428results about How to "Reduce feature dimension" patented technology

The invention provides a method and a system for deep nerve translation based on character encoding. A combined nerve network model is established by using an RNN to cover the whole translation process, and translation tasks are directly completed from the perspective of an encoder-decoder framework. The method comprises the following steps: A, word vector generation: performing word segmentation on character-level input data through neural network modeling and generating a word vector; B, language model generation: establishing grammar rules by utilizing the characteristic of memory of the recurrent neural network in time; C, word alignment model generation: obtaining the probability of translating multiple words in a source language statement into target language words; D, output: translating an inputted source language into a target language; E, translation model combination: establishing a deep nerve translation model (RNN-embed) based on character encoding in combination with neural network models in the four steps and accelerating model training by using CPU parallel computation.

The invention discloses a voiceprint identification method. The voiceprint identification method comprises the following steps of: 1, preprocessing segmented speech data of each speaker in a training speech set to form a group of sample sets corresponding to each speaker; 2, extracting Mel-frequency cepstrum coefficients from each sample in all sample sets; 3, selecting a sample set one by one and randomly selecting the Mel-frequency cepstrum coefficients of part samples of the sample set, and training a Gaussian mixture model for the sample set; 4, performing incremental learning on the samples which are not selected and trained in the step 3 and the Gaussian mixture model of the sample set corresponding to the sample set one by one to obtain all optimized Gaussian mixture models, and optimizing a model library by utilizing all optimized Gaussian mixture models; and 5, inputting and identifying test voice data, identifying the Gaussian mixture model of the sample set corresponding to the test voice data by utilizing the optimized model library in the step 4, and adding the test voice data to the sample set corresponding to the speaker.

The invention discloses an image inquiry method based on clustering and includes the following steps: (1) extracting color characteristics and grain characteristics of target images and each image in an image database; (2) reducing the dimensions for high-dimensional characteristic to describe image contents, so as to gain characteristic subsets; (3) clustering data sets together to form characteristic subsets; (4) establishing indexes for each cluster in the clustering respectively; (5) using the index to perform image inquiry. The invention has the advantage that simple and effective partition can be performed for images according to concerned areas in images, so as to optimize the index tree structure and make the searching more accurate and effective.

The invention discloses a vivo-face detection method based on HSV (hue, saturation, value) color space statistical characteristics. The method includes firstly, converting a face image captured by a camera from RGB (red green blue) color space to YCrCb color space; secondly, subjecting the face image to skin color segmentation, denoising, mathematical morphological treatment and connected region boundary calibration so as to obtain coordinates of a face matrix region; thirdly, acquiring a facial image to be detected from the face image according to the coordinates of the face matrix region; fourthly, segmenting the facial image to be detected into image blocks, and acquiring characteristic values of three color components of each image block in the facial image to be detected; and fifthly, using normalized characteristic values as a sample to be detected, and sending the sample to a trained support vector machine for detecting whether the face image is a vivo real facial image or not. The vivo-face detection method based on HSV color space statistical characteristics has the advantages that delay of a face recognition system is reduced, computational complexity is lowered, and detection accuracy is improved.

The invention discloses a multi-gesture and cross-age oriented face image authetication method. The method comprises the following steps of: rapidly detecting a face, performing key point positioning, performing face alignment, performing non-face area filtration, extracting the face features by blocks, performing feature dimension reduction and performing model prediction. The method provided by the invention can perform the face alignment, realize the automatic remediation for a multi-gesture face image, and improves the accuracy rate of the algorithm, furthermore, the feature extraction and dimension reduction modules provided by the invnetion have robustness for aging changes of the face, thus having high use value.

The invention discloses an image type fire flame identification method. The method comprises the following steps of 1, image capturing; 2, image processing. The image processing comprises the steps of 201, image preprocessing; 202, fire identifying. The fire identifying comprises the steps that indentifying is conducted by the adoption of a prebuilt binary classification model, the binary classification model is a support vector machine model for classifying the flame situation and the non-flame situation, wherein the building process of the binary classification model comprises the steps of I, image information capturing;II, feature extracting; III, training sample acquiring; IV, binary classification model building; IV-1, kernel function selecting; IV-2, classification function determining, optimizing parameter C and parameter D by the adoption of the conjugate gradient method, converting the optimized parameter C and parameter D into gamma and sigma 2; V, binary classification model training. By means of the image type fire flame identification method, steps are simple, operation is simple and convenient, reliability is high, using effect is good, and the problems that reliability is lower, false or missing alarm rate is higher, using effect is poor and the like in an existing video fire detecting system under a complex environment are solved effectively.

