367results about How to "Fast classification" patented technology

A welding visual detection method based on a convolutional neural network includes the following steps that firstly, at the training stage, a training sample is input into the convolutional neural network, and the connection weight and offset value of the convolutional neural network are obtained; secondly, at the testing state, a welding picture is read in and preprocessed with the digital picture processing technology, and a region of interest is extracted and then subjected to picture size normalization processing to serve as input of the convolutional neural network. The invention further provides a welding visual detection device based on the convolutional neural network. The welding visual detection device based on the convolutional neural network comprises a crawling mechanism, a power transmission mechanism, visual detection equipment and a weld defect detection and analysis system. By means of the welding visual detection method and device based on the convolutional neural network, the automaton and intelligence level is improved, the detection precision is effectively improved, and the detection speed is effectively increased.

The invention discloses a parallel data mining method for identifying a mass of mobile client bases. The parallel data mining method includes the steps of building a client value model and a client behavior model, classifying clients according to the client value model and the client behavior model, popularizing assigned preference services to the clients with the high purchasing power and the high potential purchasing inclination, and then achieving accurate marketing. According to the parallel data mining method, the mass of mobile client bases can be identified, the aspects such as Internet surfing time preferences, Internet surfing place preferences and browsed website preferences of the clients can be identified, and the social group classes of the clients can be accurately judged. Clustering and classifying can be rapidly carried out through the adopted parallel clustering algorithm and the adopted parallel classifying algorithm. By means of the parallel data mining method, different strategies can be formulated for the different client bases by an enterprise, and the important guiding function for profit maximization of the enterprise is achieved.

The invention discloses a panoramic image CNN based tunnel disease automatic identification device. According to the device, firstly, a panoramic image of a tunnel inner wall is quickly acquired with a panoramic vision sensor; then the panoramic image is processed mainly by panorama unrolling, image preprocessing, binarization processing and the like to extract suspected disease region; and finally, the disease is subjected to automatic detection and classification identification by adopting a convolutional neural network. According to the scheme, the structure of the detection device in extracting a tunnel inner wall panoramic image is greatly simplified, and various tunnel disease characteristics can be automatically extracted, detected and identified by the end-to-end convolutional neural network, and an effective technical support is provided to tunnel maintenance and completion acceptance.

The invention discloses a face detection method based on a convolutional neural network. The face detection method comprises a training stage and a testing stage, wherein the training stage comprises the following steps: inputting a training sample into the convolutional neural network, and obtaining a connection weight and an offset value of the convolutional neural network. The testing stage comprises the following steps: reading a video image; when a moving object is detected, extracting an interested area, moving for one pixel each time by utilizing a block with a n*n pixel size, and carrying out partitioning processing to the interested area to obtain a plurality of pictures; zooming the partitioned pictures to a size which is the same with the size of the training sample; and inputting the pictures into the convolutional neural network which finishes training for classification, wherein n is greater than or equal to 50 and less than or equal to 70. The convolutional neural network is used as a classifier, so that detection precision and speed can be improved.

The invention relates to an RGB-D image classification method and system. The method comprises the steps of: S1, utilizing a convolution neural network (CNN) to process a source RGB image and a Depth image respectively, and extracting low level characteristics; S2, utilizing a recursion neural network (RNN) to carry out feedback learning on the image low level characteristics, and extracting image middle level characteristics; S3, adopting a block interior constraint dictionary learning method, carrying out characteristic set sparse expression on the image middle level characteristics, and obtaining high level characteristics of the RGB-D images; and S4, inputting the high level characteristics of the RGB-D images into a linear SVM to complete the classified identification of the RGB-D images. According to the invention, automatic characteristic extraction of the images is realized, learning RGB-D image characteristic expressions can effectively distinguish classification of noise data from high similarity images, and the classification precision of the RGB-D images is improved; in addition, the linear SVM is utilized, and the image classification speed is improved.

