149 results about "Svm classifier" patented technology

The invention relates to a photovoltaic power generation system fault diagnosis method based on a dual-channel CNN and time-frequency characteristics. The method comprises the following steps: S1, collecting and preprocessing historical time sequence data of a photovoltaic power generation system; s2, copying the preprocessed data set to obtain a one-dimensional time sequence data set and a two-dimensional time-frequency graph data set; s3, inputting a dual-channel CNN for feature extraction, enabling a one-dimensional time sequence data set to enter a 1DCNN channel, and enabling a two-dimensional time-frequency graph data set to enter a 2DCNN channel; s4, performing dynamic fusion on the features extracted by the dual-channel CNN through an SE attention module; s5, constructing a classifier based on the SVM, and optimizing two key parameters of a kernel function of the classifier by adopting an adaptive inertia weight particle swarm algorithm to obtain a final SVM classifier; and S6, inputting the fused features into a final SVM classifier, and carrying out fault diagnosis classification. The complex fault diagnosis efficiency and reliability of the photovoltaic power generation system are effectively improved.

The invention belongs to the technical field of green hydrogen preparation, discloses an intelligent monitoring method for hydrogen impurities in oxygen in an electrolytic cell system, and solves the problems of response lag, high false alarm rate, incapability of dynamically adjusting working conditions and lack of prediction capability in the prior art. According to the method, the hydrogen concentration in oxygen and key operation parameters of an electrolytic cell are preprocessed, then multi-dimensional correlation analysis is carried out, a dynamic characteristic matrix is constructed, then a hydrogen concentration prediction model is trained based on the constructed dynamic characteristic matrix, and an SVM classifier is adopted to carry out abnormal mode recognition; in the monitoring process, if the predicted hydrogen concentration exceeds a threshold value or an abnormal mode is recognized, grading response is triggered, operation of the electrolytic cells is controlled by adopting a closed-loop control strategy according to real-time detection values of key operation parameters of the electrolytic cells, and finally transverse data comparison of the multi-electrolytic-cell system is achieved through a cloud collaborative optimization technology. And identifying common problems and providing maintenance suggestions.

The invention provides a transient stage electromyographic signal gesture recognition method based on electromyographic activation intensity transfer characteristics and self-adaptive analysis window length. The invention aims to solve the problems of imbalance between spatial features and operation cost, imbalance between recognition accuracy and response speed and low algorithm generalization between individuals and between tasks in the traditional electromyographic signal processing technology when transient stage signals are utilized. The method comprises the following steps: firstly, performing signal acquisition and preprocessing, starting point detection, signal frame formatting, feature extraction and SVM classifier training in an offline mode; in the online mode, the preprocessed signals are collected, a starting point is detected, the signals are formatted, myoelectricity activation intensity transfer features are extracted and input into an offline SVM classifier, the length of an analysis window is dynamically adjusted according to the confidence coefficient, and finally the threshold value of the confidence coefficient is fed back and adjusted in real time. Practice verifies that the method can effectively give consideration to the spatial features and the operation cost, achieves the balance between the recognition accuracy and the response speed in different individuals and tasks, remarkably improves the generalization of the algorithm, and provides more efficient and accurate technical support for the fields of artificial limb control, exercise rehabilitation equipment and the like.

The invention discloses an acceleration method, device and equipment for partial discharge detection of a switch cabinet and a storage medium, and relates to the technical field of power equipment fault detection, partial discharge signals are collected in real time through a multi-parameter sensor, multi-source signal alignment is realized by combining a hardware trigger mechanism and timestamp synchronization, noise components are separated through multi-scale wavelet transform, and the partial discharge detection of the switch cabinet is realized. The method comprises the following steps: extracting time-frequency joint features through discrete wavelet transform, reducing dimensionality by adopting a principal component analysis method, taking low-dimensional feature vectors as input limited by a Gaussian mixture model (GMM), taking feature vectors limited by the GMM as input of an SVM classifier, outputting a classification result, and performing online judgment according to the classification result. When the output is partial discharge, an alarm prompt is triggered through the interface module; and when the output is normal, real-time monitoring is continued. According to the invention, signal processing and classification are accelerated by using an FPGA parallel processing architecture, and real-time early warning of power equipment faults is realized.

