108 results about "Knn classifier" patented technology

The invention provides a probabilistic latent semantic model object image recognition method with the fusion of a significant characteristic of a color, belonging to the field of image recognition technology. The method is characterized by: using an SIFT algorithm to extract a local significant characteristic of an image, adding a color characteristic simultaneously, generating a HSV_SIFT characteristic, introducing TF_IDF weight information to carry out characteristic reconstruction such that the local significant characteristic has discrimination more, using a latent semantic characteristicmodel to obtain an image latent semantic characteristic, and finally using a nearest neighbor KNN classifier to carry out classification. According to the method, not only is color information of theimage considered, but also the distribution of a visual word in a whole image set is fully considered, thus the local significant characteristic of an object has discrimination more, and the ability of recognition is raised.

A fake fingerprint detection method based on markov random field (MRF) and support vector machine-k nearest neighbor (SVM-KNN) classification includes steps: (1) fingerprint image feature extracting: (1.1) first-order statistics (FOS), (1.2) a gray level co occurrence matrix (GLCM) and (1.3) an MRF; (2) SVM training: training the FOS and the GLCM feature vector and the MRF feature vector to obtain a model A and a model B; (3) SVM-KNN classification: (3.1) the SVM classification mechanism and (3.2) SVM-KNN classifier forming; and (4) decision fusion for true and false fingerprint detection. Presently related articles for fake fingerprint detection by aid of the GLCM and the MRF are not found, and the fake fingerprint detection method achieves the purpose of identifying true and false fingerprints by aid of physical structures of the two feature quantized fingerprint images. Experiment results prove that the false accept rate and the false reject rate of the algorithm are respectively 1.84% and 1.79%, and therefore the fake fingerprint detection method is high in accuracy and good in practicality.

The invention discloses an open domain Chinese text naming entity identification method based on semi-supervised learning. The method comprises two steps of model training and prediction by a model. In the model training stage, a training set text is subjected to word segmentation preprocessing; then, in virtue of word vector space constructed by a word2vec tool, obtaining a word vector expressedby a word distribution type form in the training text; and utilizing the word vector in the training set and the existing entity type tag of each word vector to train a KNN (K-Nearest Neighbor) classifier and a CRF (Conditional Random Field) annotator, and generating a prediction model of a KNN-CRF naming entity type. In the model prediction stage, an empty reliable result set is imported, and when a new prediction result is generated by prediction, the prediction result is added into the reliable result set; when an amount in the reliable result set achieves a threshold value, previous KNN and CRF models are abandoned, the results in the reliable result set are added into the training set, and the KNN classifier and a CRF annotation model are trained again; and the above steps are repeated until a condition is met.

The invention relates to a method for detecting a landslip from a remotely sensed image by adopting an image classification technology. The method comprises the following steps: firstly, partitioning the remotely sensed image after being pretreated according to an overlapping-based area averaging method, thereby obtaining square image blocks with the same areas; dividing the obtained image blocksinto two sets: a training set and a test set; extracting SIFT (Scale Invariant Feature Transform) features of all the image blocks in the training set and the test set; processing the SIFT features in the training set according to a k-means classifying method, thereby obtaining words and a dictionary; representing each image block in the training set and the test set by a BoVW model; and lastly, utilizing a pLSA model to extract the theme of each image block, and utilizing a KNN classifier to divide the image blocks in the test set into landslip type and non-landslip type image blocks, thereby realizing the landslip detection for the remotely sensed image. The method provided by the invention has the advantages of small calculation volume, high detecting efficiency and ultrahigh detectingaccuracy.

The invention relates to a special human body recognition method based on a head and shoulder model, belonging to the fields of pattern recognition and artificial intelligence as well as computer vision. In the invention, a camera is adopted as a video collector, and an industrial personal computer is used as a video processing device. The method comprises the steps of firstly detecting a movement object, counting the human body number in the obtained binary images of the movement object and extracting the head and shoulder models of the human bodies, and numbering in sequence if a plurality of human bodies exist; sequentially extracting the invariant moments of the head and shoulder models of the human bodies according to numbers, classifying the head and shoulder models into front backsor side surfaces according to classification standards; and realizing the recognition on specific human bodies and outputting a recognition result by respectively using the trained front back KNN (K-nearest neighbor) classifier or side surface KNN classifier for classification according to the classification of the head and shoulder models of the human bodies. The use of the special recognition method can well improve the recognition accuracy.