The invention discloses a Chinese word segmentation method based on two-way LSTM, CNN and CRF which improves and optimizes traditional Chinese word segmentation base on deep learning algorithm. The method comprises following specific steps: preprocessing the initial corpus, extracting corpus character feature information and pinyin feature information corresponding to characters; using the convolutional neural network to obtain pinyin feature information vector of the characters; using the word2vec model to obtain the character feature information vector of text; splicing pinyin feature vectors and character feature vectors to obtain context information vectors and put the context information vectors to a bidirectional LSTM neural network; decoding the output of the bidirectional LSTM using the linear chain condition random field to obtain the word segmentation sequence; decoding the word segmentation label sequence to obtain word segmentation results. The invention utilizes the deep neural network to extract text character features and pinyin features and combines the conditional random field decoding, can effectively extract Chinese text features and achieve good effect on Chinese word segmentation tasks.

The invention discloses target detection system and method based on covariance and a binary-tree support vector machine. The system comprises a video data collecting unit, an image preprocessing unit and a background modeling vehicle partitioning and displaying unit. The video data collecting unit is used for acquiring information in real time and carrying out digitization and system conversion on analog videos. The image preprocessing unit comprises an FPGA (Field Programmable Gate Array) and a DSP (Digital Signal Processor), wherein the FPGA is used as a coprocessor; and the DSP is used as a main processor to accomplish the background modeling of video images, the partitioning and the extraction of vehicle targets and the realization of a model identification algorithm. Utilizing the combination of the FPGA and the DSP, the invention can realize multi-video real-time model identification through combining with the model identification algorithm based on the covariance features of the images and the support vector machine. The invention can be widely used for a plurality of fields of intelligent traffic management, intelligent video monitoring and the like.

The invention discloses a polarimetric SAR classification method based on a semi-supervised convolutional neural network. According to the method, first, image data is input, and super pixel segmentation is performed; training samples and test samples are extracted; a neighborhood preserving and sparse filtering deep learning network (NDSFN) is adopted to perform unsupervised pre-training on parameters of the convolutional neural network, and a needed filter set is obtained; a feature map is obtained through convolution operation; fuzzy processing is performed on the feature map through downsampling; parameter settings of the NDSFN are adjusted according to the filter size in a second convolution layer to obtain a new feature map; a Softmax classifier is utilized to perform image classification on the training samples; and a small quantity of marked samples are adopted for slight adjustment to obtain a polarimetric SAR terrain classification. Through the method, by constructing a novelsemi-supervised convolutional neural network (SNCNN) model, the problem that a traditional supervised convolutional neural network has a high requirement for marked samples is effectively solved.

A cityscape image semantic segmentation method based on multi-feature fusion and Boosting decision forest includes the following steps of carrying out super-pixel segmentation on images, carrying out multi-feature extraction, carrying out feature fusion and carrying out training learning and classification recognition. The method effectively integrates 2D features and 3D features and remarkably improves recognition rates of targets. Compared with the prior art, segmentation results are consistent, connectivity is good, edge positioning is accurate, a Boosting decision forest classification mechanism is introduced, and stability of target classification is guaranteed.

Disclosed is a multi-camera system target matching method based on a deep-convolution neural network. The multi-camera system target matching method based on the deep-convolution neural network comprises initializing multiple convolution kernels on the basis of a local protective projection method, performing downsampling on images through a maximum value pooling method, and through layer-by-layer feature transformation, extracting histogram features higher in robustness and representativeness; performing classification and identification through a multi-category support vector machine (SVM) classifier; when a target enters one camera field of view from another camera field of view, performing feature extraction on the target and marking a corresponding target tag. The multi-camera system target matching method based on the deep-convolution neural network achieves accurate identification of the target in a multi-camera cooperative monitoring area and can be used for target handoff, tracking and the like.

The invention discloses a multi-mode non-contact emotion analyzing and recording system. The system is characterized by being composed of a voice receiving module, a voice feature extracting and processing module, a speech recognition module, a textural feature extracting and processing module, a comprehensive scheduling module, a displaying module and a clock module; the voice receiving module is used for completing receiving of voice from outside environment, the voice feature extracting and processing module is used for acquiring voice frequency emotion labeling information of speech, the voice recognition module is used for completing conversion from speech content to textural content, the textural feature extracting and processing module is used for acquiring textural emotion labeling information of the speech, the comprehensive scheduling module is used for completing processing, storing and scheduling of all data, the displaying module is used for completing displaying of detected speech emotion state, and the clock module is used for completing time recording and providing a time labeling function. By the multi-mode non-contact emotion analyzing and recording system, a textural mode and a voice frequency mode can be integrated to recognize speech emotions, so that accuracy of recognition is improved.

