206results about How to "Guaranteed classification accuracy" patented technology

The invention discloses an infrared human body object recognition method based on HOG-PCA and transfer learning. The method comprises the steps: firstly constructing an infrared and visual light human body or non-human object training sample library; secondly extracting the HOG features of each sample, carrying out the dimension reduction of the features through employing PCA, and obtaining an HOG-PCA target feature set after dimension reduction; thirdly carrying out the training based on the feature set through the transfer learning theory, and obtaining an infrared human body object recognition classifier; and finally completing the recognition of a to-be-detected infrared image. Compared with a conventional human body object recognition algorithm, the method provided by the invention achieves the transferring from a visible light image region to an infrared image field, effectively irons out the defect that a conventional object recognition classifier cannot achieve the training of an effective classifier because of the extremely small number of training samples in the infrared image field, and improves the object recognition rate.

The present invention discloses a deep neural network compression method based on block item tensor decomposition. The method comprises the steps of: obtaining a framework of a deep neural network, respectively converting a weight matrix W and an input vector x to a high-order tensor (u)W(/u) and a high-order tensor (u)X(/u), performing block item tensor decomposition processing of the high-ordertensor (u)W(/u), replacing a full-connection layer of the deep neural network with a block item tensor layer, and employing a back-propagating algorithm to perform training of the deep neural networkreplaced in the step D. The deep neural network compression method based on the block item tensor decomposition employs a block item tensor decomposition method to construct the block item tensor layer to replace the full- connection layer of the original deep neural network, the features of symmetry and index expression capability of the block item tensor layer are employed to greatly compress the parameter amount of the full-connection layer and maintain the classification precision of the original network.

The invention discloses an intelligent steel cord conveyer belt defect identification method and an intelligent steel cord conveyer belt defect identification system. The identification method includes the following steps: (1) electromagnetic loading; (2) defect signal acquisition; (3) feature extraction; (4) training sample obtainment; (5) class priority determination; (6) multi-class model establishment; (7) multi-class model training; (8) real-time signal acquisition and synchronous class: electromagnetic detection units are adopted for real-time detection, detected signals are synchronously inputted into a data processor, features are extracted and then sent into established multi-class models, and the defect class of a detected conveyer belt is automatically outputted. The identification system comprises an electromagnetic loader, a plurality of electromagnetic detection units, the data processor and an upper computer, the data processor can automatically output the defect class of the detected conveyer belt, and the upper computer bidirectionally communicates with the data processor. The design of the invention is reasonable, the invention is easy to operate and convenient to put into practice, moreover, the using effect is good, the practical value is high, the reliability of conveyer belt defect detection is enhanced, and the efficiency of defect identification is increased.

The invention provides a sorting device provided with impurity-clearing mechanisms used for potato cleaning and sorting machine. The sorting device is characterized in that serial numbers of five rhombuses of an expansion bracket are I, II, III, IV and V; a motor frame is mounted on a hinged part of the end part of a long connecting piece with the III rhombus; a gear motor mounted on the motor frame drives each sorting roller on the expansion bracket to rotate through a transmission mechanism; the two ends of each sorting roller which passes through the expansion bracket are connected to a single row of chain wheels; one end of the sorting roller which passes through the rhombus II and the rhombus III is connected to the gear motor; a protective sleeve is arranged on the external circle of each sorting roller in a sleeved manner; the end part, towards the gear motor, of each sorting roller is provided with a impurity-clearing frame I of each clearing mechanism, and the other end of the corresponding sorting roller is provided with a impurity-clearing frame II; a bevel gear shaft is driven by a bevel gear mounted on the sorting roller I to operate; and a plurality of impurity-clearing cones are driven by a synchronous belt which is hung on four synchronous belt wheels of each group of impurity-clearing mechanisms to operate along a rail. The sorting device can be used for conveniently grading the cleaned potatoes and timely clearing mud and impurities, is low in potato damage rate, and is uniform and thorough in sorting.

The invention discloses a multi-strategy combined document automatic classification method which comprises the steps: extracting keywords from a given to-be-classified document, obtaining a vector of the to-be-classified document according to the keywords, and obtaining a class center vector of each class in a standard document base; obtaining a class threshold value of a classified document according to the class centre vector; carrying out comparison of the similarity degree of the vector of the to-be-classified document and the class centre vector with the class threshold value, when the similarity degree of the vector of the to-be-classified document and the class centre vector is larger than the class threshold value, classifying the to-be-classified document through adoption of a space vector model; and if the similarity degree of the vector of the to-be-classified document and the class centre vector is not larger than the class threshold value, classifying the to-be-classified document through adoption of a k-nearest neighbor (KNN) method. Compared with the prior art, through the technical scheme, the multi-strategy combined document automatic classification method can greatly improve efficiency of the document classification and cannot reduce classification precision at the same time.