The invention relates to an environment noise identification classification method based on a convolutional neural network. The method comprises the following steps of: S1, extracting natural environment noise, and editing the natural environment noise into noise segments with duration of 300ms to 30s and a converted frequency of 44.1kHz; S2, carrying out short time Fourier transformation on the noise segments, and converting a one-dimensional time-domain signal into a two-dimensional time-domain signal to obtain a sonagraph; S3, extracting a MFCC (Mel Frequency Cepstrum Coefficient) of the signal; S4, forming a training set with 80% of all the noise segments and forming a testing set with the residual 20% of all the noise segments; S5, carrying out noise classification by a convolutionalneural network model; and S6, training a classification model by the training set, and verifying accuracy of the model by the testing set so as to complete environment noise identification classification based on the convolutional neural network. According to the invention, the sound segments are input, sound feature information is extracted, an output is a classification result, and automatic extraction on the sound feature information can be implemented.

The invention discloses an image classification method based on MapReduce, on Hadoop platform, firstly, using the MapReduce frame for parallel extracting image SIFT characteristic; using the MapReduce frame for sparse coding for extracted SIFT characteristic of each image and obtaining the corresponding sparse vector of the image, generating the sparse characteristic of the image; then, based on the sparse characteristic of the image, using MapReduce frame for training the decision-making tree, generating the random forest aiming at the image characteristic set; using MapReduce combined with the random forest for classified counting each image. Through experimental verification, the image classification method based on MapReduce can obviously raise the classification speed while guaranteeing not lower than single platform classification precision.

The invention discloses a photo classifying method and a photo classifying device. The photo classifying method comprises the following steps: judging whether a to-be-classified photo has a face area by adopting a face detection algorithm; if the to-be-classified photo has a face area, performing feature extraction on all to-be-classified photos with a face area by use of a lightweight deep model network, and obtaining the face recognition features of the to-be-classified photos and at least one face attribute information; clustering the to-be-classified photos based on at least one face attribute information to obtain multiple clusters; and clustering the to-be-classified photos of each cluster based on the face recognition features to obtain multiple classes of photo albums, wherein each class of photo album stores the to-be-classified photos of the same person. The photo classifying method can realize offline automatic classification of the to-be-classified photos, the classification result is accurate, and the classification speed is high.

The invention discloses a binary tree-based SVM (support vector machine) classification method. The binary tree-based SVM classification method comprises the following steps: 1, acquiring signals, namely detecting working state information of an object to be detected in N different working states through a state information detection unit, synchronously transmitting the detected signals to a data processor, and acquiring N groups of working state detection information which corresponds to the N different working states; 2, extracting characteristics; 3, acquiring training samples, namely randomly extracting m detections signals to form training sample sets respectively from the N groups of working state detection information which are subjected to the characteristic extraction; 4, determining classification priority; 5, establishing a plurality of classification models; 6 training a plurality of classification models; and 7, acquiring signals in real time and synchronously classifying. The binary tree-based SVM classification method is reasonable in design, easy to operate, convenient to implement, good in use effect and high in practical value; and optimal parameters of an SVM classifier can be chosen, influence on the classification due to noises and isolated points can be reduced, and classification speed and precision are improved.

The invention discloses a semi-supervised dimension reduction-based hyper-spectral image classification method for mainly solving the problems of high calculation quantity caused by over high hyper-spectral image data dimensions and low classification accuracy of the conventional method. The method comprises the following steps of: expressing each pixel point of a hyper-spectral image by using a feature vector, and selecting a marked training set, a test set and a total training set; constructing local inter-class and local intra-class dissimilarity matrixes of the marked training set respectively to obtain a total local dissimilarity matrix; constructing and solving a feature value equation to obtain a projection matrix; projecting the marked training set and the test set to a low-dimensional space respectively to obtain a new marked training set and a new test set; and inputting the new marked training set and the new test set to a support vector machine, and performing classification to obtain class information of the test set. By adopting the thought of semi-supervision, higher classification accuracy can be acquired; and the method can be applied to the fields of map design, vegetation survey and military intelligence acquisition.