The invention relates to the technical field of electroencephalogram signal processing, in particular to an emotional electroencephalogram signal recognition method based on EMD domain multi-dimensional information. The method includes the steps that electroencephalogram signals are adaptively decomposed into intrinsic mode functions (IMF) with different oscillation frequencies through EMD; then, the waveform difference, phase difference and normalized energy of the IMFs are extracted; feature vectors are formed by the extracted multi-dimensional information to serve as the representation of different emotional electroencephalogram signals; the emotional electroencephalogram signals are classified and recognized through a KNN classifier and an SVM classier. Thus, the classification accuracy is greatly improved.

The invention discloses a specification-based patent classification method and belongs to the field of text processing and data mining. The method comprises the steps of firstly performing text preprocessing on patent specifications; secondly establishing reverse index files and selecting feature words by utilizing a feature selection method combining information gain with a word frequency; thirdly calculating weights of the feature words by utilizing an improved TF-IDF formula, and creating patent eigenvectors; fourthly establishing a training patent field set; and finally classifying patents by utilizing an optimized KNN classifier. The research provides a new idea for patent literature classification and lays a foundation for further researching patent literature intelligent retrieval and the like.

The invention belongs to the field of machine learning and privacy protection, and particularly relates to a KNN classification service system and method supporting privacy protection. The architecture of the system comprises a model owner and a client; the method of the KNN classification service system supporting privacy protection comprises the following steps: 1) a preparation stage: generating a public key and a private key, and encrypting training data according to the public key; 2) a classification stage: two parties interact with keys; and the client encrypts to-be-tested data throughthe public key, the model owner completes encrypted data classification by cooperating with the client through a security protocol based on the encrypted training data, and finally obtains a classification result and sends the classification result to the client. According to the method, training data and to-be-tested data are encrypted by using homomorphic encryption calculation, a secure basicprotocol is constructed by combining a secure multi-party calculation technology and homomorphic encryption, and a secure KNN classifier is constructed based on the secure basic protocol, so that thetwo parties realize analysis and prediction of personal data on the premise of ensuring that the privacy of the personal data is not leaked.

The invention provides a method for performing particle source analysis by using a single-particle aerosol mass spectrometer and relates to the field of atmospheric particle source analysis. Accordingto the method, a KNN (K Nearest Neighbor) classifier and a CMB (Cosmic Microwave Background) source analysis model are combined, single-particle mass spectrometry information of the pollution sourceas well as environmental receptor single-particle mass spectrometry characteristic and particle size information are fully utilized, the contribution of the primary emission source and secondary emission source of the particles on single environmental receptor particle can be rapidly solved, and refined source analysis results with high time resolution and high particle size resolution can be obtained. The method has excellent popularization and application prospects.

Disclosed is a model identification method based on machine learning. The method comprises the following steps: 1), carrying out vehicle detection by use of a background difference method, performing contour tracking on a motion object, obtaining external contour characteristics of the motion object, and performing vehicle predetermination and image preprocessing on an object image; 2), extracting vehicle characteristics: (2.1) extracting vehicle geometrical characteristics; and (2.2), extracting characteristics of seven invariant moments of a vehicle; 3), model classification training, i.e., training an input fifteen-dimensional model characteristic sample by use of a metric learning based KNN classifier, and obtaining four model classifications; and 4), model classification based on local linearity reconstruction error minimization, i.e., performing local linearity reconstruction error calculation and classification on a newly introduced test sample by use of a reconstruction error minimization method. The model identification method based on the machine learning, provided by the invention, is quite high in accuracy and good in real-time performance in case of quite a large number of types.