The invention belongs to the technical field of intelligent traffic, and discloses a traffic state estimation and queuing state discrimination method based on a fuzzy traffic wave model, comprising the following steps: S1, collecting and preprocessing ETC gantry data, thunder-vision fusion trajectory data and artificial report event data, and extracting traffic flow parameters and event features after preprocessing; s2, fuzzy parameters are introduced to improve a traditional traffic wave model, the traffic wave velocity is calculated through the fuzzy parameters, wave front position transmission based on a state transition model is researched, and a fuzzy traffic wave model is constructed; s3, a VMD-GA-ConvLSTM network is constructed to carry out traffic state estimation; s4, determining a traffic state membership degree by adopting a GWO-FCM clustering algorithm, and judging a vehicle queuing state through an SVM classifier; and S5, improving the fuzzy traffic wave model in the network connection environment, and dynamically correcting the model from two dimensions of road section macroscopic characteristics and vehicle microscopic behaviors. The method provided by the invention can provide theoretical and technical support for accurate perception and dynamic regulation and control of the highway traffic situation.

The invention provides a vehicle over-limit type detection method and system based on wagon balance vibration and point cloud data, relates to the technical field of intelligent traffic detection, and aims to solve the problems that in the prior art, misclassification is caused by dependence on single data, the over-limit types of a whole loading vehicle and a bulk vehicle are difficult to effectively distinguish, an over-limit management strategy is lack of pertinence and effectiveness, and the detection accuracy is poor. And real vibration characteristics and the like generated in the lockage process of the vehicle are difficult to accurately separate. According to the method, a wagon balance vibration signal and original vehicle point cloud data are obtained, processing and feature extraction are carried out, vibration features and point cloud geometric features are fused through an SVM classifier, a vehicle type and a loading state are obtained, and judgment is carried out based on a result. According to the method, the problems existing in the prior art are solved, misclassification caused by traditional single data detection is effectively avoided, the overrun types of the whole loading vehicle and the bulk vehicle can be effectively distinguished, high robustness can be still kept in an extreme environment, and misjudgment and missing detection are reduced.

The invention discloses a geological exploration data intelligent cloud management and analysis system. The system comprises a data access module which is responsible for supporting automatic or manual importing of geological exploration data from multiple sources; the data management and integration module is used for receiving the data transmitted from the data acquisition and uploading module, performing version control and integrating the data; the data processing module receives the data provided by the data management and integration module, is responsible for data cleaning, format conversion and normalization processing, introduces an adversarial training technology and generates an adversarial sample; the intelligent analysis module is used for constructing a prediction model, the prediction model is used for modeling time series geological data by using LSTM, capturing a long-term dependency relationship in the data and identifying a geological change trend, then the output of the LSTM is used as a feature to be input into an SVM classifier, classification or regression prediction is carried out on a geological structure, and the geological process and a disaster occurrence mechanism are simulated to obtain a geological model; and generating more synthetic data similar to real data and used for training a prediction model.

The invention discloses a grain quality index intelligent detection method based on machine vision, and relates to the technical field of grain quality safety detection, and the method comprises the steps: collecting an upper surface image and a lower surface image of a grain sample through a high-resolution camera, carrying out the preprocessing, obtaining a preprocessed two-dimensional image, extracting two-dimensional features through a convolutional neural network, and carrying out the recognition of the two-dimensional features; the method comprises the following steps: extracting two-dimensional features by combining a convolutional neural network, identifying external defects of grains, analyzing internal defects of the grains by applying a physical model of a light scattering theory, and obtaining a preliminary quality evaluation result. The external defects of the grains are identified by combining the convolutional neural network to extract the two-dimensional features, and the internal defects of the grains are analyzed by applying the physical model of the light scattering theory; therefore, comprehensive assessment of the internal and external quality of the grain is realized. And the SVM classifier is adopted to carry out final quality state judgment on each grain, so that the user experience and the working efficiency are further improved.

The invention discloses a transformer fault intelligent diagnosis system and method based on a K-Medoid and SMOTE optimization support vector machine, and the system comprises a data collection module which is used for collecting dissolved gas and operation parameters in transformer oil in real time; the data preprocessing module is connected with the data acquisition module and is used for completing data standardization and abnormal value detection; the sample balancing module is connected with the data preprocessing module and used for reducing majority class redundancy through K-Medoid clustering and expanding minority classes through an SMOTE algorithm; the intelligent classification module is connected with the sample balance module and adopts a weighted SVM classifier to train and predict; and the result output and visualization module is connected with the intelligent classification module and is used for displaying the classification result, the confidence coefficient and the key performance indexes. According to the method, the problem of classification performance degradation caused by sample imbalance in the prior art is solved, and the recognition capability of the SVM model on minority class faults and the overall diagnosis precision are improved.

