57results about How to "Improve classification recognition rate" patented technology

The invention discloses an indoor human body behavior recognition method. The method comprises the following steps that: human body three-dimensional skeleton information is obtained based on Kinect equipment; three-dimensional skeleton features in each video set are extracted; the three-dimensional skeleton features are trained, and the features are described, and the training of the three-dimensional skeleton features further includes the following steps that: online dictionary learning is performed on the features, and then, sparse principal component analysis is performed on the features, and finally, a multi-task large margin nearest neighbor algorithm and a linear support vector machine are utilized to classify the features, so that a training feature set can be obtained; three-dimensional skeleton features of test videos are extracted; and the multi-task large margin nearest neighbor algorithm and the linear support vector machine are utilized to classify the features, so that feature descriptions can be obtained, and optimum judgment is performed on the training feature set and the test features with a scoring mechanism. The indoor human body behavior recognition method of the invention has a bright application prospect in intelligent video surveillance, patient monitoring systems, human-computer interaction, virtual reality, smart home, intelligent security and prevention and athlete assistant training, and has high feasibility and great social economic benefits.

The invention relates to an electroencephalogram feature classification method based on a PSO-SVM (Particle Swarm Optimization-Support Vector Machine). The electroencephalogram feature classification method comprises the following steps: firstly, utilizing a regularizable CSP (R-CSP) (Cryptographic Service Provider) algorithm to carry out electroencephalogram feature extraction; secondly, utilizing the particle swarm optimization to optimize the penalty factor C and the nuclear parameter g of the support vector machine; and finally, training a SVM classifier by an obtained optimal parameter, and utilizing the trained classifier to carry out classification prediction on a sample. Compared with traditional SVM classification identification, a result indicates that the classification identification algorithm based on the PSO-SVM can effectively improve the classification identification rate of the electroencephalogram. Compared with the traditional classification identification method, the electroencephalogram feature classification method has obvious advantages.

The invention discloses a sequence deeply convinced network-based pedestrian identifying method. The method comprises the following steps of preprocessing a training image in a pedestrian database to obtain a training sample image, extracting an HOG (Histograms of Oriented Gradients) feature from the obtained training sample image, building and training a sequence restricted Boltzmann machine-based sequence deeply convinced network, using the sequence deeply convinced network to further extract features from the obtained HOG feature to form a feature vector of the training sample, inputting the obtained feature data into a support vector machine classifier, and finishing training; preprocessing a to-be-tested pedestrian image to obtain a test sample; using an HOG and the sequence deeply convinced network to extract pedestrian features from the test sample to form a feature vector of the test sample; inputting the feature vector of the test sample into the support vector machine classifier, and identifying whether the test image is a pedestrian or not. According to the method, better classification performance can be obtained, the accuracy of pedestrian identification is improved, and the robustness of a pedestrian identifying algorithm is enhanced.

The invention discloses a small sample radio signal enhancement identification method based on ACGAN. The method comprises the following steps of 1) a data preprocessing stage of converting a sample format, dividing a data set, and carrying out onehot coding on the category of the data to obtain a corresponding category label sequence; and 2) based on a signal enhancement method of a generative adversarial network, adjusting a radio signal sample to a set size, performing pixel normalization on the radio signal sample, setting the initial parameters of each network layer of the deep convolutional network, inputting the training data divided in the step 1) into the deep convolutional network, and training the network; and 3) extracting the features of the training data by using an LSTM network to construct a signal identification classifier, and classifying the test data and the verification data by using the classifier.

The invention discloses a data discretization method based on category-attribute relation dependency, belonging to the field of data mining. The method is characterized by comprising: first, based on CAIM algorithm, comprehensively considering about the influence of attribute importance and inconsistent rate of a decision table for discretization result, and providing an improved CAIM algorithm; and second, adopting a lambda correlation coefficient as discretization discriminant for evaluating category-attribute relation, and providing a new monitor discretization algorithm which does not need artificial input parameters and can automatically select discretization points. The method has the advantages that the method can maintain the high efficiency of extracting information for an original data set, balances the consideration of accuracy, and can obtain higher accuracy when in machine learning.