The invention belongs to the technical field of network flow data analysis and particularly relates to a network flow data exception detection method. The method comprises the following steps of processing original network flow data captured in real time to obtain network flow data; if the network flow data is abnormal data, outputting an exception, inputting the exception data into a pre-trainedfirst exception classifier, judging that the attack type of the exception data is a known attack type, and outputting the attack type of the exception data; if the network flow data is not abnormal data, further detecting whether the network flow data is abnormal or not by adopting an unsupervised anomaly detection method; if the network flow data is abnormal data, inputting the abnormal data intoa pre-trained second abnormal classifier, judging that the type of the abnormal data is an unknown attack type, and marking the abnormal data as the unknown attack type; and if the network flow datais not abnormal data, the output being normal.

The invention discloses an fMRI dynamic brain function sub-network construction and parallel connection SVM weighted recognition method which comprises the steps that (1) data are preprocessed; (2) the time series of each brain area is extracted; (3)interested brain areas are selected; (4) dynamic brain function sub-networks of all the brain areas are constructed; (5) all the sub-network classifiers are trained; (6) values are assigned to all sub-classifiers to form parallel connection SVM classifiers; (7) unknown samples are classified. Compared with a traditional static function network, information on the time dimension is added on the constructed dynamic brain function networks; prior knowledge is combined for constructing dynamic sub-networks on different interested brain areas, and the feature dimensions are reduced while useful information is reserved; SVM classifiers of all the sub-networks are trained, the parallel connection SVM classifiers is formed by determining the weight of the sub-classifiers through the recognition rate, the brain areas are integrally weighed and classified, and the classifiers have better robustness.

The invention discloses a disk fault detection method using multi-model prediction, which extracts multiple characteristics of disk SMART indexes through a sequential data processing means, and establishes a classification model to predict disk state. Step 1, data input: acquiring a data set composed of monitoring data of a plurality of disks in a period of time; step 2, SMART screening: adoptingmutation point detection mode to select SMART index; step 3, feature engineering: using SMART index as the input of the user-defined feature extraction module to extract the features of the SMART index, then extracting the corresponding parameter configuration, and transmitting the parameter configuration to the feature extraction module as a parameter, so as to extract the feature sets of the training set and the test set; step 4, data set balance: desampling the negative sample which occupies a large amount by adopting dimension reduction clustering; step 5, algorithm selection and modeling:on the basis of the step 4, training the classification model and testing whether the current disk belongs to the normal state or the fault state that needs to be replaced.

The invention relates to a layered observation vector decomposed hidden Markov model-based method for identifying behaviors. The method comprises the following aspects of: target detection, target tracking, characteristic extraction, motion modeling, behavior identification and the like. The method improves the conventional hidden Markov model by aiming at deficiencies of a present method for identifying behaviors of multi-user, separates an individual state and an interactive state in the model to stand out interactive relation among targets, and reduces calculated amount by decomposing observation nodes. Meanwhile, the model allows the number of the targets taking the movement to be changed, so that the method has more flexibility in characteristic selection problem than the traditional method and allows simultaneously using discrete characteristics and continuous characteristics. The method plays an important role in analyzing the multi-user interactive behaviors in the fields of video monitoring, and content-based video retrieval and the like.

The invention relates to the technical field of pattern identification and artificial intelligence, in particular relates to an extracting method for smiling face identification on a picture of a human face. The invention provides the extracting method for smiling face identification on the picture of the human face, wherein smiling faces and non-smiling faces can be differentiated by using characteristics of a layered gradient column diagram PHOG (PYRAMID HISTOGRAM OF ORIENTATIONGRADIENTS). The smiling face identification method comprises the following steps: (11) human face detection based on Haar characteristics and AdaBoost algorithm; (12) location of a mouth area of the human face; (13) characteristic extracting algorithm of the layered gradient column diagram; (14) training and identification of a support vector machine (SVM). The method differentiates smiling faces and non-smiling faces by proposing to use PHOG characteristics against over-high Gabor characteristic dimension, and can acquire experimental results equivalent to the identification rate of the Gabor characteristics by lower characteristic dimension. Therefore, the method can greatly improve the efficiency of algorithm, and is easier for usage in reality.