The invention relates to a method and system for automatically identifying the authenticity of entry and exit certificates, which comprises the following steps of: acquiring an integral image of a certificate, positioning a textural feature extremal region, extracting key details, matching an anti-counterfeiting mark feature library by combining a feature matching algorithm, and extracting an anti-counterfeiting standard detail feature image in the key details; a fuzzy algorithm is adopted to unify the size of an anti-counterfeiting region, and gray mapping and normalization are combined to realize standardization processing of an anti-counterfeiting standard detail feature image, so that deformation and illumination interference are solved. According to the method, a graph neural network based on a pre-training attention mechanism network is constructed, a convolutional layer, a pooling layer, a full-connection layer, a nonlinear kernel SVM classifier and a classification module are superposed to form a multi-stage classifier, generalization is improved through positive and negative sample training, an anti-counterfeiting standard detail feature image subjected to standardization processing is used as input, and an authenticity judgment result is output. The multi-level feature analysis and the deep network learning are fused, the authenticity identification precision and the anti-counterfeiting capability are remarkably improved, and the method is suitable for multi-type certificate anti-counterfeiting detection scenes.

An intelligent and multi-layered approach that uses real-time analysis to identify and confirm the authenticity and inauthenticity of bulk digital documents. Support Vector Machine (SVM) learning is implemented to perform significant attribute validations, such as barcode validation, image-specific validations, and signature validations. An SVM classifier is implemented to compare, analyze, predict the accuracy of the document (i.e., quantify the certainty of authenticity) and decision the documents as either valid / authentic or invalid / tampered-state. Neuro-symbolic Artificial Intelligence (AI) technology is subsequently implemented to confirm or deny the authenticity decision resulting from the SVM classifier.

Disclosed in the present invention is an image classification method based on collaborative optimization algorithms and a feature selection mechanism, which effectively improves the image feature extraction quality and the image classification accuracy. The technical solution comprises: step S1, preprocessing a collected image; step S2, constructing an image feature extraction model on the basis of a genetic programming algorithm, performing image feature extraction, using the concept of individual information optimization to assign position and velocity information to each individual in the algorithm, and updating the position and velocity information of each individual to adjust a selection operator; step S3, constructing a feature selection model to perform selection on the extracted features; step S4, by using the selected features as inputs, training an SVM classifier; and step S5, using the trained SVM classifier to classify express delivery images.

The invention relates to the field of electroencephalogram signal processing and neural engineering, and discloses a multi-mode electroencephalogram feature fusion decoding method. The method comprises the following steps: carrying out noise reduction, calibration and standardization processing on an original electroencephalogram signal to generate a standard signal; extracting mu / beta rhythm time-frequency features and Hjorth time-domain features of the signals, and generating a fusion feature vector through PCA dimension reduction fusion; performing pre-classification and weight optimization by using an SVM classifier, performing space and time sequence feature extraction and classification through a CNN-LSTM network, and outputting a classification probability; model parameters are updated according to the probability and verified, and finally real-time control signals suitable for various communication interfaces are generated. According to the method, the accuracy and robustness of electroencephalogram signal decoding are improved, and efficient and self-adaptive motor imagery brain-computer interface control is realized.

The invention discloses a BiLSTM-Logistic regression-based production line fault identification method, and belongs to the technical field of industrial fault diagnosis. The method comprises the following steps: determining characteristic parameters representing the operation condition of a production line, and respectively collecting data of each characteristic parameter at different moments under different working conditions; constructing a sample set and a time sequence data set; a fault recognition model is constructed, the fault recognition model comprises a BiLSTM network unit and a Logistic regression model which are connected in parallel, the BiLSTM network unit and the Logistic regression model extract features in the time sequence data set and the sample set respectively, and the features extracted by the BiLSTM network unit and the Logistic regression model are spliced to serve as input of an SVM classifier; and training the fault identification model and utilizing the fault identification model to predict production line faults. According to the method, the BiLSTM network structure is used for mining the long-term and short-term time sequence dependence of the data, the logistic regression model is used for carrying out rapid explicit modeling on the non-time sequence features, excessive smoothness of the deep model on the structured data is avoided, parallel output of the two is classified through the SVM classifier, the recall effect is good, the accuracy rate is high, and the production line fault can be efficiently recognized.