The invention discloses an image classification method based on combination of SRC and MFA, which is mainly used for solving the problem that the image classification result is not ideal because only the reconstruction relationship or local discriminant structure is considered and the sample information cannot be accurately described in the conventional feature extraction method. The method comprises the following steps: 1. inputting a training sample and a test sample, constructing the same kind and different kinds of sample matrixes, and initializing a projection matrix; 2. projecting the training sample, respectively taking the same kind and different kinds of samples as dictionaries, solving sparse representation coefficients of the samples, and constructing the same kind and different kinds of sparse weight matrixes; 3. constructing an objective function to solve a novel projection matrix; 4. iterating the steps 2 and 3 until the cycle index is larger than the set initial value, outputting the final projection matrix, and projecting the test samples; and 5. classifying the test samples by utilizing a sparse representation classifier. According to the method disclosed by the invention, the accuracy of image classification is enhanced, and the method can be used for discriminating identity of characters or searching objects during image shooting in a police work system.

The invention provides a zero sample image classification method based on a relative attribute random forest. An attribute ranking score model is built for an image with an unknown class according to the relative relationship between image classes and image attributes, the attribute ranking score models for all images serve as training samples to train a random forest classifier, and finally, according to an attribute ranking score for a tested image and the random forest classifier obtained through training, the label of the tested image is predicted. The method of the invention can realize zero sample image classification and has the advantages of high classification recognition rate, strong model stability and the like.

The invention discloses a motor imagery electrocorticogram (EEG) signal classification method based on independent component analysis, which comprises the following steps: S1, collecting the EEG signals and preprocessing the EEG signals, and randomly dividing the preprocessed EEG signals into a training set and a test set; S2, sequentially selecting the training set data to carry out independent component analysis and calculation on the single test sample data, and automatically identifying and acquiring the motion related component based on the spatial distribution pattern of the source; S3:classification and recognition of motor imagery based on zero training classifier; S4: optimizing the selection of leads by using the training set data, substituting the optimized leads into the testset, and looping the steps S2 and S3 to obtain the final classification recognition rate. The invention can reduce the spatial model matching problem caused by the difference between the collected EEGdata, and has high recognition accuracy to the motor imagery EEG signals.

The invention discloses a hyperspectral data dimensionality reduction method based on ultra-pixel and maximum boundary distribution, which overcomes a defect of high inter-band correlation for lack of enough spatial and spectral domain information in the prior art. The hyperspectral data dimensionality reduction method comprises the steps of (1) dividing a sample set; (2) generating a regular matrix; (3) generating a judgment matrix; (4) solving an optimal projection matrix; and (5) carrying out projection dimensionality reduction. The hyperspectral data dimensionality reduction method has the advantages of abilities of maintaining the consistency in space and spectrum of neighboring samples and reducing redundant bands, and can be applied to dimensionality reduction of hyperspectral remote sensing images.

The invention discloses an observation vector difference based method for classifying SAR image textures. The method mainly solves the problem of terrain classification of SAR images. The method comprises the steps of (1) randomly selecting r images from a training set for partitioning processing, converting to obtain a column vector difference matrix P; (2) observing the P with an observation matrix to obtain a texture observation vector difference matrix X, and conducting clustering on the X to obtain a texture dictionary D; (3) calculating images of the training set according to Step (2) to obtain an observation vector difference matrix Xtr; (4) projecting the Xtr onto the texture dictionary D to form a training image texture column diagram h; (5) representing images of a test set by a test image texture column diagram he; (6) calculating the distance between the he and the h, and determining the classification to which the he belongs according to the distance; and (7) calculating all test images according to Step (6) to obtain a final classification rate. According to the method, the latest compressed sensing theory is applied, the process is simple, the classification identification rate is high, and the method is applicable to terrain texture classification of SAR images.

The invention provides an electronic nose signal feature fusion method based on the separability degree and the dissimilarity degree and belongs to the technical field of electronic noise signal and information processing. The method comprises a step one of subjecting an electronic nose signal to feature extraction, a step two of performing feature selection and a step three of performing feature weighted fusion. The invention reserves classification information to the greatest degree while reducing dimensions and eliminating redundancy, and thus the classification identification rate is greatly improved, and the classification identification performance of an electronic nose is improved further.