The invention relates to the technical field of classification of pedestrians and vehicles, and discloses a method and a system for achieving the classification of pedestrians and vehicles based on a neural network. The method includes the following steps of collecting a plurality of training samples, classifying the training samples by a convolutional neural network, and thereby obtaining a classifier including tab results, when the pedestrians and the vehicles are classified, reading a video image to be detected, detecting moving objects in the image, and processing the image in blocks according to the moving objects; and then classifying the image blocks by the classifier to obtain a detection result; therefore, the neural network system can be simply constructed as the classifier; the system is trained by using different pedestrian and vehicle samples such that the system automatically learns the complex class conditional density of the samples, and problems caused by an artificial hypothesis class conditional density function are avoided. Compared with existing methods for classifying pedestrians and vehicles, the method for achieving the classification of pedestrians and vehicles based on the neural network has the advantages of improving classifying accuracy as well as classifying speed.

The invention relates to a method for carrying out mangrove forest map making on intermediate resolution remote sensing image by utilizing object-oriented classification method, which relates to a method for mangrove forest map making and solves the problems of time and labor waste, bad timeliness and serious neglected and wrong classification of mangrove forest in positioning and map making of mangrove forest map making through conventional means at present. The method comprises the following steps: firstly, carrying out ortho-rectification and geometric exact correction on Landsat TM data so as to obtain Landsat TM images after registration; secondly, carrying out multi-layered multi-dimensioned division on the Landsat TM images after registration, wherein each division unit is used as an object; thirdly, extracting textural and topological characteristics, and calculating normalized vegetation index and ground surface humidity index; fourthly, removing a non-vegetated object so as to obtain a vegetated object; fifthly, extracting a mangrove forest object from the vegetated object; sixthly, exporting the mangrove forest object so as to generate a mangrove forest vector; and seventhly, manufacturing a mangrove forest thematic map. The method disclosed by the invention is used for mangrove forest map making.

A website classification method and system are disclosed for determining the website category more rapidly and accurately, the method comprises the steps of: setting feature vectors for each website, in which each single dimension of the feature vector is a different user query word, the value of which is equal to the emergence times of correspondent query word, classifying the website. The query words are more representative than the common words in the webpage, and can benefit the classification of websites and improve the classification accuracy because the emergence times of the query words is counted by the click times of the user which represents the close relationship between the clicked website and the query word, in addition, the generated feature vector is very short which can improve the classification efficiency greatly.

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 discloses a vehicle logo detection method based on two-layer classification. The method comprises the following steps: firstly acquiring a vehicle logo and background in a windshield image in a training data set, and then respectively training a vehicle logo image block detector and a vehicle logo image block classifier; for a test image, firstly acquiring gradient feature vectors of image blocks in the image, and acquiring a candidate vehicle logo image block by using the trained vehicle logo detector, and then computing a HOG (histogram of oriented gradients) feature vector, an LBP feature vector and a color histogram feature vector for the acquired candidate vehicle logo image block, splicing the computed vectors into the feature vector of the image block, and classifying by using the trained vehicle logo classifier to acquire the category of the vehicle logo. The vehicle logo detection process disclosed by the invention is simple and easy to implement, the detection accuracy is high, and the detection speed is fast.

The invention discloses a medical image disease classification method based on naive Bayes. According to an equipment type and image finding, diagnosis and other text information in a diagnosis report form, the disease type which an image examination result belongs to is automatically judged. Considering the influence of the independence assumption of naive Bayes classification in actual application, the method carries out disease clustering analysis by utilizing a K-Means clustering algorithm, data with the high similarity level are classified into the same cluster, data with the low similarity level are classified into different clusters, and meanwhile the number of disease categories is determined. The characteristics of high efficiency and high speed of a naive Bayes algorithm are utilized, classification precision is guaranteed, and meanwhile classification speed of medical image search is improved to a large degree.