The invention discloses an SDN (Software Defined Networking) controller for flow classification based on DPI (Deep Packet Inspection) and a machine learning algorithm. In an existing controller module, a DPI mechanism and a machine learning algorithm are combined by adding a self-designed module to perform an application request classification function of a flow data packet. The self-designed module mainly consists of three parts: a data flow feature building module, a feature extraction and classifier training module, and a classifier-based flow classification module. Firstly, a K-means clustering algorithm is used to construct an information feature base; secondly, a DPI technique is used to extract features to form the training module; and finally, the Naive Bayes algorithm is used to perform detailed flow classification on data packets. The SDN controller for flow classification based on the DPI and the machine learning algorithm, which is disclosed by the invention, is relativelyhigh in classification precision and relatively good in flexibility.

The invention discloses a method for estimating and classifying motion states of a front object based on a vehicle-mounted millimeter wave radar. The method is characterized by comprising the following steps of on the basis of the lateral speed information of the limited motion of the front object which is directly measured by the vehicle-mounted millimeter wave radar, establishing a motion equation of the front object under a ground coordinate system, utilizing a self-adaptive Kalman filter estimation algorithm, and accurately estimating the motion state of the front object in real time; according to the motion information of the front object, classifying according to the specified motion state division speed threshold value and the motion state conversion rule. By adopting the technical scheme, the method has the advantages that compared with the prior art, the estimation accuracy of the motion state is greatly improved, the strict front object motion state conversion condition is specified, and the classification accuracy is guaranteed.

The invention provides a mass image sorting system based on deep character learning. The mass sorting system comprises the following steps that firstly, non-label image data and label image data are input, the non-label image data are pre-processed, interference information is removed, and key information is remained; secondly, K-means character learning is carried out on pre-processed images, and a dictionary of the layer is obtained; thirdly, if the layer is the Nth layer, character mapping is carried out on the dictionary of the layer and the label image data, the fifth step is executed after the deep characters are obtained, or otherwise, character mapping is carried out on the dictionary of the layer and the non-label image data, and the deep characters are obtained; fourthly, the high correlation characters are combined into a receptive field according to the correlation of the deep characters, if the layer is the (N-1)th layer, the fifth step is executed, or otherwise the layer serves as the input information of the next layer and is sent to the second step; fifthly, in the Nth layer, the learned characters are input to an SVM sorter, and sorting is carried out.

The present invention discloses a brain electrical emotion identification method for extracting features based on the empirical wavelet transformation. The brain electrical emotion identification method comprises the steps of according to the coherence between the brain electrical emotion data peripheral physiological signals and the brain electrical signal channels, selecting the brain electricalchannels having the highest coherence degrees as the feature extraction channels; extracting the brain electrical emotion features based on the empirical wavelet transformation, and obtaining a series of intrinsic mode functions after the empirical wavelet transformation decomposition; calculating the sample entropy of each intrinsic mode function, and constructing the feature vectors based on the sample entropy values to form a feature vector set; dividing the brain electrical emotion data into a plurality of types according to the two dimensions of arousal and titer, and using a support vector machine to identify the brain electrical emotion. The method of the present invention considers the non-linear features and the non-stationary features of the data in a brain electrical emotion data set simultaneously, thereby guaranteeing the classification precision, accuracy and execution speed of the method.

The invention provides a deep multi-feature active migration network-based hyperspectral image classification method. The method is used for solving the technical problem that the classification precision is low in the prior art. The method comprises the following steps of acquiring to-be-classified source domain and target domain hyperspectral images; preprocessing to-be-classified source domainhyperspectral images; acquiring a source domain marking sample set, a training sample set, a candidate sample set and a test sample set; constructing a source domain depth spectral feature extractionnetwork, and extracting the depth spectral features of the source domain marking sample set; constructing a source domain depth spatial feature extraction network, and extracting the depth space features of the source domain marking sample set; constructing a source domain depth spectrum-spatial joint feature extraction and classification network; and classifying to-be-classified target domain hyperspectral images by adopting a source domain depth multi-feature active migration network. According to the hyperspectral image high-precision classification method provided in the invention, the high-precision classification for different areas of one hyperspectral image or two hyperspectral images shot at the same place and at different time moments can be realized by utilizing a small number of training samples.