The invention discloses a method for connecting an intelligent operation and maintenance large model and a public large model, and relates to the technical field of artificial intelligence, and the method comprises the steps: obtaining equipment operation data from a sensor node through a preset data collection frame, and carrying out the preprocessing of the equipment operation data, thereby obtaining a first data set; performing abnormal state recognition on the first data set by using a convolutional neural network and an SVM classifier, extracting an abnormal data subset, and clustering to obtain an abnormal mode; acquiring historical equipment operation parameters and environmental parameters, and performing correlation analysis on the abnormal state, the historical equipment operation parameters and the environmental parameters by adopting a Bayesian network to obtain a potential reason set of the abnormality; constructing an equipment fault knowledge graph to perform path reasoning on reasons in the potential reason set, and generating a fault association mapping table; and a multi-objective optimization function is established based on the fault association mapping table and the real-time state data of the equipment, and an optimal disposal scheme combination is solved, so that the intelligent level of equipment fault diagnosis and operation and maintenance operation optimization is improved.

The invention discloses a neuroblastoma pathological image classification method based on a CMSwinKAN neural network, and the method comprises the steps: firstly building a neuroblastoma pathological image data set, and dividing the data set into a training set and a test set; a CMSwinKAN neural network model is constructed, wherein the CMSwinKAN neural network model comprises a SwinKANsform feature extraction module, a comparison driving multi-scale aggregation module and a kernel activation network classification head; performing feature classification on the slices through an SVM classifier to obtain classification confidence; dynamically weighting the output of the CMSwinKAN neural network model according to the classification confidence, and outputting a final classification result; and taking the training set as input, training the CMSwinKAN model and the SVM classifier, and adjusting and optimizing parameters of the CMSwinKAN neural network model and the SVM classifier.

The present invention relates to a wire rope surface defect recognition method based on feature fusion, which is in the field of wire rope surface defect recognition technology. The method comprises the following steps: graying a wire rope defect image, dividing the image into blocks, improving a traditional LBP algorithm using a central multi-scale local binary pattern based on the image blocks, extracting texture feature information of the divided image, performing PCA dimensionality reduction based on the obtained image texture features, and finally extracting global image texture features using GLCM and performing feature fusion with the reduced image texture features; and performing wire rope surface defect recognition and classification using an SVM classifier. The method solves the problem that traditional local binary patterns (LBP) are easily affected by central pixels and noise and cannot accurately identify wire rope surface defects. The method achieves an overall recognition rate of 97.5% for wire rope surface defects, which is at least 5% higher than other algorithms. The method can effectively identify various defects on the wire rope surface.

The invention belongs to the field of radar automatic target recognition, and relates to an SAR target recognition method and device based on projection features and improved self-learning. The method comprises the following steps: respectively processing each sample parameter of a training sample by a classifier T respectively trained by each prototype set to obtain integrated projection features of the sample parameters; initializing an SVM classifier based on the labeled sample feature set; taking out part of samples from the unlabeled sample feature set, and constructing a temporary sample pool; classifying the samples in the temporary sample pool by using an SVM classifier; evaluating the classification confidence of the samples in the temporary sample pool by using the classification posterior probability; adding a plurality of samples with high confidence into the labeled sample feature set, and retraining the SVM classifier; repeating for multiple times until an optimal SVM classifier is obtained; and target identification is carried out based on the optimal SVM classifier. The target recognition performance under the condition of a small number of labeled samples and a large number of unlabeled samples is effectively improved.

The application discloses a wool and cashmere fiber identification method based on near-infrared spectroscopy, which comprises the following steps: step one, establishing a near-infrared spectroscopy dataset; step two, preprocessing to obtain a spectrum sample set; step three, generating enhanced samples to constitute positive and negative sample pairs; step four, constructing a pre-task model of self-supervised learning; step five, constructing a downstream task model of self-supervised learning; step six, the pre-task model and an SVM classifier constitute a self-supervised learning model based on near-infrared spectroscopy, and the self-supervised learning model of near-infrared spectroscopy is used for classification. The application has the advantages of simple structure and reasonable design, sample enhancement based on spectral band scaling, local smoothing disturbance and spectral peak shift, automatic generation of positive and negative sample pairs, reduced cost and time of data labeling, deeper features extracted by the pre-task model, improved precision and efficiency of wool and cashmere fiber identification, and wide application prospect.