The invention provides a human body posture identification method and system, and belongs to the technical field of security engineering. The method comprises the following steps of (1) at least collecting six parameters, including linear acceleration and angular acceleration, of a human body in X, Y and Z directions of space; (2) analyzing data features presented by the parameters, and selectingt data statistics features for each type of data as condition attributes for judging human body postures, wherein t is an integer greater than 1; (3) extracting the data statistics features of a dataset under various human body postures to form a sample set, and dividing the sample set into training samples and test samples; (4) according to the training samples and the test samples, building a decision tree classification model; and (5) based on the decision tree classification model, performing real-time classification on collected data to identify the human body postures. The human body posture identification method and system is high in identification accuracy, high in identification speed, high in adaptability to multiple input samples, and strong in robustness; and output of a classification result has high error tolerance.

The invention provides an improvement-based KNN (K Nearest Neighbor) text classification method. The method comprises the following steps: preprocessing a training text, computing the feature vector of each training sample, and constructing a feature vector spatial model of a training set; defining a density and a distance, defining a density and a distance, defining a whole sample space into a plurality of spherical regions and outliers according to types, and storing as a training set library; during testing, judging whether a text to be tested falls into a certain spherical region, judging the type of the text to be tested according to a corresponding mark number, otherwise, using the outliers and the center point of each sphere as a training set library, calling a KNN algorithm, and judging the type of the text to be tested. By adopting the method provided by the invention, the classification speed, classification accuracy and data skew sensitivity are considered. The method can be well applied to the classification problem of non-spherical distribution, and is particularly suitable for a text classification problem having a high-dimension feature vector and a distribution irregularity feature.

The invention discloses a pedestrian identification method based on a gradient cascade SVM (Support Vector Machine) classifier. In the method, pedestrian identification is carried out by use of a risk-sensitive classifier and a voting mechanism, the detection precision is improved to a certain degree; and furthermore, a gradient cascade architecture is judged by use of an assembly line type from coarse grains to fine grains, negative samples which are easily judged can be completely detected in first levels, the classification speed in the detection phase is greatly accelerated, the operation efficiency and the total processing speed are improved. The method can be adapted to pedestrian identification in scenes with a small pedestrian flow and a lot of non-pedestrian targets.

The invention relates to a convolutional-neural-network-based first lead electrocardiogram heartbeat classification method. The method comprises: S1, preparing trained first lead electrocardiogram heartbeat data and a corresponding class label; S2, designing a convolutional neural network structure for the first lead electrocardiogram heartbeat by using a one-dimensional convolutional neural network and setting the number of convolutional layers as well as the number and sizes of characteristic maps of each convolutional layer; S3, carrying out forward calculation; to be specific, inputting normalized 196-length first lead heartbeat training data into the CNN, the data are processed by all intermediate layers successively, and extracting features of the intermediate layers; S4, carrying out error back propagation; to be specific, calculating classified losses and carrying out back propagation of the loses based on a chain rule; S5, updating a weight by using a gradient descent method;and S6, carrying out iterative training.

The embodiment of the invention discloses a method and device for classifying voice data. The method for classifying the voice data comprises the steps that first voice data of the category to be recognized are obtained; windowing is carried out on the voice time shaft of the first voice data according to a preset windowing algorithm; an MFCC feature vector is extracted for the voice data in each window; vector quantization is carried out on each MFCC feature vector to obtain a one-dimensional first feature value; all the first feature values are calculated according to a preset histogram drawing algorithm to obtain a first histogram of the first voice data; similarity calculation is carried out on the first histogram and preset histogram feature templates corresponding to the voice data of various voice categories, and a first histogram feature template most similar to the first histogram is obtained; the voice category of the feature template is the voice category of the first voice data. Compared with the prior art, according to the technical scheme, the accuracy and speed of classifying the voice data are improved.