The invention discloses an image segmentation method based on immunity spectrum clustering, which includes: 1. extracting texture characteristic of the input image, representing each pixel point in the image with an eigenvector to obtain a characteristic set; 2. mappings the characteristic set to a linear measure space by spectrum clustering to a mapping set; 3. dividing the category number according to the given image, accidentally selecting the corresponding number of data from the mapping set as the initial clustering center, executing cloning, variation, selection and judgement in sequence, to find out a optimum clustering center with the same category number with the initial clustering center; 4. dividing all pixel points of the characteristic set to an optimum clustering center nearest to the pixel points, and giving each pixel point a category mark according to the category of optimum clustering center where the pixel point locates to complete the image segmentation. Compared with the prior technology, the invention has advantages of insensitivity to initialization, quick convergence to global optimum and high specification accuracy, which can be used in the image segmentation of SAR image processing and computer visual sense field.

The invention discloses a method of face recognition. Firstly, known and classified face samples are trained for modeling via pairs of training modules to produce a support vector machine classifier capable of classifying and recognizing unknown face samples, then, features of the unknown face samples are extracted via recognition modules, and to-be-recognized recognition samples after being extracted out features are transferred to a trained support vector machine classifier to judge the faces which category belong to. The invention provides the method of face recognition based on nuclear stochastic mapping and support vector machines, and the method of nuclear stochastic mapping is initially used for extracting more effective face features in the process of extraction of the face features. In addition, the support vector machine classifier used for dichotomy classification is employed in the problem of face multiple classification, a pair of multi-strategies are employed to convert multi-problems into two-problems, and the invention not only can largely reduce feature dimensions for recognizing faces, but also can remarkably increase the recognition capability for the faces.

The invention relates to a feature extraction method for image scene recognition. The feature extraction method comprises the first step of mining information in a group of training images of which the classes are known and the second step of recognizing test images to be recognized. The first step comprises the sub-steps of preprocessing the images, extracting target features of the images, decreasing the dimensions of the target features, executing the LDA model training algorithm, generating scene environment features of the training images, carrying out feature combination and executing the SVM training algorithm. The second step comprises the sub-steps of preprocessing the testing images, generating code words of the testing images, generating scene environment features of the testing images, extracting target features of the testing images, decreasing the dimensions of the target features of the testing images, carrying out feature combination on the testing images and generating image classes through a trained SVM classifier. By means of the feature extraction method, the calculation amount of an existing method is deceased, the application range is expanded, and the recognition accuracy is improved.

The invention relates to a human body detection method based on Gauss shape feature, belongs to the field of computer vision and mode identification, and solves the problem that the detection rate and detection speed of the prior detection method are low. The method comprises: a first step of extracting feature, namely extracting the Gauss shape feature of each rectangular area of each training sample; a second step of constructing cascade detectors, namely studying the current training sample to construct the cascade detectors; and a third step of detecting human body, namely using the cascade detectors to scan and detect an image to be detected to determine the position and size of the human body in the image. The method has strong robust property for changes of illumination, background and the like by using the constructed Gauss shape feature, and has low feature dimension; the construction of the Gauss shape feature is added with a mean value of areas so as to enhance the capability of identifying the human body and the background; therefore, the constructed cascade detectors can greatly improve the detection rate of the human body, and can be applied to intelligent monitoring, assistant driving and human-computer interaction systems.

The invention discloses a deep learning based medical gas identifying method. The deep learning based medical gas identifying method specifically includes utilizing original frequency response signals to perform simple uniformization, inputting shed typed own coding network, and extracting layer by layer to learn abstract characteristics of original data. The whole network shield the abstract characteristics, dimensionality reduction, drifting restraining and the like to the outside environment, and a classifying layer is added in the network, so that the characteristics can be classified into a classifier to classify. The training process is classified into a pre training step and a micro adjusting step, network learning ability can be effectively increased, and new samples are input into the network to directly acquire predicted category after the training is finished. With the deep learning based medical gas identifying method, effective identifying characteristics of medical gas can be automatically abstracted, steps such as characteristics abstraction, characteristics selection and drifting restraining are combined, complexity in traditional methods is greatly simplified, and gas detecting and identifying efficiency is increased.

The invention provides a synthetic aperture radar target identification method based on diagonal subclass judgment analysis, which mainly solves the problem that the prior synthetic aperture radar has poor target identification performance. The method comprises the following processes: the self-adapting threshold segmentation, the morphological filtering, the geometric clustering operation and the pretreatment of image enhancement are carried out for an original image; the optimal subclass division to each target after pretreatment is carried out by adopting a two-dimension rapid global K-means clustering algorithm; the diagonal subclass judgment analysis or the diagonal subclass judgment analysis and two-dimension subclass judgment analysis are used for finding out an optimal projection matrix; training and testing images after pretreatment are projected towards the projection matrix to obtain characteristic matrixes of the training and testing images; the Euclidean distance between a testing target and the characteristic matrix of each training target is calculated, and the category attribute of the testing target is determined by adopting a nearest neighbor rule. Simulation experiments show that the invention has the advantages of good effect of inhibiting background clutter, high quality of the target image and low characteristic dimensionality and can be used in a remote sensing system.