The invention belongs to the technical field of data stream mining, and particularly relates to a method for handling missing values during data stream decision tree classification. The method includes reading data samples in data streams and updating sliding windows; updating missing handlers if the missing handlers corresponding to attributes are available when the detected attributes in the current data samples have the missing values, or adaptively selecting and creating missing handlers according to characteristics of data if the missing handlers corresponding to the attributes are not available; supplementing the missing values in the data samples by the aid of the missing handlers to obtain complete data samples, training the complete data samples according to a Hoeffding decision tree classification process and returning data stream decision tree classification results. Compared with existing methods, the method has the advantages that the method is superior in time performance, the classification accuracy of decision tree models can be sufficiently guaranteed, accordingly, the time expenditure can be reduced, the time performance can be improved, the data stream classification handling speeds can be increased, and requirements of actual data stream handling application can be met.

The invention relates to the technical field of network traffic classification, and relates to an unbalanced-like network traffic classification method and device and computer equipment. The method comprises the steps of obtaining to-be-classified network traffic data, and extracting features of network traffic; deleting irrelevant features and redundant features by adopting a feature selection algorithm, and performing dimension reduction on the remaining features so as to select an optimal feature subset; and inputting the optimal feature subset into a weight-based multi-classifier, performing network traffic classification training in an incremental learning mode, optimizing classifier performance, and classifying the network traffic. Aiming at the problem of unbalanced distribution ofnetwork traffic samples, irrelevant features and redundant features are deleted, and the recognition rate of small categories is effectively improved on the premise of ensuring the overall classification accuracy; an incremental learning thought is introduced, the flexibility of model updating training is improved, and the model updating period is shortened; and by utilizing multiple classifiers based on weight, the influence caused by concept drift is reduced.

The invention discloses an active learning and deep learning combined aluminum material surface defect detection method. The method comprises the steps of performing data enhancement on data in a training set by randomly adjusting image saturation, adjusting image brightness, adjusting image contrast and randomly rotating an image; adopting a Weighted-Entry evaluation standard for active learning,ranking the samples to be labeled according to a Weighted-Entry value in an increasing sequence, selecting K highest samples for labeling, and adding the samples into a training set to serve as training samples; and meanwhile, sequencing the samples to be labeled according to an increasing sequence of Entropy (information entropy) by adopting a pseudo labeling algorithm, selecting the lowest H samples, and taking a prediction result of the model as a pseudo label to serve as additional temporary training data of next training. An SEResNet-152 neural network structure is adopted, the neural network structure is based on a ResNet-152 network model, an SE module is added behind each Resinual module, and the SE module is used for calculating the weight proportion relation between channels ofa feature map.

The invention discloses a classification method for a multi-kernel support vector machine, which relates to the artificial intelligence field, in particular to the data mining technology, and comprises a data pretreatment section, a kernel function selection section, a support vector machine realizing section, and a human-computer interaction section. The work processing comprises that users submit classification request of data to the DPP, then KSP chooses the kernel function, an SILP solution module converts a multi-kernel support vector machine problem to an SILP problem and then solves the problem, a condition detecting module detects whether the condition is satisfied, and if the condition is satisfied, the HIP returns the result to users, otherwise, the parameter and the objective function are updated, and the SILP solution module is transferred to solve. The invention improves the capability of processing complex data of the support vector machine through multi-kernel functions, promotes the complexity of a module and the calculation, and converts the multi-kernel support vector machine problem to a semi-infinite linear program for avoiding the increasing of kernel functions simultaneously, and solves through a method of global convergence.

The invention provides an industrial big data early failure detection method based on image semi-supervising cost-sensitive, and relates to the technical field of failure detection and diagnosis. Theindustrial big data early failure detection method based on image semi-supervising cost-sensitive includes the steps: acquiring the data of the industrial process, and utilizing an image semi-supervising tag propagation method to update tags without marked data and delineating the suspected early failure; for the suspected early failure points, performing cost-sensitive Bayesian classification, and completing secondary updating of the tags of the suspected early failure points; and finally establishing an EDC-SVM (Example dependent cost-sensitive support vector machine) classifier, and performing on-line failure diagnosis on the industrial process. The industrial big data early failure detection method based on image semi-supervising cost-sensitive takes minimization of the diagnosis costas the failure diagnosis target, divides the suspected early failure, and solves the problem that the industrial failure detection error classification cost is high. And at the same time, on the premise of guaranteeing accuracy of classification, the industrial big data early failure detection method based on image semi-supervising cost-sensitive can reduce the misjudgment cost during the failuredetection process, and can improve safety of the industrial process.

