50 results about "Nonlinear support vector machine" patented technology

The invention provides an image object recognition method based on SURF features. Firstly, the image is preprocessed, and then SURF corner points and SURF descriptors are extracted from the image to describe the image features, and then the features are processed by PCA data whitening and dimensionality reduction. The processed features are clustered by Kmeans to establish a bag-of-words model, and the bag-of-words model is used to construct the visual vocabulary histogram of the image. Finally, the non-linear support vector machine (SVM) classification method is used for training, and the classification of different categories of images is completed. After the classification model modeling of different images in the training stage is completed, the images in the test set are detected in the test stage, and the function of object recognition in different images is realized. The invention has excellent performance in both recognition rate and speed, so that it can reflect the content of the image more objectively and accurately. In addition, the classification result of the SVM classifier is optimized, and the error rate of classifier judgment and the category of training samples are reduced. limitations.

The present disclosure proposes a multivariable monitoring method and system based on Savitzky‑Golay filter, including: obtaining historical operating data of the multivariable system, the historical operating data contains a plurality of process variables indicating the operating state of the system, and marks the history The normal or abnormal condition of the operating data; use the Savitzky-Golay filter to filter the historical time series data of each process variable to obtain the filtered signal amplitude and derivatives of each order; filter the filtered signal amplitude of each process variable and each The order derivative forms the feature space; in the feature space, a nonlinear support vector machine model is obtained based on historical data training, and is used to judge whether to give an alarm to realize multi-variable monitoring. The proposed method has the advantages of high accuracy, overcomes the defects of the traditional univariate alarm threshold design method, can effectively improve the limitation of the monitoring system, and greatly reduce the number of false and missed alarms.

The invention discloses a low-power epilepsy detection circuit based on a master-slave support vector machine, which belongs to the field of intelligent medical applications. The circuit includes: a clock module, a feature extraction module, a master-slave support vector machine module and a decision module; the master-slave support vector machine module includes a master support vector machine and a slave support vector machine, the master support vector machine is a linear support vector machine, and the slave The support vector machine is a non-linear support vector machine; the main support vector machine controls the startup and shutdown of the slave support vector machine; during the detection process, the main support vector machine detects the beginning of the seizure, starts the slave support vector machine, and corrects the slave support vector machine The end of the epileptic seizure; the detection result of the master-slave support vector machine module is the logic AND of the detection result of the master-slave support vector machine. This application utilizes the master-slave support vector machine and continuous sequence detection, so that under the premise of ensuring the detection performance, the calculation complexity is greatly reduced, the power consumption is reduced, and the requirements of intelligent medical applications are better adapted.