32 results about "Nonlinear classifier" patented technology

The invention provides a pulmonary nodule benignity and malignancy predicting method based on convolutional neural networks. The pulmonary nodule benignity and malignancy predicting method based on the convolutional neural networks comprises obtaining multi-scale pulmonary nodule image blocks from pulmonary nodule coordinates; according to the multi-scale pulmonary nodule image blocks, structuring the convolutional neural network corresponding to every multi-scale pulmonary nodule image block; according to a loss function, training the convolutional neural networks through the multi-scale pulmonary nodule image blocks; extracting low-dimensional features of pulmonary nodules through the trained convolutional neural networks; training a non-linear classifier through the low-dimensional features and predicting unknown pulmonary nodule image blocks. The pulmonary nodule benignity and malignancy predicting method based on the convolutional neural networks can accurately predict the benignity and malignancy of the unknown pulmonary nodule image blocks.

The invention provides a deep learning algorithm-based classification method of bacterial pneumonia and viral pneumonia in children. According to the method, a source data set is manually labeled; onthe basis of the combination of a full convolutional network semantic segmentation algorithm and a convolutional neural network algorithm, the full convolutional network semantic segmentation algorithm is adopted to perform lung region foreground segmentation on an image so as to obtain a region of interest, the extracted region of interest is inputted to a convolutional neural network model so asto train a classifier, and therefore, the category of an unknown chest X-ray image can be predicted, and the high-dimensional features of the region of interest are extracted; and a traditional imageprocessing method is adopted to extract the low-dimensional features of the region of interest; and the high-dimensional features and the low-dimensional features are used to train a non-linear classifier; and the category of the unknown X-ray image is predicted, and the type of the pneumonia of a patient can be judged. Since a main component analysis algorithm is used to perform dimensionality reduction on the features, and therefore, the amount of calculation can be reduced; and the features which have been subjected to mixed dimensionality reduction are inputted into the nonlinear classifier, and the category of the unknown X-ray image can be predicted.

A determination is made in real-time regarding whether a bicyclist is present in a target image. A target image is received. The target image is classified and an error value for the target image is determined using a linear classifier. If the error value does not exceed the threshold value, the classification is outputted. Otherwise, if the error value exceeds the threshold value, the target image is classified using a non-linear classifier.

The invention discloses a method and a system for the rapid electrocardiograph identity recognition. The method mainly comprises a model training stage and a real-time testing stage. According to theinvention, the identity information of an individual is recognized and verified by collecting the electrocardiograph signals of the individual, so that the method belongs to the technical field of biological characteristic identity recognition. According to the method, firstly, electrocardiograph windows are cut at any initial point position, wherein any window is larger than one heart beat length. Each electrocardiograph window is randomly divided into fixed-length fragments, and the characteristic extraction is conducted by utilizing a full-automatic characteristic extraction layer. Secondly, extracted characteristics are preliminarily classified by utilizing a non-linear classifier. Thirdly, a recognition result is generated through the highest entropy voting process. Finally, the rapidelectrocardiograph identity recognition is achieved through the parallel treatment of the characteristic extraction and the non-linear classifier. According to the invention, the detection of characteristic points and the fusion of characteristics are not required, so that the recognition speed is high. The electrocardiograph signal-based identity recognition can be carried out in real time, andthe verification crowd is wide.

This invention provides a system and method for providing Indicator Features in high resolution micro-ultrasound prostate images, wherein the ultrasound device operates at a center frequency of 15 MHz and above. Indicator Features are features, which have been identified alone and/or in combination with other features in high resolution micro-ultrasound images, and have been determined to be significantly statistically correlated to either benign tissue or some grade of cancerous tissue on the basis of predictive probabilities. These Indicator Features can be used to train a linear or non-linear classifier, which can be used to classify patient images of the prostate. These Indicator Features can be used to guide a clinician as to where to take a biopsy core from the prostate, as well as assist in diagnosis of the tissue.