The invention discloses a pulmonary nodule feature extraction method based on a convolutional neural network and principle component analysis, which comprises the steps of firstly automatically learning pulmonary nodule features by using the convolutional neural network, performing feature extraction through convolution, and performing feature mapping through downsampling; and secondly, performing dimension reduction on the output of each feature mapping layer in a convolutional neural networks model for feature extraction (FeCNN) by using PCA (Principal Component Analysis) in feature extraction, and integrating with mapping of an output layer to acquire final multi-layer deep integration features. The pulmonary nodule feature extraction method not only can effectively recognize medical symptoms of a pulmonary nodule, but also avoids the complex feature extraction and feature reconstruction process in a traditional method, and plays a role of assisting diagnosis from an objective aspect.

The invention discloses a remote sensing image classification method based on partially random supervision discrete Hash. The method comprises the following steps that: according to calibrated real data, carrying out object segmentation on a remote sensing image, and carrying out characteristic extraction on a segmented object, wherein each object is represented by a characteristic vector, and thecharacteristic vectors of all objects are combined into a characteristic matrix; according to the same ratio, dividing each category of samples into training samples and testing samples; carrying outdiscrete Hash coding on all samples; carrying out partially random Hash coding on all samples; combining the discrete Hash coding with the partially random Hash coding, carrying out iterative optimization on parameters, and finally, obtaining more accurate Hash coding; and according to the generated Hash coding, calculating Hamming distance to finish classification. By use of the method, the problem of high calculation complexity caused by an overlarge data size in a remote sensing image processing process is solved, and the remote sensing image can be quickly and effectively classified.

The invention discloses a traffic real-time classification method based on a shell vector type SVM incremental learning model. According to the method, redundant features and irrelevant features of ahigh-dimensional training set are deleted through a symmetric uncertain FCBF algorithm, then dimension reduction processing is conducted through a linear superposition PCA algorithm, a new training set is obtained, an SVM model is trained according to the new training set, and then network flow classification is achieved; In the incremental learning process, the same data preprocessing is adoptedfor the real-time network flow, the union set of the new sample data violating the KKT condition and the shell vector serves as a new training set to train the SVM model again, and therefore the incremental learning of the SVM is achieved.

The invention provides blood vessel image processing method, device and equipment and a storage medium. The blood vessel image processing method comprises the steps: extracting an original feature mapfrom a to-be-processed blood vessel image; determining the probability value of each pixel point belonging to the blood vessel in the to-be-processed blood vessel image based on the feature of each pixel point in the original feature map so as to form a blood vessel segmentation probability map of the to-be-processed blood vessel image; distributing a corresponding weight for each pixel point based on the distribution condition of the probability values corresponding to each pixel point in the blood vessel segmentation probability graph; fusing the weight of each pixel point in the to-be-processed blood vessel image with the features of the corresponding pixel points in the original feature map to obtain a fusion feature map; and based on the characteristics of each pixel point in the fusion feature map, determining probability values of each pixel point in the to-be-processed blood vessel image belonging to different types of blood vessels, so as to form a blood vessel classificationprobability map of the to-be-processed blood vessel image. According to the invention, blood vessel segmentation and blood vessel classification can be automatically realized with high precision.

The invention provides a remote sensing image classification network robustness improvement method based on self-supervised learning, which not only utilizes labeled data, but also fully utilizes a large amount of non-labeled data existing in the remote sensing field, and effectively improves the robustness of a model by mining the information of images through a twin network. Feature extraction is conducted on the clean sample and the adversarial sample simultaneously by using a twin network to obtain feature vectors, and model training is completed by comparing and learning the feature vectors approaching the clean sample and the adversarial sample, so that the image has stable expression in a deep remote sensing image encoder network in an online network in the twin network, and therefore, the robustness is improved. According to the method, the robustness of the model to adversarial sample noise and natural noise is effectively enhanced, meanwhile, the classification effect of a clean data set is hardly influenced, and application is facilitated.