The invention provides a communication emitter signal classification and identification method under the condition of Gaussian white noise channel, which comprises the following steps of transmittingand receiving the communication emitter signal; extracting envelope information from the received signal of the communication emitter; constructing the characteristic data set dictionary of communication emitter signal; classifying and identifying the received signals from untagged communication emitters. On the premise that the labeled signal characteristic dictionary database is artificially continuously supplemented and perfected, the invention can realize very good classification and identification of the communication emitter signal under the condition of Gaussian white noise channel.

The invention relates to a continuous attribute discretization method based on Chi2 statistics, which belongs to the field of data mining, and is characterized in that Chi2 series algorithm is analyzed, a discretization standard is re-defined, and continuous attributes are discretized more reasonably and more effectively; and a difference sequence interval combination method serves as a standard, so that all adjacent areas have fair combination opportunities, thereby well solving the problem of unfairness. In addition, the imprecise Eji value in Chi2 statistics is analyzed, and two improving schemes are presented. The continuous attribute discretization method based on Chi2 statistics has the advantages and the benefits that the high efficiency in extracting information from original data, the discretization process is more precise, and higher accuracy is realized in machine learning.

The invention discloses a human-vehicle classification and identification method for low-resolution radar ground targets, and relates to target classification and identification of ground radars in the field of radio measurement. The method mainly comprises the processing steps of target multi-dimensional feature extraction, feature smoothing and selection, speed coarse classification, probabilityadjustment, SVM classifier, D-S evidence pushing, threshold judgment and the like. According to the method, the moving targets forming the stable track are classified, and ground moving target classification with high recognition probability and low cost is realized. The method has the characteristics of high target recognition probability, target rejection outside the library, high calculation speed and simple engineering implementation, solves the problem that a low-resolution radar system does not have the target classification and recognition capability or is poor in recognition performance, and is particularly suitable for the ground pedestrian and vehicle target classification and recognition process of ground surveillance radar and battlefield reconnaissance radar.

The invention discloses a pedestrian recognition method based on positive-negative generalized max-pooling. The method comprises: preprocessing acquired traffic videos to obtain required training sample images, extracting local features of the training sample images by means of gradient-based HOG local descriptors, encoding the local features by the depth hierarchical encoding method which is completed via a spatial aggregating restricted Boltzmann machine, forming feature encoding vectors of training samples, obtaining high-level image feature representation vectors by adopting the positive-negative generalized max-pooling method, and inputting feature data obtained to a support vector machine classifier to finish the training; preprocessing to-be-tested pedestrian images to obtain test samples, and obtaining feature representation vectors of the test samples by the same method; inputting the feature representation vectors of the test samples to the support vector machine classifier which is already trained, and identifying whether test images are pedestrian images or not. The invention improves the accuracy of pedestrian recognition and enhances the robustness of the pedestrian recognition algorithm.

The invention provides a high-voltage cable partial discharge mode recognition method, and the method comprises the following steps: obtaining training data through laboratory data, dividing the training data into a training set and a test set, and collecting onsite to-be-recognized data; respectively inputting the training set into a plurality of classifiers for training, respectively inputting the test set into the trained classifiers for testing, and inputting to-be-identified data acquired on site into the trained classifiers to obtain a preliminary identification result; taking the preliminary identification result as input data, constructing a partial discharge mode identification model by adopting a learning method taking a support vector machine as a classifier, and then inputtingthe input data into the partial discharge mode identification model for testing to obtain a tested partial discharge mode identification model; and inputting to-be-identified data acquired on site into the tested partial discharge mode identification model, and outputting to obtain a high-voltage cable partial discharge integrated mode identification result.

The embodiment of the invention provides a rehabilitation prediction method and system of a deaf patient after a CI operation based on machine learning, and relates to the field of computer signal processing. The method comprises the steps of obtaining multiple electroencephalogram signals of the patient with congenital deafness, and pre-processing the obtained electroencephalogram signals; extracting characteristic quantities of the electroencephalogram signals from the preprocessed electroencephalogram signals; conducting normalization preprocessing on all electroencephalogram signal data; selecting a kernel function; selecting a training set and a testing set from the electroencephalogram signal data; loading the training set and the testing set into a support vector machine, and conducting rehabilitation prediction. The rehabilitation prediction method and system of the deaf patient after the CI operation based on machine learning have the advantages of being high in recognition rate, accuracy rate, practicability and the like.