity targets one by one, and classifying the traffic entity targets to which the traffic entity targets belong to realize tracking if the matching succeeds; and extracting the trajectory of the traffic entity target subjected to trajectory matching tracking, and determining the driving direction of the traffic entity through judgment of starting and ending coordinate positions of the target trajectory. According to the invention, the extracted multi-target trajectory can be automatically matched to different traffic entities and different traffic flow directions of non-human and non-human, and the robustness is good.

The invention relates to a BiLSTM-based composite characteristic optical fiber sensing disturbance signal mode identification method, which comprises the following steps that vibration signals including optical fiber sensing disturbance signals in different modes are collected, data is stored, and type labels are added; the time domain feature extraction unit is used for calculating short-time energy and a short-time over-level rate for the acquired vibration signals, setting thresholds of the short-time energy and the short-time over-level rate, and preliminarily judging intrusion disturbancesignals according to judgment conditions; the frequency domain feature extraction unit is used for carrying out four-layer wavelet packet decomposition on each vibration signal, solving 16 sub-band energy spectrum distribution, splicing short-time energy and short-time over-level rate to form a composite feature vector, carrying out normalization processing on the composite feature vector, and taking the normalized feature vector as an input feature vector; and a bidirectional LSTM network model is constrcuted, the normalized feature vector is taken as an input, the event label is taken as aclassification output result, and a classifier is trained by using the test sample to realize optical fiber sensing disturbance signal mode identification.