45 results about "Support vector classifier" patented technology

The invention discloses a real time filtrating method for large-scale garbage message based on the content, including the steps as following: 1, pre-filtrating by using the black list and the white list; 2, carrying out the online filtrating by using the filtrating module based on the frequency; 3, carrying out the fast filtrating for the message content by using the method of twice hashing; 4, carrying out the pretreating of the message text for suspicion message, and converting the same into the phase vector; 5, judging the suspicion message by using the method of combination of Naive Bayesian classifier and support vector classifier. The invention can greatly improve the filtrating speed of garbage message, and efficiently reduce the produced erroneous judgement rate in the conventional key word filtrating method; can efficiently solve the problem of group sending garbage messages with malicious intent in the short time; can efficiently avoid to mistake the common message as the garbage message so as to reduce the erroneous judgement, and efficiently improve the filtrating accuracy of whole system by analyzing the message content on the semantics.

The present invention provides a system and method for building fast and efficient support vector classifiers for large data classification problems which is useful for classifying pages from the World Wide Web and other problems with sparse matrices and large numbers of documents. The method takes advantage of the least squares nature of such problems, employs exact line search in its iterative process and makes use of a conjugate gradient method appropriate to the problem. In one embodiment a support vector classifier useful for classifying a plurality of documents, including textual documents, is built by selecting a plurality of training documents, each training document having suitable numeric attributes which are associated with a training document vector, then initializing a classifier weight vector and a classifier intercept for a classifier boundary, the classifier boundary separating at least two document classes, then determining which training document vectors are suitable support vectors, and then re-computing the classifier weight vector and the classifier intercept for the classifier boundary using the suitable support vectors together with an iteratively reindexed least squares method and a conjugate gradient method with a stopping criterion.

The invention discloses a cerebrum cognitive status recognition method based on cerebral function imaging, comprising the following steps of: inputting experimental data series of cerebral nuclear magnetic resonance images; preprocessing the experiment data of the cerebral nuclear magnetic resonance images; establishing a multidimensional data set which includes a cerebrum coordinate dimension, acerebrum area dimension, an experiment time sequence dimension and a cognitive status dimension; extracting features; classifying and identifying the statuses. The invention extracts feature vectors from cerebral images of cerebra knowing the task statuses to train a support vector classifier, and classify the cognitive status for the feature vectors of an unknown cerebral image using the classifier. If the unknown cerebral image belongs to some category, then the cognitive state the category belongs to is identified as the cognitive status of the unknown cerebral image, thereby achieving theaim of cerebrum cognitive status recognition. The multidimensional data set of cerebral images of the invention are used for secondary analysis, treatment and integration of the cerebral images, thereby enabling the data calculation and analysis of the cerebral images to be more convenient, faster and more visual.

The invention discloses a method for estimating a safety domain of a track irregularity amplitude in the technical field of track traffic safety. The method comprises the following steps of: establishing a wheel / rail dynamics simulation model; inputting track irregularity data into the wheel / rail dynamics simulation model to obtain derail coefficient, wheel load reduction rate and wheel / rail cross force, and smoothing the derail coefficient; marking the track irregularity data once; performing grid division of the track irregularity data, counting the track irregularity data in grid to obtain dangerous point distribution rate of the grid, and completing secondary marking on the track irregularity data; and training a support vector classifier; and finally testing the trained support vector classifier, if a reference standard is reached, determining that the support vector classifier is available, otherwise, training the support vector classifier again. The method can be used for quantitative evaluation of a geometrical shape of a high-speed railway track on traffic safety.

The present invention provides a blasting vibration predicting method based on a particle swarm algorithm optimization support vector machine. According to the method, firstly, blasting vibration influence factors are subjected to feature extraction. Secondly, the kernel function, the penalty factor, the slack variable and the kernel parameters of the support vector machine are subjected to combined optimization thorough the improved PSO algorithm, and then an optimal support vector regression machine and an optimal support vector classifier model are respectively obtained. In this way, the classified prediction of blasting vibration data is realized. Compared with the traditional blasting vibration velocity predicting method for support vector machines, optimized combined parameters are obtained, so that the performances of models can be better improved. The prediction accuracy of models is improved, and the prediction accuracy of the blasting vibration strength is greatly improved.