The invention discloses a bag-of-visual-word model-based monitor video vehicle type classification method. The method comprises the steps of collecting vehicle images; extracting vehicle eigenvectors from the vehicle images in a training image library; clustering the obtained vehicle eigenvectors to generate a visual dictionary; performing spatial pyramid decomposition on the extracted vehicle eigenvectors and the generated visual dictionary to obtain final image eigenvectors, and forming a training image model library; and extracting the final image eigenvectors from the to-be-classified images, performing classification by utilizing a KNN classifier, and outputting the types of the to-be-classified images. According to the method, the dimensions of the eigenvectors can be freely reduced by using a PCA-SIFT algorithm, so that the calculation amount is greatly reduced; and in addition, for an improved k-means algorithm, outliers and isolated points are removed by using a local factor anomaly algorithm, and then clustering is performed, so that the clustering precision of the k-means algorithm can be remarkably improved, the initial center selection is more targeted, and the calculation amount and the iterative frequency are reduced.

The present invention provides one kind of tongue coating and tongue proper color extracting and classifying method based on pixel, and the method has high accuracy and wide application value. The method includes the first selecting typical tongue picture samples from tongue picture database, performing color clustering on each of the tongue picture samples in FCM algorithm, calibrating the color centers of the results and superposing the calibrated clustering color values of all the color sorts; the subsequent purifying treatment of each color sample set and screening samples to obtain color sample set; and final classifying and analyzing the tongue coating color and tongue proper color on each of the tongue picture with the improved KNN classifier. The present invention realizes the quantitative, objective and standardized analysis on tongue coating and tongue proper color and automatic computerized classification of 14 kinds of tongue coating and tongue proper colors.

The invention discloses a bearing fault diagnosis method based on KNN-AdaBoost, which comprises the following steps of: preprocessing vibration signals into modal components for determining a center frequency and a bandwidth according to a variational modal decomposition algorithm, respectively selecting the modal components with maximum kurtosis in each group of vibration signals, calculating toobtain a plurality of groups of bearing fault characteristic vectors; training to obtain a plurality of KNN classifiers; training according to the plurality of groups of bearing fault characteristic vectors and the plurality of KNN classifiers to obtain an AdaBoost strong classifier; and classifying the collected bearing vibration signals according to steps by using the AdaBoost strong classifier.According to the invention, based on KNN-AdaBoost, fault classification can be carried out according to a large amount of real-time bearing monitoring data, and the fault types of the bearing can bequickly identified.

The invention relates to a three-level inverter open-circuit fault diagnosis method based on an optimized support vector machine. The method comprises following steps: (1) extracting a fault characteristic, and allocating samples; (2) performing nuclear Fisher dimension reduction, and optimizing a nuclear function parameter; (3) determining a range in which a support vector is; (4) extracting a KNN reference point; (5) allocating test samples; and (6) classifying test samples. In the method, characteristics are extracted through intrinsic mode decomposition and singular value decomposition, sothat time-varying nonlinear signal characteristics can be better extracted; a support vector is extracted by using nuclear Fisher algorithm and is used as a training sample, effectively improving training speed of the vector machine; test samples nearby a classification super plane are classified by using a KNN algorithm, thereby being classified more accurately; and the SVM-KNN classifier builtbased on the nuclear Fisher algorithm can correctly classify three-level inverter open-circuit faults more accurately within shorter test time.

The invention relates to a steganalysis hybrid integration method based on deep learning. The method comprises the following specific operation steps: dividing a data set, obtaining a residual image through a high-pass filter, constructing a convolutional neural network, training a network model, storing a plurality of optimal models, respectively loading the models, and storing the output of a pooling layer as a feature; Changing a high-pass filter, generating different residual images to obtain difference features, and performing feature fusion to obtain high-dimensional features; Inputtingthe high-dimensional features into the PCA for dimension reduction; Inputting the features subjected to dimension reduction into an xgboost classifier, an SVM classifier and a KNN classifier for classification; And carrying out integrated learning on the obtained classification result, and carrying out weighted voting to obtain a final classification result. According to the invention, the classification accuracy of the classifier can be effectively improved.