The invention discloses a deep-network-based tissue segmentation method of a panoramic digital colorectum pathology image. The method comprises the following steps: step one, acquiring a panoramic digital colorectum pathology image; step two, segmenting the panoramic digital colorectum pathology image; step three, establishing a training sample image; step four, extracting deep features of tissuesof different types; step five, determining types of tissues in the segmented images by using a classifier based on the extracted deep tissue features; step six, splicing image classification resultsand determining a tissue class of an overall picture; and step seven, splicing the images according to block coordinates. According to the invention, the panoramic digital colorectum pathology image is segmented and tissue type marking is carried out on all segmented images by using a sliding window and a trained model; and tissue type determination is carried out by using the classifier based onthe extracted deep tissue features to obtain an image classification result. The classification becomes accurate and the classification speed is increased.

The invention provides a motor imagery electroencephalogram classification method and system based on Riemannian distance. The motor imagery electroencephalogram signal classification method based onthe Riemannian distance comprises the steps of expressing Riemannian space conversion, specifically, motor imagery electroencephalogram vector signals of known category labels by adopting a sample covariance matrix, and acquiring a sample covariance matrix of the known category labels; calculating a Riemannian average value: calculating the Riemannian average value between the sample covariance matrixes of the labels of the known categories to obtain Riemannian average values with the same number as the labels of the known categories; a Riemannian distance calculation step: respectively calculating Riemannian distance values between the sample covariance matrix corresponding to the to-be-classified motor imagery electroencephalogram vector signals and Riemannian average values with the same number of known category labels; and a category output step: taking the category corresponding to the minimum one of the Riemannian distance values as a category label of the motor imagery electroencephalogram vector signal to be classified.

The invention provides a high-performance OpenFlow virtual flow table searching method. Aiming at the problem of high flow table searching overhead in an OpenFlow virtual switch, a flow table item caching mechanism is firstly designed, so that most of data packets directly hit a cache to find corresponding flow table items, and a mask detection process and a flow table searching process corresponding to mask detection are bypassed; secondly, an SMA1 mask heuristic strategy is designed, and the successfully detected active masks forward are moved by one position each time, so that all the active masks are adjusted to the front positions of the mask arrays after a series of accesses, thereby reducing the average mask detection times of the data packets. In addition, an extensible counting type SCBF filter is adopted to quickly judge a flow table searching failure result, meanwhile, it is guaranteed that the false positive error rate is always kept at a low level, and therefore the flowtable searching traversal process when mask detection fails is avoided. According to the method, the average flow table searching overhead of the data packets can be remarkably reduced, the packet classification speed is greatly increased, and the packet switching performance of the OpenFlow virtual switch is effectively improved.

The invention discloses a coin classification mechanism, comprising a receiving tray and a plurality of classification tracks, wherein the receiving tray is used for receiving coins entering the coin classification mechanism and provided with an outlet through which coins slip out; the classification tracks approach the front end of the receiving tray and leave the rear end of the receiving tray, the classification tracks are used for receiving coins slipping out from the outlet and guiding the coins to move from the front end to the rear end, a plurality of through holes through which coins fall are formed in the inner wall of each classification track, the through holes are arranged in the length direction of the classification track, the through holes are gradually enlarged from the front end to the rear end, and coins with one size fall from one through hole. The coin classification mechanism is provided with a plurality of classification tracks, and a plurality of coins can slip in one classification track, so that the classification efficiency and the classification speed are improved, and the coin classification mechanism is suitable for processing a large amount of coins.