The invention discloses a universal no-reference image quality evaluation method based on natural scene statistics of a transformation domain and a spatial domain. The method comprises the following steps of (1) carrying out NSCT (Non Subsampled Contourlet Transform) decomposition on an image to obtain sub-band coefficients, which have the same size as an original image and are in different scales and different directions; (2) extracting a mutual information statistics characteristic between a relative coefficient and a father coefficient of an NSCT sub-band; (3) extracting a structural information statistics characteristic between the relative coefficient and the father coefficient of the NSCT sub-band; (4) extracting an MSCN (Mean Subtracted Contrast Normalized) coefficient of the original image and a neighborhood coefficient statistics characteristic thereof; and (5) respectively constructing a no-reference image quality evaluation model and an image distortion type recognition model by utilizing SVR (Support Vector Regression) and an SVC (Support Vector Classifier) on the basis of the characteristics. In comparison with the prior art, the method has the advantages of consistent evaluation result and human subjective evaluation height, high classification accuracy, low computing complexity, stronger application value and the like.

The invention discloses a method for detecting an adverse state of inclined sleeve part screws of a high-speed train overhead line system. The method comprises steps that firstly, a sample database of inclined sleeve parts is established, an AdaBoost classifier cascaded with HOG characteristic training of samples is extracted, and a supporting vector classifier is trained; secondly, Hough transformation is employed to realize extraction of an inclined sleeve inclination angle of a target image, and the inclination angle is made to rotate to a vertical direction; during fault determination, a bolt length and diameter ratio is taken as a criteria of a bolt loosing fault, and a relevant threshold is set to determine the bolt loosing fault; the bolt loosing fault is determined according to the position of a thin nut, differential processing on pixel accumulated distribution in a horizontal direction is carried out, and whether loosing occurs is determined according to a relevant horizontal pixel distribution change rate. Through the method, the state of inclined sleeve screw parts of the high-speed train overhead line system can be directly detected, an objective, true and accurate detection analysis result is acquired, and disadvantages of a traditional manual detection method are overcome.

The invention discloses a reference-free image quality evaluation method based on image entropy. The method comprises the following steps of (1) extracting image features to be tested, and identifyingthe probability pi of each distortion type of the image to be tested by using a support vector classifier (SVC), wherein the image features include mutual information between different color channels, two-dimensional entropy of gray image and two-dimensional entropy and mutual information of filtered sub-band image; 2 respectively assuming that the images to be tested belong to a certain distortion type, calculating a score qi corresponding to the distortion type by the support vector regression (SVR), and obtaining a quality index of the image to be tested by weight sum. Experiments on LIVEand TID2013 databases show that the algorithm has good performance and consistency of subjective and objective evaluation.

The invention provides a depth feature-based high-spatial resolution remote sensing image migration learning method. According to the method, a newly obtained remote sensing image is directly classified by using existing remote sensing images and the sample information, so that the method provides the support for the remote sensing rapid monitoring. The method comprises the following steps of carrying out the main component transformation for a source domain image and a target domain image, and respectively extracting previous three main components of the images; for newly generated images of three wave bands, extracting an image block with each pixel as the center thereof and inputting the image block into a well trained multilayer convolution neural network; outputting the last one full-connection layer of the convolution neural network and obtaining the depth feature representation of the pixel; subjecting the training samples of the source domain image and the target domain image to the support vector machine classifier training process based on extracted depth features and then obtaining a classifier; classifying the target domain image directly by using the obtained support vector classifier, and completing the migration learning process of the source domain image and the category corresponding relation to the target domain.

The present invention provides a system and method for building fast and efficient support vector classifiers for large data classification problems which is useful for classifying pages from the World Wide Web and other problems with sparse matrices and large numbers of documents. The method takes advantage of the least squares nature of such problems, employs exact line search in its iterative process and makes use of a conjugate gradient method appropriate to the problem. In one embodiment a support vector classifier useful for classifying a plurality of documents, including textual documents, is built by selecting a plurality of training documents, each training document having suitable numeric attributes which are associated with a training document vector, then initializing a classifier weight vector and a classifier intercept for a classifier boundary, the classifier boundary separating at least two document classes, then determining which training document vectors are suitable support vectors, and then re-computing the classifier weight vector and the classifier intercept for the classifier boundary using the suitable support vectors together with an iteratively reindexed least squares method and a conjugate gradient method with a stopping criterion.