The invention discloses a risk identification method and device based on multiple classifiers, computer equipment and a storage medium, and relates to the technical field of intelligent decision making, which comprises the following steps: if an item handling instruction sent by a user side is detected, obtaining user video data; acquiring corresponding user identity information to judge whether user identity verification is passed or not; if so, obtaining current video data in the current transaction process; extracting emotion information and content information in the current video data; calling a combined classifier composed of a plurality of nonlinear classifiers and linear classifiers; inputting the emotion information and the content information of the current video data into a combined classifier to obtain a corresponding user risk category; and if the user risk category corresponding to the current video data belongs to the high risk category, terminating the connection with the user side and giving a prompt of terminating event handling. The method achieves the judgment of the user risk types of the user video data based on the combined classifier, and improves the recognition accuracy.

The invention discloses a picture garbage recognition method based on deep learning, and the method comprises the following steps: S1, building a large-scale picture training set, training a deep learning classifier based on a convolutional neural network in the training set, and extracting a convolutional layer in the deep learning classifier as a deep feature extractor; S2, constructing a singlenonlinear classifier, and inputting pictures in the high-quality picture library and the picture garbage library into the depth feature extractor obtained in the step S1 to obtain a feature representation vector; S3, taking the feature vector obtained in the step S2 as the input of a nonlinear classifier, and carrying out supervised learning by utilizing a corresponding tag to obtain an availablepicture garbage classifier; And S4, cutting the long image through a sliding window to obtain a plurality of pictures, extracting each segment of feature by using a depth feature extractor, classifying by using a picture garbage classifier, and comprehensively integrating each result to obtain a final result.

The invention provides a generalized zero sample target classification method based on active learning and a variational auto-encoder. The method is used for solving the problem of bias caused by loss of unknown class supervision information and the problem of low-dimensional feature aggregation caused by projection from high-dimensional features to low-dimensional space in the prior art, and effectively improving the classification accuracy. The method comprises the following implementation steps: acquiring a training sample set Ptrain and a test sample set Ptest; constructing a generalized zero sample classification model H based on the variational auto-encoder; carrying out iterative training on the variational auto-encoder f and a nonlinear classifier fclassifier in the generalized zero sample classification model H based on the variational auto-encoder; and obtaining a target classification result of a generalized zero sample. The method can be applied to the fields of classification of rare species lacking training data, biomedical image recognition and the like.

The invention discloses a judgment and identification method and system for a suspicious violent and horrific person based on ontology identity features. The method includes the following steps of: (1) collecting a plurality of ontology identity feature information for each object of interest and defining it as a corresponding ontology feature variable; (2) assigning the ontology feature variablesbased on the ontology identity feature information, and using a discretized variable assignment method; (3) building a training data set; and (4) using at least one data classification method for classification training of the training data set to construct a non-linear classifier. The method and system considers and analyzes the possible correlation between the feature variables of the object ofinterest, performs the discretized assignment of the feature variables, forms the training data and constructs a non-linear classifier accordingly, and accurately and efficiently discriminates the risk of persons involved in terrorism through the non-linear classifier.

The invention provides a multi-task dictionary sheet classification method. The method comprises the steps of obtaining to-be-classified tasks; each to-be-classified task learning a comprehensive dictionary and an analysis dictionary; utilizing adictionary learning model to analyze the non-correlation item, the analysis coefficient code extraction item and the multi-task single classification itemto establish a target optimization function; wherein the dictionary learning model comprises a comprehensive dictionary and an analysis dictionary; solving the optimization function to respectively obtain a linear classifier and a nonlinear classifier; and classifying the to-be-classified tasks by utilizing the linear classifier and the nonlinear classifier. According to the method, one task is adopted to learn one comprehensive dictionary and one analysis dictionary, and the coding coefficient is as sparse as possible for other tasks, so that the potential structure of the data can be betterrepresented. And meanwhile, a multi-task learning model is utilized, so that the calculation complexity is greatly reduced. The invention further provides a multi-task dictionary sheet classificationsystem and device and a computer readable storage medium, which have the above beneficial effects.