The invention discloses a garbage sorting device and method based on deep reinforcement learning. The device comprises a workbench, a mechanical arm, an AI calculation unit and a camera which are arranged on the workbench. The workbench is provided with a to-be-sorted region for placing to-be-sorted garbage. The camera acquires image data regularly and sends the image data to the AI calculation unit, and the image data comprises the situation of the to-be-sorted garbage in the to-be-sorted region and the state of the mechanical arm; the AI calculation unit adopts a deep reinforcement learningmodel to calculate the current optimal action of the mechanical arm according to the image data acquired by the camera, outputs a corresponding control instruction and sends the control instruction tothe mechanical arm; and the mechanical arm acts according to the control instruction outputted by the AI calculation unit, grabs, moves and throws the to-be-sorted garbage so as to complete garbage sorting. The optimal action of the mechanical arm is obtained by adopting the deep reinforcement learning model without depending on garbage positioning and recognition, the classification speed is high and the classification accuracy is high.

The invention discloses a convolutional neural network cloth defect detection method based on an extreme learning machine. The method comprises the steps of carrying out feature extraction by constructing a convolutional neural network; fusing the extracted features; extracting cloth defect candidate boxes on the fused feature layer; in a cloth defect detection stage, carrying out regression on the extracted cloth defect candidate boxes by using the convolutional neural network; classifying the cloth defect candidate boxes by using an extreme learning machine; calculating loss by combining classification and regression results with real labels of sample pictures; updating the weight in the network by using a random gradient descent method based on the obtained loss; carrying out continuousiterative training until the loss of the network converges to a minimum value or reaches a preset training round number, and thus obtaining a trained cloth defect detection network model based on theextreme learning machine, so cloth defect detection can be carried out. The overall performance of cloth defect detection is effectively improved, and the cloth defect detection method has higher cloth defect detection accuracy.

The invention discloses a method and equipment used for quickly sorting and searching music; wherein, the method comprises the steps as follows: a music file is input; acoustic feature of each frame of the input music file based on MDCT is extracted; the energy of each frame is calculated; the acoustic features of each frame are sorted according to the energy size; the method also comprises the steps as follows: the sorted acoustic features are divided into a plurality of sections; average value and standard offsets are calculated according to one or more sections; the calculated average value and standard offset are combined into vectors. The short-time music feature (namely MFCC and timbre characteristic) used by the method and the equipment are directly gained from MDCT coefficients; therefore, the speed of feature extraction is extremely fast. In order to sort one piece of music, only part of the music file (12s duration) needs to be decoded.

The invention discloses a hyperspectral image classification method based on K nearest neighbor filtering. The classification process mainly includes (1) support vector machine (SVM) classification: rough classification of a hyperspectral image using a SVM classifier to obtain an initial probability graph; (2) principal component analysis dimensionality reduction: dimensionality reduction of the hyperspectral image by way of principal component analysis to obtain a first principal component image; (3) K nearest neighbor filtering: extraction of spatial information of the hyperspectral image under the guidance of the first principal component image based on a non local K nearest neighbor filter to optimize the initial probability graph; and (4) accurate classification of the hyperspectral image according to the optimized probability graph. The greatest advantage of the method in the invention over a traditional hyperspectral classification algorithm is that the non local spatial information of the hyperspectral image can be extracted for optimized classification without solving for a complex global energy optimization problem. Thus, the classification speed is high, and the accuracy is high.

The invention discloses a motor imagery electroencephalogram signal classification method based on deep learning, and belongs to the technical field of mode recognition and brain-computer interfaces.The method comprises the following steps: analyzing and preprocessing electroencephalogram signals, carrying out two-dimensional conversion on the electroencephalogram signals, and constructing a convolutional neural model of motor imagery electroencephalogram signal classification; and training the neural network to obtain a motor imagery electroencephalogram signal classification model, and carrying out test and performance evaluation on the model. Compared with the prior art, the method has the advantages that the data of each time point are used as samples, 2D conversion and deep learningmethods are carried out, the classification accuracy of the motor imagery electroencephalogram signals is improved, and meanwhile the classification instantaneity and stability are improved.