Embodiments of the invention disclose a method and device for forecasting and pre-warning downhole abnormal working conditions in real time in shale gas fracturing processes. The method comprises the following steps of: constructing a training data set on the basis of offline data, and training a monitoring parameter forecasting model by utilizing a support vector regression machine; extracting trend features of feature parameter offline data through calculating a plurality of discretized slope values, and establishing downhole abnormal working condition monitors under corresponding working procedure stages on the basis of a support vector classifier; and further carrying out real-time forecasting and pre-warning on downhole abnormal conditions on the basis of online monitoring data by utilizing the established monitoring parameter forecasting model and the downhole abnormal working condition monitors. According to the method and device r forecasting and pre-warning downhole abnormal working conditions in real time in shale gas fracturing processes, the precision of forecasting downhole abnormal working conditions is improved.

The invention discloses a method for detecting SAR image changes based on neighborhood clustering kernels. The method mainly solves the problems that no-label sample information can not be utilized by an existing method for synthesizing kernels based on differential values and lower detection precision is caused. The method includes the first step of extracting strength features and textural features of two time phase images, the second step of manually selecting label training samples and no-label training samples, the third step of building differential synthetic kernels by the utilization of the label training samples, the fourth step of correcting the differential synthetic kernels by utilization of the no-label training samples to obtain the neighborhood clustering kernels, the fifth step of inputting the neighborhood clustering kernels into a support vector machine to carry out training, and obtaining a support vector classifier, the sixth step of inputting the label training samples and the neighborhood clustering kernels formed by all pixel points into the support vector classifier to carry out testing, and obtaining a final change detection result. Compared with the method for synthesizing the kernels based on the differential values, the method for detecting the SAR image changes has the advantages of being high in detection precision, good to speckle noise resistance of the SAR images, and capable of being used in SAR image change detection.

The invention discloses an image emotion classification method based on a ResNet-GCN network. The method includes following steps: (1) designing an image emotion classification network formed by a front part structure ResNet-50 network and a rear part structure GCN network; (2) designing an image emotion classification framework including one image emotion classification network ResNet-GCN and onesupport vector classifier used for deciding fusion network characteristics; (3) performing substantial main body extraction and pyramid cutting on an original image; (4) training the image emotion classification network; (5) testing the image emotion classification framework; and (6) performing classification by employing the trained image emotion classification framework by a user image to realize image emotion classification. According to the method, an image emotion classification result is in accordance with the mankind emotion standard, human participation is avoided in a determination process, and automatic image emotion classification of a machine is realized.

Provided is an excitation surge current identification method based on a support vector classifier. The method selects seven characteristics of a secondary second harmonic, a third harmonic, a current dead angle, a wave width, a waveform distortion amount, a waveform correlation coefficient and an excitation side measured impedance as inputs of a support vector machine, then various running states of a transformer are trained, and a decision function identifying an excitation surge current and a fault current is constructed. When the transformer is out of order, data collected by a protection device collection system is calculated to obtain seven characteristics, the seven characteristics are put into the decision function and determination of an excitation surge current and a fault current is carried out. The identification method integrates advantages of principles of harmonic wave braking, a dead angle, waveform similarity and the like and avoids respective limitation, and the surge current identification credibility is raised. An algorithm can be converted into a convex optimization problem finally, and the local minimum problem which cannot be solved by a neural network is avoided. The method is free from a transformer wiring mode and is free from model parameters, the applicability is strong and the flexibility is good.

The invention discloses a method and a system for detecting an abnormal user based on a water-electricity ratio and support vector clustering. The method comprises the following steps: firstly, acquiring water and electricity data of the user; secondly, obtaining non-zero consumption data in the water and electricity data; thirdly, calculating the water-electricity ratio of the non-zero consumption data; fourthly, inputting the water-electricity ratio to a support vector classifier for performing classification to obtain classification data; finally, judging whether the classification data is a noise cluster or not, and if the classification data is the noise cluster, the classification data is abnormal data; if the classification data is not the noise cluster, the classification data is normal data. The electricity consumption and the water consumption are effectively integrated with a water-electricity ratio or electricity-water ratio method and serve as input vectors for support vector clustering analysis in sequence for performing detection, so that the false detection rate and the missing detection rate of the water consumption and the electricity consumption during separate detection can be effectively avoided. A support vector clustering design can obtain relatively high classification accuracy by utilizing relatively good distinguishing performance. With a method for dynamically setting a support vector kernel function and a maximum number of iterations, the sensitivity of the detection method can be improved.

