12 results about "Support vector machine" patented technology

The invention relates to an online fault detection method for a reduced set-based downsampling unbalance SVM (Support Vector Machine) transformer. At present, the research of improving the performance of an unbalance data downsampling SVM algorithm comprises upsampling and downsampling. The SVM model calculating cost of the upsampling algorithm is increased. The downsampling algorithm is selected improperly sometimes, and thus the poor classifying effect is caused. The online fault detection method comprises the following steps: (1), acquiring a vibration signal of a transformer; (2), obtaining a noise reduction vibration signal; (3), obtaining multiple groups of fault detection feature data; (4), clustering by using a K-mean algorithm; (5), figuring out a weight value of each sample; (6), establishing a majority sample reduction vector solution optimization model; (7), obtaining an SVM fault diagnosis model; and (8), inputting a sample to be tested to an unbalance SVM detector trained in the step 7, analyzing a result output from the detector to obtain a working state of the transformer, and realizing online fault detection of the transformer. The online fault detection method is used for detecting the fault of the transformer online.

The invention discloses a real-time gesture recognition method which comprises the steps of (1) decomposing an obtained gesture video into image sequences sorted in a chronological order and preprocessing obtained images and then carrying out hand-region segmentation, (2) extracting a hand shape feature of a hand region in each image and using an SVM support vector machine to identify the hand shape feature as a corresponding gesture value, (3) combining the gesture value of each image and a direction feature of a motion trajectory obtained by an iterating LK pyramid optical flow algorithm asa feature vector of each dynamic gesture image, (4) carrying out loop execution of steps (2) and (3) with a loop end condition that all images of a current video are processed so as to obtain a complete set of feature vector sequences, (5) creating a gesture template library, (6) carrying out optimization DTW match on the obtained feature vector sequences and all templates in a template library, calculating the degree of distortion of the match, wherein recognition is failed if the degree of distortion is larger than a distortion threshold, and a recognition result is outputted if the degree of distortion is smaller than the distortion threshold.

The invention provides a method for acquiring an optimal classification surface of a hypersonic speed air inlet starting/non-starting mode, which relates to the technical field of hypersonic speed air inlets and aims to solve the problem that a control system cannot fast detect the current working states of the air inlets because the optimal classification surface of the hypersonic speed air inlet starting/non-starting mode cannot be accurately acquired in the prior art. The method comprises the following steps of: acquiring training samples classified in the starting/non-starting mode by a wind tunnel test or numerical simulation; defining operation modes and classification rules according to flow capturing characteristics; selecting characteristics by a support vector machine method; and optimizing different sensor combinations by a linear discriminant analysis method to acquire the optimal classification surface. By the method, the optimal classification surface of the hypersonic speed air inlet starting/non-starting mode can be fast acquired, an isolation strip between the hypersonic speed air inlet starting mode and the hypersonic speed air inlet non-starting mode can be obtained, and the defects of measurement noise and interference are overcome to some extent.

The present invention relates to the technical field of image processing, and particularly to a lung tumor recognition method based on support vector machine MRI image segmentation. The method mainly adds a non-linear optimization model. A segmented image is optimized via the non-linear optimization model. The accuracy of the segmented image obtained by using this method is higher, so that the recognition accuracy of a lung tumor is higher.

The invention discloses a method for detecting a crack fault of a bogie of a maglev train. The method comprises the following steps: S1, collecting a vibration acceleration signal of a bogie of a maglev train by a sensor and acquiring a modal parameter of the bogie of the maglev train in a running state from the vibration acceleration signal; S2, carrying out the principal component analysis (PCA)on the modal parameter obtained in the S1 and extracting a principal component vector sensitive to a crack fault of the maglev train bogie; S3, establishing a support vector machine (SVM) model and carrying out SVM training on the principal component vector extracted in the S2 to obtain a classification surface of the principal component vector point sensitive to a crack fault of the maglev trainbogie; and S4, according to the position of the principal component vector obtained in the S2 in the classification surface in the S3, determining whether the maglev train bogie has a crack fault. Therefore, whether the bogie has a crack fault is determined accurately. The method has characteristics of high sensitivity, high precision and high efficiency.

The invention relates to the field of anomaly detection and new high-risk alarm identification by adopting a One Class SVM (Support Vector Machine), and particularly provides an anomaly alarm identification method and device based on incremental learning, a server and a storage medium, the method comprises the following steps: firstly, training a One Class SVM model by utilizing an impure original data set to identify an anomaly alarm; and then, enabling a safety operator to regularly mark abnormal alarms obtained by the model, and iteratively optimize the model based on incremental learning, so that the model detection performance is improved, and missing report and false report are reduced.

The invention discloses a multi-modal neural image feature selection method based on sample weight and low-rank constraint. The method comprises the following steps: (1) collecting multi-modal brain image data; (2) after data of multiple modals are obtained, feature selection being carried out in a multi-modal feature collaborative analysis mode, a regression model from each modal datum to a classification class label being established, group sparse constraint being carried out on regression vectors, and therefore a common feature subset related to all tasks being obtained; (3) multi-modal data feature modeling; (4) writing the target function into an augmented Lagrange form, and changing the target function into a convex function; (5) obtaining the weight matrix of each mode of each sample and the weight of each feature; and (6) using a multi-kernel support vector machine method to fuse the sample multi-modal features for classification. According to the technical scheme, the importance of samples among modals and the correlation among the modals are considered, more meaningful features are found out, and the classification and prediction accuracy is improved.