The invention relates to a network security situation assessment method, and aims to provide a network security situation assessment method based on a fuzzy rough set. The network security situation assessment method based on the fuzzy rough set comprises the following steps: acquiring situation factors and situation levels to construct a decision table of the situation levels relative to the situation factors; and performing attribute reduction on the situation factors of decision rules in the decision table to obtain a reduced decision table, and making a decision on data needing to be judged in a current network through a KNN classifier by using the reduced decision table in order to obtain a situation level. Through adoption of the method, the problem that discretization of a real number continuous attribute is needed in a rough set method is solved. On the other hand, a rule aggregation way is adopted, thereby lowering the calculation complexity of a fuzzy rough set method. Meanwhile, a good decision result can be provided.

The invention relates to a remote sensing hyperspectral image classification method based on a local binary pattern and a KNN classifier. Compared with the prior art, the defect that remote sensing hyperspectral image classification is difficult to carry out is overcome. The method comprises the following steps: acquiring training data; extracting a spectral feature vector; extracting a spatial feature vector; extracting a color feature vector; stacking a plurality of feature vectors; constructing and training a KNN classifier; acquiring a remote sensing hyperspectral image to be classified; preprocessing the remote sensing hyperspectral images to be classified; and obtaining a remote sensing hyperspectral image classification result. According to the method, spectrum, space and color features are combined, the similarity of similar pixels is enhanced; meanwhile, the dissimilarity of different pixels is increased, and compared with a partially proposed classification method, the overall classification precision, the average classification precision and the Kappa coefficient are improved to different extents.

The invention discloses an improved Online Boosting and Kalman filter improvement-based TLD tracking method, and belongs to the technical field of machine vision, artificial intelligence, man-machineinteraction and target tracking. The method comprises the following steps of: (1) initialization: initializing an improved Online Boosting classifier and a P-N learning device by utilizing an initialsample set formed through selecting a target and carrying out affine transformation; (2) image tracking: selecting a feature point, tracking the feature point for twice by using an L-K optical flow method, and comparing an error between the twice tracking with a threshold value so as to obtain a tracking result; (3) image detection: obtaining a detector result through a Kalman filter, a variance classifier, the Online Boosting classifier and a KNN classifier; (4) tracking result and detection result integration: assessing confidence coefficients of the tracker result and the detector result soas to determine which module result is finally adopted; and (5) online learning: correcting the tracker result and the detector result by using the P-N learning device, and enriching the sample set.The method is capable of effectively overcoming the shielding problems, improving the speed of original methods and effectively the precision and robustness of detectors.

The invention discloses a feature database-based industrial wastewater pollutant tracing analysis method and belongs to the technical field of industrial wastewater pollutant supervision. The method is characterized by comprising the steps of establishing a feature weight database M of parameters of all types of pollutants of each factory; acquiring pollutant data of wastewater discharged by each factory in a target area; establishing a feature data sample library N of pollution discharge of the factory in the target area; establishing a class variable of a KD-tree by utilizing the feature data sample library N; taking the input pollutant measurement values of mixed industrial wastewater as a to-be-identified vector z; and performing matching identification by adopting a kNN classifier and data in the feature data sample library N, thereby finishing pollutant tracing. According to the method, the pollutant tracing of the mixed industrial wastewater in various areas can be finished, an order of target discharge factories is quickly and effectively given, applicability and universality are good, technical support is provided for related functional departments of government to check pollution source factories in order, the check efficiency is greatly improved, and the check success rate is greatly increased.

The invention, which belongs to the mobile robot environment sensing field, especially relates to a bag-of-visual-word-model-based indoor scene cognitive method. The method comprises an off-line part and an on-line part. At the off-line part, scene types are determined based on an application need; a robot uses a carried RGB-D sensor to scan all scenes to obtaining enough scene images to form an image training set; and an ORB 256-dimensional descriptor of each image in the image training set is generated by using an ORB algorithm, wherein each image includes thousands of ORB vectors usually. At the on-line part, the robot receives a current scene type inquiring instruction; and the system is initialized and is prepared for scene query. With the ORB algorithm, the image pretreatment process including feature extraction and matching is completed, so that the algorithm rapidness can be guaranteed; and the scene identification rate is improved by using a KNN classifier algorithm, so that the demand of common indoor scene inquire application of the mobile robot can be satisfied.