The invention discloses a KNN-based text classification method which is suitable for nuclear safety software verification and reliability verification. The KNN-based text classification method comprises a training process processing step and a test process processing step: representing training sample data sets by information of original texts and all the titles in the texts; and constructing two DBM models according to characteristic hierarchical structures in the texts from shallow to deep, extracting and storing deep characteristics with low dimensionality and high discrimination, and determining the categories of to-be-tested texts through considering the contribute brought from text titles to the similarity calculation via proper weights in the test process. According to the method disclosed in the invention, information of the text titles are fully utilized, the classification performance, relative to the condition of taking the shallow characteristic vectors as training sets, is remarkably improved, and meanwhile, the storage demand and online calculation amount are reduced at the same time, so that the problem of characteristic vector high-dimensionality disaster is solved and the classification correctness is improved; and the method can be used for the matching of rules in safety level software reliability evaluation analysis and the establishment of a failure mode library.

The invention provides a non-reference tone mapping image quality evaluation method based on multi-feature fusion. The method mainly comprises a feature extraction stage and a training regression stage. In the feature extraction stage, features of an image are extracted in three fields, and entropy features of the image, texture features based on a gray level co-occurrence matrix and natural scenestatistical features are extracted in the pixel field. In the field of ambiguity, local phase consistency is adopted to evaluate the overall ambiguity, and small block features are extracted from anedge region to measure the halo effect. In the color field, an image is converted into an opposite color space, and overall color information and contrast information of all channels are measured. Finally, the quality of the tone mapping image is predicted by using a machine learning method. The algorithm provided by the invention can accurately and effectively predict the quality of the tone mapping image.

The invention discloses a hyperspectral image classification method based on compression spectrum clustering integration. The hyperspectral image classification method comprises a classification process including the steps that (1) image characteristic sets of hyperspectral images are obtained; (2) an image characteristic set sub space subjected to dimension reduction is obtained; (3) a plurality of hyperspectral image dividing results are obtained; (4) the final hyperspectral image dividing result is obtained; (5) the hyperspectral image classification result is obtained; (6) the accurate classification of the hyperspectral image is obtained. Compared with the prior art, the hyperspectral image classification method has the advantages that the defect that a k-means algorithm adopted in the classical spectral clustering is sensitive to initialization is overcome; the characteristic dimension of the hyperspectral images is reduced; meanwhile, the classification precision is obviously improved, and the dividing effect is good.

The invention relates to a human motion emotion identification method based on Gauss feature, belongs to the field of computer mode identification, and solves the problems of low identification rate and low learning identification speed in the prior emotion identification method. The method comprises a training classifier step and an emotion identification step, wherein the training classifier step comprises a training data acquisition substep, a motion segmentation substep, a feature extraction substep and a training substep; and the emotion identification step comprises a data to be detected acquisition substep, a motion segmentation substep, a feature extraction substep and an identification substep. The method adopts the Gauss feature to describe human motion, has the advantages of strong descriptive power, low feature dimension, good Lie group structure, capability of effectively analyzing spatial structure, and the like, adopts a LogitBoost machine learning method based on a Lie group space to carry out multi-emotion identification, makes full use of the Lie group structure of the Gauss feature in a machine learning process, and has high training and identification efficiency and strong practicability.

The invention discloses a multi-view angle gait recognition method based on Kinect. The multi-view angle gait recognition method based on Kinect comprises the steps that three-dimensional space position information of a framework joint point is collected and normalized to side view angles through view angle normalization; gait features existing after view angle normalization are extracted, and neural network modeling and recognition are conducted on gait system dynamics at different view angles in a training set; a constant value neural network is established, a dynamic estimator is established, and through the differences between gait modes at different view angles on the gait system dynamics, a test mode is recognized according to the minimum error rule. According to the multi-view angle gait recognition method based on Kinect, the three-dimensional space position information of the framework joint point is obtained, the help of other sensing devices is not needed, image processing is not needed, the complexity of a system is lowered, and the extraction precision of feature data is improved.

Nano-scale devices and methods provide reduced feature dimensions of features on the devices. A surface of a device substrate having a pattern of spaced apart first nanowires is consumed, such that a dimension of the first nanowires is reduced. A second nanowire is formed in a trench or gap between adjacent ones of the first nanowires, such that the nano-scale device includes a set of features that includes the first nanowires with the reduced dimension and the second nanowire spaced from the adjacent first nanowires by sub-trenches.