The application relates to an encoding method and device based on unit characteristics, electronic equipment and a storage medium. The method comprises the following steps: obtaining a to-be-divided encoding unit; calculating the texture complexity of the to-be-divided encoding unit, wherein the texture complexity comprises horizontal texture complexity and vertical texture complexity; if the texture complexity is less than a preset flat threshold, the to-be-divided encoding unit is not divided any more; if not, the to-be-divided encoding unit is divided by using a binary tree or an extended quadtree; if the extended quadtree is used, a trained SVM classifier is used to determine whether the to-be-divided encoding unit is subjected to horizontal extended quadtree division or vertical extended quadtree division; and the divided encoding unit is used to encode image information. The application fully utilizes the texture characteristics of the encoding unit itself and the surrounding encoding information, realizes fast division of the encoding unit before the encoding is completed, greatly reduces the trial space, and further improves the encoding speed.

The present invention discloses a method and system for identifying interactive behaviors of people in post office scenes, belonging to the field of video analysis technology. The method utilizes background subtraction to determine if abnormal behavior is triggered in the unpacking area, then captures a video stream. The captured video stream image is input into a pre-trained UNet network model to extract the box and human hands in the foreground. Feature descriptors for the box and human hands are constructed by combining depth and visible light information to determine the interactive relationships between people. Finally, a pre-trained SVM classifier is used to generate a judgment result, achieving accurate recognition of interactive behaviors in post office scenes. The method has a strong inhibitory effect on complex background interference, a high recognition accuracy rate for unpacking and visual inspection behaviors, and excellent robustness. It can meet the supervision requirements of express delivery stations for staff, and contributes to improving the automation and modernization level of intelligent video surveillance systems.

The invention belongs to the technical field of posture recognition, and provides a posture recognition method and system based on identity feature desensitization, a medium and a server, and the method comprises the steps: carrying out the preprocessing of three-dimensional point cloud data containing human body posture information, and carrying out the feature extraction through a deep learning algorithm, and carrying out standardization processing, covariance matrix eigendecomposition and principal component analysis dimension reduction processing on the extracted eigenvector in sequence to obtain a desensitized eigenvector, inputting the desensitized eigenvector into a pre-trained support vector machine classifier, and outputting an identification result including the basic posture of the human body. According to the method, after the original point cloud is converted into the feature vector, the posture data is detected through the SVM classifier, human body joint point recognition and data output are not involved, meanwhile, identity relevance in the point cloud data is effectively eliminated through identity feature desensitization processing, and privacy protection is higher; the deep learning algorithm is adopted to perform feature extraction and posture recognition, so that the recognition accuracy is improved.

The application discloses a kind of based on machine vision's grain quality index intelligent detection method, it is related to grain quality and safety detection technical field, including, using high-resolution camera to collect the upper surface and lower surface image of grain sample, pre-processing is carried out, obtains the two-dimensional image after pre-processing, utilizes convolutional neural network to extract two-dimensional feature, identifies grain external defect, and applies the physical model of light scattering theory to analyze grain internal defect, obtains preliminary quality evaluation result, the application is extracted two-dimensional feature by combining convolutional neural network to identify grain external defect, and applies the physical model of light scattering theory to analyze grain internal defect, to realize the overall evaluation of grain internal and external quality.SVM classifier is used to determine the final quality state of each grain, further improve user experience and work efficiency.

The present invention discloses a method and a system for detecting track risks based on big data analysis, which relates to the technical field of track safety. The present invention uses the YOLO model and the SVM classifier to detect the track, extracts multi-scale features of the track image, and can quickly and effectively detect the deformation and damage positions in the track; based on the detection results, the RF model and the BP neural network are respectively used to further analyze the fault points of the track, and the fault points in the track are accurately determined by combining the outputs of each model, reducing the misjudgment and omission of fault points, providing effective support for the subsequent detection path planning; according to the fault points, a detection path is formulated by combining the greedy algorithm and the ant algorithm, which can effectively reduce the detection time, improve the efficiency of the staff, improve the safety of train operation, and assist train operation management.

The application discloses a trajectory matching method combining target motion characteristics and joint discrimination, and relates to the field of geographic information processing. The application randomly generates trajectory starting points in three sub-regions, generates a simulation trajectory set according to the trajectory starting points and simulation trajectory parameters, obtains a positive and negative sample set in a cross matching mode by introducing noise, resamples the positive and negative sample set to obtain a simulation trajectory training matching sample pair set, calculates distance features and direction features of the simulation trajectory training matching sample pair, inputs the distance features and the direction features into an SVM classifier, constructs a trajectory matching classification model, inputs two trajectories to be matched into the trajectory matching classification model, and obtains a corresponding matching result. The application can effectively utilize target motion element information in the trajectory, construct a matching model, realize matching of a target trajectory, and is more accurate and reliable in the trajectory matching result, realizes matching of a most similar path target trajectory, and thus provides strong technical support for optimization and management of marine transportation.