The invention discloses an image classification algorithm and system combining the superpixel saliency features and the HOG features, comprising the following steps of 1 extracting the HOG features from the original data set to obtain a feature set A; 2 processing the original data set by using the superpixel segmentation to obtain the reconstruct salient image data set; 3 obtaining a new featureset B by using the reconstructed salient image set in an SLBP coding mode, wherein the SLBP coding mode is an improvement of the traditional LBP coding mode; 4 carrying out the feature fusion, addingthe feature set A and the feature set B obtained in steps 1 and 3; 5 combining the training data obtained in step 4, using the KNN classifier to supervise and classify the test data and calculating the classification accuracy. The method of the invention well meets the requirements of the small-scale engineering design on the traditional machine learning algorithm, and to a certain extent improvesthe image classification accuracy.

The invention relates to an elastic optical network modulation format identification method based on KNN. The method comprises the specific steps of (1), constructing training data, sending the training data from a sending end, and through transmission of an OFDM (Orthogonal Frequency Division Multiplexing) transmission system, receiving signals at a receiving end; (2), carrying out FFT on received signals and carrying out channel estimation and channel balancing, thereby obtaining a series of complex data; (3), extracting real parts of the complex data, thereby obtaining a real part histogram, and preprocessing the histogram, thereby obtaining feature vectors; (4), modeling features through utilization of a KNN algorithm and obtaining a classifier; (5), for newly received data, obtainingdata features of each subcarrier, and carrying out prediction classification on the feature vectors through utilization of a KNN classifier, thereby obtaining a modulation format of the subcarrier; and (6), demapping respective carrier data through utilization of the obtained modulation format, and obtaining original data.

The invention discloses a medical image lesion area positioning and classification method. The method comprises the following steps: the step 1, acquiring medical images, and dividing the images into a training set and a test set; the step 2, extracting WLD histogram information from the images of the training set, marking characteristic points of the training set according to the histogram information, placing the marked characteristic points into a KNN classifier, training the classifier, testing the trained KNN classifier using the images of the test set, and completing positioning of a medical image lesion area; the step 3, segmenting the images after the positioning through adoption of a histogram threshold method, and reserving the lesion area; and the step 4, putting the segmented images into a CNN depth model to extract characteristics, performing lesion area characteristic classification using an SVM classifier, and outputting a classification result. When the lesion area occupies a small part of the whole image, the accuracy rate is substantially improved.

The invention relates to a pulse signal subhealth detection method and a detection system based on PCANet. The method comprises the following steps: selecting research objects and initially dividing and screening health conditions of the research objects during detection, and then collecting pulse signal data of the research objects through a pulse sensor; normalizing the collected pulse signal data, wherein the pulse signal data is mapped to a range of [-1, 1], and removing noises and abnormal waveforms in the pulse signal to obtain pure pulse signals; separately extracting health and subhealth pulse signal characteristics on the pre-processed pulse signals by PCANet to reconstruct errors as a loss function, learning a PCA filter nucleus, calculating a sample set by means of a PCA algorithm, and extracting characteristic vectors corresponding to previous L maximum characteristic values of a covariance matrix of the sample seat to form a characteristic mapping matrix; and recognizing health and subhealth states on the extracted characteristic mapping matrix through an SVM classifier and a KNN classifier. The detection process is simple and efficient.

The invention discloses a text classification method based on a CNN-SVM-KNN combined model, and relates to a text classification method based on a combined model. The objective of the invention is tosolve the problem of low text classification accuracy of an existing method. The method specifically comprises the steps of 1, text preprocessing; 2, performing feature extraction on the text preprocessed in the step 1 to obtain a text subjected to feature extraction; 3, establishing a CNN model based on the step 2; 4, establishing a CNN-SVM model; 5, establishing a CNN-KNN model; 6, setting a distinguishing threshold d; 7, calculating the distance: calculating the optimal classification surface distance tmp from the to-be-classified sample points to the CNN-SVM classifier; 8, comparing distances: when tmp is greater than d, selecting a CNN-SVM classifier; otherwise, selecting a CNN-KNN classifier; and 9, repeatedly executing the steps 6 to 9, and searching for the optimal d value of the evaluation index. The method is applied to the field of text classification.