The invention relates to a power transmission line live working technology and especially relates to a helicopter live working combination air gap discharge voltage prediction method. An air gap between a high voltage lead and a suspended conductor is defined as a first gap. The air gap between a suspended helicopter and a grounding body is defined as a second gap. The method is characterized by through a static electric field, calculating and extracting an electric field characteristic set for the first time, inputting to a support vector classifier and carrying out the primary prediction ofa discharge voltage, and determining which gap breaks first and a discharge voltage value; according to a potential change condition after breakdown, carrying out the secondary calculation of the static electric field and electric field characteristic extraction, using the support vector classifier to carry out the secondary prediction of the discharge voltage, and determining the discharge voltage value of an after-discharge gap; and comparing the discharge voltage values of the previous discharge gap and the after-discharge gap, and taking a higher value as the discharge voltage prediction value of a helicopter live working combination air gap. The method can be used to replace a discharge test, test cost is saved, and guidance can be provided for acquiring the minimum safe distance of helicopter live working.

The invention discloses an emotion recognition method and system. The method comprises the steps of acquiring an electroencephalogram signal to be recognized; extracting to-be-identified time-frequency domain characteristics, to-be-identified nonlinear characteristics and to-be-identified brain network attribute characteristics; wherein the to-be-identified time-frequency domain feature is a wavelet entropy calculated according to a wavelet decomposition coefficient of the to-be-identified electroencephalogram signal; wherein the to-be-identified nonlinear feature comprises a power spectrum density obtained by performing discrete Fourier transform on the to-be-identified electroencephalogram signal and a Hurst index obtained by performing average error calculation on the to-be-identified electroencephalogram signal; wherein the attribute characteristics of the to-be-identified brain network reflect the correlation among the to-be-identified electroencephalogram signals; and carrying out emotion recognition on the to-be-recognized electroencephalogram signal features by adopting a trained support vector classifier. The method and the system provided by the invention have the advantage that the emotion recognition accuracy can be improved.

The invention provides a real-time vehicle-mounted weighing method. The real-time vehicle-mounted weighing method comprises the following steps: S2, establishing a mapping relationship between the numerical value of a strain gauge and the weight of a standard weight; S3, extracting a feature vector of the time sequence weight data by using fast Fourier transform; S4, inputting the feature vector into a multi-classification support vector classifier model, and recognizing the running state of the vehicle; and S5, outputting the weight of the vehicle in real time according to the running state of the vehicle. All-weather monitoring of overload is achieved, the overload problem is completely eradicated, real-time weight data at the forefront end are provided for vehicle dispatching, and the operation efficiency is greatly improved.

The present invention provides an electric appliance type determination method for a students' dormitory. An electric appliance identification device consisting of an information acquisition module, an information processing module and a communication module is employed to realize the electric appliance type determination method for the students' dormitory. The electric appliance type determination method is abundant in feature information through adoption of the starting current features, the fundamental wave voltage current phase difference and the load current frequency spectrum features of an electric appliance, and is combined in the features of two classifiers to perform comprehensive identification and high in identification accuracy through adoption of a combination classifier consisting of a support vector classifier and a Bayes classifier to perform identification classification. The fundamental wave voltage and current phase difference, starting current feature and load current frequency spectrum feature obtaining methods are simple and reliable. The electric appliance identification device may be used at collective public places requiring electricity load management such as a students' dormitory and the like, and may be also used at the occasions requiring electricity device management for determination and statistics of electricity load types.

The invention provides an FPGA (Field Programmable Gate Array) parallel computing implementation device of a support vector machine, which solves the problem of low speed of present software training of SVM (support vector machine). The FPGA parallel computing implementation device of SVM implements a data preprocessing and storage module and a task configuration module on a PC (personal computer) host computer, and implements an SVC (Support Vector Classifier) training module, an SVR (Support Vector Regression) training module and a model test module / . Multiple SVC / SVR training modules can be provided, the VC / SVR training modules divide the training samples into several parts and train the divided training samples simultaneously. The PC host computer and the FPGA perform data transmission through a RS232 communication interface. The FPGA parallel computing implementation device of SVM can be used for processing all the data of real number type, is universal, makes full use of parallelism of FPGA, enables a plurality of SVM modules to perform parallel computation, and greatly improves the training speed of SVM.