The embodiment of the invention discloses a defibrillating heart rhythm recognition method based on order statistics and a support vector machine and a defibrillating heart rhythm recognition device based on order statistics and the support vector machine, and solves the technical problems that the present ShR/NshR recognition method needs to acquire long electro-cardio signals (usually 8 seconds)to guarantee certain classification performance, which delays the defibrillation rescue time. The method comprises the steps that S1, differential operation is performed on the acquired first electro-cardio signals so as to obtain the second electro-cardio signals; S2, the second electro-cardio signals are arranged according to an ascending order so as to obtain the order statistics; and S3, theorder statistics is inputted to a nonlinear classifier so as to obtain the classification result of ShR and NshR.

The invention discloses a bad information website detection method, and the method comprises the steps: carrying out the preliminary classification of a to-be-detected website through a nonlinear classifier according to webpage data based on a statistical principle, obtaining a classification result, and determining the information type of bad information; since a corresponding recurrent neural network model is set for each information type in advance, further using the recurrent neural network model corresponding to the classification result to perform secondary classification on the to-be-detected website to obtain a final detection result. Therefore, the purpose of quickly searching for and accurately positioning illegal and bad information websites in massive webpages is achieved, thedetection efficiency and the detection accuracy are effectively improved, and great convenience is brought to network supervision. In addition, the invention further provides a bad information websitedetection device and equipment and a readable storage medium, and the technical effect of the bad information website detection device and equipment corresponds to the technical effect of the method.

The invention provides a fraudulent transaction identification method, system and device based on dynamic weighted information entropy. The method comprises the following steps: screening one-class-SVM models through the dynamic weighted information entropy, and selecting a one-class-SVM model Mocsvm with the maximum dynamic weighted information entropy of an overlapped data subset; dividing original data into an overlapped data subset and a non-overlapped data subset by using a one-class-SVM model Mocsvm; training a non-linear classifier model Mclf is by using an overlapping data subset obtained through division of a one-class-SVM model Mocsvm, and distinguishing fraudulent transactions and normal transactions in the overlapping data subset by using the non-linear classifier model Mclf; and generating a fraudulent transaction identification model composed of the one-class-SVM model Mocsvm and the nonlinear classifier model Mclf. According to the method, a division and treatment strategy is adopted, and a large amount of normal transaction data easy to recognize is discharged for a nonlinear machine learning model, so the model can only pay attention to learning of data difficult to divide, the capacity of the nonlinear model is fully played, and the performance of a fraudulent transaction identification model is improved.

The invention relates to an on-line soft-spaced kernel learning algorithm (OSKL) based on step size control. By introducing kernel function to construct nonlinear classifier and soft interval parameter to control the effect of noise data, a robust on-line kernel learning algorithm is designed based on the basic framework of on-line gradient descent algorithm. The algorithm can reduce the storage space of the model and effectively control the influence of noise. The computational complexity of model updating is only O (1). It is a natural tool to deal with and analyze the data flow problems because of its advantages of real-time and easy implementation. The online learning algorithm of the invention overcomes the problem that the traditional classification method based on batch processing technology can not efficiently process the data stream, and also overcomes the problem that the existing online learning algorithm such as Kernel Perceptron and Pegasos and the like can not effectivelysuppress the influence of noise.

A linkage mapping process incorporating an RBF non-linear classifier pattern-matching approach for use in the genetic analysis and modification of botanical organisms by comparing infrared mass spectroscopy, isomeric-level quantitative chemical analysis patterns, with single-molecule real-time genetic and genomic isoform analysis sequence patterns to build a searchable pattern library of those phenotype patterns producing commercially desirable traits for directed selection of desirable attributes and identification of their associated phenotype patterns.