The application belongs to the field of electroencephalogram signal processing, and is a fatigue driving classification method and system based on electroencephalogram multi-scale fuzzy entropy features of CEEMDAN. The method comprises the following steps: collecting electroencephalogram data, and labeling part of the data; pre-processing the electroencephalogram signal to remove artifacts in the electroencephalogram signal; extracting features from the pre-processed electroencephalogram signal, constructing and training an SVM classifier, classifying unmarked data using the trained SVM classifier to obtain pseudo-labeled data; performing CEEMDAN processing and scale transformation processing on the pseudo-labeled data and the previously obtained labeled data to obtain electroencephalogram multi-scale fuzzy entropy features based on CEEMDAN, and establishing a fatigue state classification model to classify the electroencephalogram signal. The collection method is simple, is more suitable for application in an intelligent driving system, and has a high fatigue state recognition rate.

The invention relates to the field of motor fault identification, and discloses a permanent magnet synchronous motor fault identification method, which comprises the following steps of: presetting a permanent magnet synchronous motor fault prediction model comprising a quantum heuristic neural network module, a double-flow convolutional neural network module and an SVM classifier which are connected in sequence, and identifying the fault of a permanent magnet synchronous motor through a quantum heuristic neural network mode, weak features of early faults of the permanent magnet synchronous motor can be effectively mined, unique rotation and entanglement operations of a quantum heuristic neural network are utilized to enhance feature expression ability, and feature extraction and fusion ability of a double-flow convolutional neural network module and accurate classification ability of an SVM classifier are combined. The detection sensitivity of early weak faults is remarkably improved, the fault degree of the permanent magnet synchronous motor is more accurately distinguished, the fault type is effectively recognized, and reliable guarantee is provided for stable operation of the permanent magnet synchronous motor.

This invention belongs to the field of radar target recognition technology, specifically relating to a radar identification method for distinguishing between complex-structured targets and simple-shaped targets. The method includes preprocessing the high-resolution range profile of the radar echo using an improved frequency-domain target length feature extraction method; extracting the target echo signal from sea clutter using the spectral differences between the target and clutter; calculating the range profile features and polarization features of each target; substituting the training feature set into a support vector machine (SVM) to obtain a trained SVM classifier; and using the classifier to classify the sample features obtained from the preprocessed radar echo signal to distinguish between complex-structured targets and simple-shaped targets. This invention addresses the situation where the radar echo contains both complex-structured target signals and simple-shaped targets, enabling effective identification between them.

The invention relates to the technical field of image incremental learning, and discloses a support vector guided discriminant playback and double-alignment distillation incremental method, which can realize balanced learning of new and old knowledge without relying on original data, and can improve the learning efficiency in each task stage. The method comprises the following steps: firstly, extracting image features by using a feature extraction network fused with a triple attention mechanism, carrying out classification training by adopting a support vector machine (SVM), generating a corresponding support vector set, and in a subsequent incremental training process, firstly, designing a joint feature retention strategy to carry out discriminative playback on support vectors, and constructing a training data set; then, support vector distillation loss is constructed, and the network is updated in combination with an existing loss function so as to maintain the discrimination capability of an old category; and finally, completing an identification task by using the updated SVM classifier in a classification stage. Experimental results on data sets of CIFAR-100, Tiny-Image Net and the like show that the method is remarkably superior to the prior art in the aspects of recognition accuracy and small sample adaptability.

The application belongs to the field of mechanical fault diagnosis of disconnectors, and provides a disconnector fault diagnosis method and system. The method comprises obtaining real-time vibration signals of the disconnector, and sequentially performing singular value filtering noise reduction and hybrid modal decomposition processing on the vibration signals to extract feature vectors; and performing fault diagnosis on the disconnector according to the feature vectors and a deep weighted fusion model based on an SVM classifier; wherein the construction process of the deep weighted fusion model based on the SVM classifier is as follows: obtaining an initial weak SVM classifier according to the feature vectors of the disconnector vibration signals and a linear support vector machine with a Gaussian kernel as an initial kernel function; and iteratively optimizing the initial weak SVM classifier by means of a deep fusion weighting algorithm and optimal allocation of the weights of the disconnector vibration signal samples to obtain an SVM classifier satisfying a preset condition and serving as the deep weighted fusion model based on the SVM classifier.