The invention provides a method for reducing a vehicle load error and a computer readable storage medium. The method comprises the following steps: establishing a mapping relationship between a numerical value of a strain gauge and the weight of a standard weight; filtering the numerical value of the strain gauge by adopting a Butterworth low-pass filter to obtain the numerical value of the straingauge after the first filtering; performing empirical mode decomposition on the numerical value of the strain gauge after the first filtering to reconstruct a signal to obtain the numerical value ofthe strain gauge after the second filtering; extracting a feature vector of the time sequence weight data by utilizing wavelet transform; inputting the feature vector into a multi-classification support vector classifier model, and identifying the running state of the vehicle; and outputting the weight of the vehicle in real time according to the running state of the vehicle and the value of the strain gauge after the secondary filtering. Errors brought to the strain gauges by the external environment are effectively eliminated, so that the weighing result is more accurate, and the real and stable vehicle load is obtained.

The invention relates to an indoor room-level positioning method based on Wi-Fi fingerprint database text classification. The indoor room-level positioning method comprises the following steps: firstly, collecting Wi-Fi signal intensity and a basic service set identifier in an indoor environment of a shopping mall; converting the Wi-Fi fingerprint library into short text data; carrying out the feature selection and word weight calculation; adopting a Crammer-Singer support vector classifier; calculating classification accuracy. The method has the beneficial effects that the signal intensity ofWi-Fi is converted into short text words, the influence of the signal intensity is directly ignored, the characteristic of the signal intensity is not considered any more, and a Wi-Fi fingerprint library is reduced; according to the method, the Wi-Fi fingerprint database is converted into the short text data set, the data dimension is reduced, meanwhile, the text classifier is based on the linearkernel SVM classifier, the training and testing efficiency is extremely high, the positioning time can be greatly shortened, and the positioning precision can be greatly improved.

The invention discloses an intelligent decision support system design and implementation method for unmanned aerial vehicle SAR image automatic target recognition. The method comprises constructing a feature library and an inference engine. The specific process of constructing the feature library is as follows: 11) performing segmentation detection on a target in the SAR image; 12) carrying out target false alarm elimination and target calibration on the SAR image; 13) constructing a multi-scale space based on a Gaussian difference model; and 14) taking multi-resolution features of the SAR image, and constructing a feature library according to the extracted multi-resolution features. The specific process of constructing the inference engine is as follows: 21) constructing a training set and a test set by using the extracted multi-resolution features of the SAR image; 22) constructing a multi-scale multi-core support vector classifier; and 23) training and testing the multi-scale multi-core support vector classifier, and performing automatic target identification of the SAR image by using the trained and tested multi-scale multi-core support vector classifier to construct an inference engine. The method can realize automatic target identification of the SAR image.

The invention discloses a selective search and machine learning classification far-infrared pedestrian detection method. The method comprises the following steps: generating a candidate region from afar-infrared image by combining a selective search algorithm with a sliding window method; constructing a double-branch support vector classifier based on local intensity difference histogram featuresaccording to different heights of the candidate regions, and classifying the candidate regions to obtain a pedestrian detection box; on the basis, realizing an auxiliary driving reminding function based on pedestrian height and pedestrian motion direction estimation. The system comprises: a candidate region generation module used for acquiring pedestrian candidate regions by utilizing a selectivesearch and local sliding window technology, a candidate region classification module used for realizing candidate region classification by a double-branch support vector machine based on local intensity difference histogram features, a double-branch support vector machine training module, and a dangerous pedestrian early warning module used for estimating based on height and pedestrian movement direction. According to the method, the detection accuracy and the detection speed can be considered, the algorithm performance meets the practical requirement, and the method has an auxiliary drivingreminding function.

The invention provides an automatic classification method and device for resource levels in a cloud operating system. Resource characteristic data in the cloud operating system are acquired, the level of the resource characteristic data is artificially determined, the resource characteristic data with the determined level serve as samples and are inputted to a support vector machine classifier, parameters of the support vector machine classifier are trained, characteristic data of resources newly added into the cloud operating system are acquired and inputted to the support vector machine classifier, and the level of the characteristic data of the resources newly added into the cloud operating system is determined based on the trained parameters. By the aid of the scheme, the resource levels in the cloud operating system can be automatically classified, and objectivity and accuracy of resource classification levels are ensured.