The invention discloses an integrated classification method for mass multi-word short texts, which comprises the following steps of: 1, acquiring a multi-word short text set, and performing word segmentation preprocessing on the multi-word short text; 2, obtaining a word vector representation model on the word segmentation result by utilizing a CBOW continuous word bag model in a Word2vec word vector representation method; 3, based on a word vector representation model, using a Sentence2vec sentence vector representation method to represent PV-in the method; The DM model is used for constructing sentence vector representation; And 4, on the basis of the sentence vector representation model, utilizing a kNN classifier to predict a category label of the labeled data. According to the method,the problem of'dimensionality disaster 'of a traditional representation learning method can be solved, so that the short text representation learning effect is improved, the text classification precision is improved, and the method has relatively high robustness and practicability.

The invention discloses an automatic footprint recognition method based on multi-feature jointed decision making. The method comprises the following steps that: S1, extracting a footprint pressure distribution feature in an off state; S2, extracting a directional gradient histogram feature of the footprint; S3, extracting a wavelet Fourier Merlin feature of the footprint; S4, carrying out feature selection on the footprint pressure distribution feature P, the directional gradient histogram feature H, and the wavelet Fourier Merlin feature F by using a two-dimensional principal component analysis and a two-dimensional linear discriminant analysis technique so as to obtain P<G>S, H<G>S, and F<G>S respectively; and S5, identifying a feature of a to-be-identified footprint and footprint data stored in a feature database in advance by using a KNN classifier.

The invention discloses a rolling bearing fault classification method based on mixed feature extraction. The rolling bearing fault classification method comprises the steps: firstly, acquiring a mixedfeature set composed of waveform features, time domain features, frequency domain features and the like of signals; introducing the internal class compactness and the internal overlap into a sequenceforward selection algorithm, and extracting a suboptimal feature group in the mixed features as the input of an enhanced KNN classifier; and finally, based on distance and density calculation, obtaining an optimal average classification probability, outputting an optimal feature group, marking a fault state corresponding to the feature group, and realizing intelligent classification of rolling bearing faults. According to the invention, the interference of correlation and redundancy between fault signals on the fault classification accuracy is effectively reduced; according to the KNN classifier, the capability that a traditional KNN classifier only adopts distance calculation for classification is improved, and the problem that the traditional KNN classifier is influenced by K value sensitivity and is not beneficial to classification of an intelligent algorithm is solved, and finally the classification accuracy is improved.

The invention discloses a feature engineering-based KNN equalization algorithm for short-distance optical communication, and the algorithm comprises the following steps: 1, inputting data which is sampled by a receiving end and contains a training sequence into a feature engineering module, constructing a feature vector, and carrying out the feature processing; Step 2, enabling the training sequence generator regenerate a training sequence as a label corresponding to the feature vector of the training sequence to form a training set; 3, using the same feature engineering to obtain feature vectors of the valid data, and taking the feature vectors of the training set and the valid data as input of a KNN classifier; and step 4, according to the categories of k training data closest to each valid data in the feature space, the KNN classification result is balanced output. According to the low-cost short-distance optical communication system, the problem that the performance of the system is reduced due to inter-code crosstalk, signal distortion and the like in the low-cost short-distance optical communication system is solved, and compared with traditional DFE and FFE equalizers, the equalization performance is improved, and the tap coefficient of the filter is reduced.

The invention provides a hyperspectral image classification method and a system based on the correlation entropy principle. Firstly, the input image is preprocessed so that the element values of all spectral vectors are between 0 and 1; secondly, a small number of samples are selected as training samples, and the spatial features of hyperspectral images are extracted by using the designed hierarchical model of fusion dimension reduction method and correlation entropy principle. Then, the spatial spectrum characteristics of the test samples are learned by using the trained hierarchical structure. Finally, the characteristics of the test samples are input into the KNN classifier to get the class label. The invention fully combines the spectral and spatial characteristics of the hyperspectraldata by using the correlation entropy principle; by using the hierarchical model of the invention, more abstract spatial spectrum features can be obtained; due to the characteristics of LDA, the invention has small sample complexity, and only a small number of training samples are needed to obtain a better classification result, so the invention is more conducive to practical application.