The invention discloses a system for detecting pressure signals of bolts for urban subway tracks. The system comprises a pressure sensor, a wireless transmission module, an upper computer and a voltage-stabilizing circuit, wherein the pressure sensor is used for acquiring pressure generated by the pressure sensor when bolt gaskets are extruded by the bolts; the wireless transmission module is usedfor sending the pressure signals to the upper computer of a train; the upper computer is used for performing fault diagnosis of the track bolts and feeding a diagnosis result back to a trainman; andthe voltage-stabilizing circuit is used for taking electricity from a contact net arranged on the track. The method for detecting states of the bolts provided by the invention comprises the followingsteps: acquiring the pressure against the bolt gaskets through the bolts by adopting the pressure sensor; acquiring pressure and position labels of the bolts by adopting the upper computer; building abolt database; inputting the pressure of the bolts into a trained support vector classifier; and marking the bolt states. According to the system and the method disclosed by the invention, the laborcost is greatly reduced, the efficiency is high, the overall size is small, the maintenance cost is low, and later repairing of the tracks is convenient.

The invention discloses a steam turbine generator set vibration fault diagnosis method and device, and computing equipment. The method comprises the following steps: S1, collecting six types of fault vibration signals of a steam turbine generator set, namely rotor misalignment, rotor imbalance, bearing looseness, rotor collision and abrasion, oil film vortex motion, and steam vortex motion, then converting the fault vibration signals into discrete frequency spectrum amplitudes through Fourier transform, dividing the discrete frequency spectrum amplitudes into N main frequency bands according to energy characteristics, and performing fuzzification processing on an amplitude of each frequency band to obtain a fuzzification value; S2, establishing an information database, wherein each piece of information contains N input features, and dividing the information in the information database into a training set and a test set; S3, establishing an optimal support vector classifier model; S4, training the optimal support vector classifier model by using the training set, and performing a test verification by using the test set. According to the method, the optimal support vector classifier model is established for carrying out a steam turbine generator set vibration fault diagnosis, the architecture is simple, and the convergence is fast.

The invention provides a novel classifier based on an information gain and an online support vector machine, and a classification method thereof. In academic researches of recent years, an online support vector classifier is concerned by some scholars in the information filtering field. The classification method of the novel classifier based on the information gain and the online support vector machine comprises the following steps of: step one, pre-treating sample information to obtain characteristics of a sample; step two, calculating an information amount of each characteristic by using an information gain method and selecting the needed characteristic according to a certain strategy; step three, establishing a characteristic vector capable of adapting to an online support vector machine model according to the selected characteristic; step four, training the novel classifier based on the online support vector machine by utilizing an online model; and step five, utilizing the classifier to classify samples. The novel classifier and the classification method, disclosed by the invention, are used for classifying texts and filtering information.

The present invention relates to the live working technology of power transmission lines, and in particular to a method for predicting the combined air gap discharge voltage of helicopter live working. The air gap is defined as the second gap. Through the first calculation of the electrostatic field and the extraction of the electric field feature set, input the support vector classification machine for the first prediction of the discharge voltage, determine which gap will break down first and its discharge voltage value; perform the second calculation of the electrostatic field and the electric field according to the potential change after the breakdown Feature extraction, use the support vector classification machine to predict the discharge voltage twice, determine the discharge voltage value of the post-discharge gap; compare the discharge voltage values ​​of the two gaps of the first discharge and the latter discharge, and take the larger one as the discharge of the air gap of the helicopter live working combination voltage prediction. This method can replace the discharge test, save the test cost, and provide guidance for obtaining the minimum safe distance for helicopter live working.

The invention discloses a method and device for identifying a somatosensory evoked potential component based on a support vector machine, comprising an acquisition building unit for acquiring median nerve somatosensory evoked potentials of known origin. The time-frequency components of median nerve somatosensory evoked potentials (MNEPs) are obtained by time-frequency conversion. The time-frequency components of MNEPs were matrices transformed to obtain the time-frequency component eigenvectors. The time-frequency component eigenvectors were used as inputs to construct a multi-stage support vector classifier (SVC). The identification unit is used for classifying and identifying the time-frequency components of the median nerve somatosensory evoked potential to be identified according to the multistage support vector classifier, and determining the source corresponding to the time-frequency components of the median nerve somatosensory evoked potential to be identified, so that a multi-stage support vector classifier is designed to improve the sensitivity and specificity of time-frequency classification of somatosensory evoked potentials and effectively overcome the problem of smallsample size.