165 results about "Linear svm" patented technology

A computerized system and method for large scale semi-supervised learning is provided. The training set comprises a mix of labeled and unlabeled examples. Linear classifiers based on support vector machine principles are built using these examples. One embodiment uses a fast design of a linear transductive support vector machine using multiple switching. In another embodiment, mean field annealing is used to form a very effective semi-supervised support vector machine. For both these embodiments, the finite Newton method is used as the base method for achieving fast training

To accurately classify a query as navigational, thousands of available features are explored, extracted from major commercial search engine results, user Web search click data, query log, and the whole Web's relational content. To obtain the most useful features for navigational query identification, a three level system is used which integrates feature generation, feature integration, and feature selection in a pipeline. Because feature selection plays a key role in classification methodologies, the best feature selection method is coupled with the best classification approach to achieve the best performance for identifying navigational queries. According to one embodiment, linear Support Vector Machine (SVM) is used to rank features and the top ranked features are fed into a Stochastic Gradient Boosting Tree (SGBT) classification method for identifying whether or not a particular query is a navigational query.

The invention discloses a quick feedback analyzing system in a tunnel constructing process. The system adopts a scheme: understanding currently adopted designing construction parameters; establishing a tunnel excavation three-dimensional finite element numerical grid calculation model; acquiring surrounding rock layering and convergent displacement monitoring information after a tunnel is excavated; establishing a non-linear support vector machine model; fixing an anchoring parameter according to the actual construction parameter, and optimally identifying rock mechanic parameters by adoptinga differential optimization algorithm; optimizing the construction parameter of an anchoring scheme by adopting a differential evolution algorithm; and optimizing the rock mechanic parameters by calling the differential evolution and optimization algorithms to further solve the construction parameter of the anchoring scheme, and outputting the construction parameter of the optimized anchoring scheme as a construction scheme through a computer display screen to guide the constructors to construct. The quick feedback analyzing system ensures that the monitoring information is used for optimizing the anchoring parameter while being used for identifying the surrounding rock parameters, so that the dynamic information construction is improved to a level of quantitative analysis.

The invention discloses an image classification method based on hierarchical SIFT (scale-invariant feature transform) features and sparse coding. The method includes the implementation steps: (1) extracting 512-dimension scale unchanged SIFT features from each image in a data set according to 8-pixel step length and 32X32 pixel blocks; (2) applying a space maximization pool method to the SIFT features of each image block so that a 168-dimension vector y is obtained; (3) selecting several blocks from all 32X32 image blocks in the data set randomly and training a dictionary D by the aid of a K-singular value decomposition method; (4) as for the vectors y of all blocks in each image, performing sparse representation for the dictionary D; (5) applying the method in the step (2) for all sparse representations of each image so that feature representations of the whole image are obtained; and (6) inputting the feature representations of the images into a linear SVM (support vector machine) classifier so that classification results of the images are obtained. The image classification method has the advantages of capabilities of capturing local image structured information and removing image low-level feature redundancy and can be used for target identification.

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 discloses a mammary gland molybdenum target image automatic classification method based on deep learning. The method comprises the following steps that step one, square image blocks are selected from the cancerous area and the normal area of a mammary gland molybdenum target image by using different sizes of sliding windows, and a training sample set and a test sample set corresponding to each size are constructed for different sizes of image blocks; step two, a convolutional neural network model corresponding to each size is established, and the model is trained by using the training sample set for each size; step three, the accuracy rate of the corresponding convolutional neural network model is tested by using the test sample set for each size, and the convolutional neural network model for the size corresponding to the highest accuracy rate is selected; step four, the overall connection layer characteristics are extracted by using the selected convolutional neural network model; and step five, the extracted characteristics are inputted to a linear SVM classifier for classification so that the classification types of the image blocks are obtained. The overall connection layer characteristics in the convolutional neural network model are extracted to act as the classification characteristics of the image blocks so that the classification speed and accuracy can be enhanced.

The invention relates to a method for re-identifying persons on the basis of deep learning encoding models. The method includes steps of firstly, encoding initial SIFT features in bottom-up modes by the aid of unsupervised RBM (restricted Boltzmann machine) networks to obtain visual dictionaries; secondly, carrying out supervised fine adjustment on integral network parameters in top-down modes; thirdly, carrying out supervised fine adjustment on the initial visual dictionaries by the aid of error back propagation and acquiring new image expression modes, namely, image deep learning representation vectors, of video images; fourthly, training linear SVM (support vector machine) classifiers by the aid of the image deep learning representation vectors so as to classify and identify pedestrians. The method has the advantages that the problems of poor effects and low robustness due to poor surveillance video quality and viewing angle and illumination difference of the traditional technologies for extracting features and the problem of high computational complexity of the traditional classifiers can be effectively solved by the aid of the method; the person target detection accuracy and the feature expression performance can be effectively improved, and the pedestrians in surveillance video can be efficiently identified.

The invention relates to a human movement significant trajectory-based video classification method. The human movement significant trajectory-based video classification method comprises the following steps that: a video set M is divided into a training set Mt and a test set Mv, and human movement information in each video is tracked in a multi-scale space by using SIFT and dense optical flow technologies, so that movement significant trajectories in each video can be obtained; feature description vectors of each trajectory are extracted; redundant information in the feature description vectors are eliminated through using a PCA method, and dimension reduction is performed on each class of feature description vectors; the feature description vectors in the training set Mt are clustered by suing a Gauss mixture model, and then, Fisher vectors of each video in the video set M are generated by using a Fisher Vector method; a linear SVM classification model is constructed on the training set Mt; and on the test set Mv, videos in the test set are classified through using the linear SVM classification model. Compared with the prior art, the human movement significant trajectory-based video classification method of the invention has the advantages of excellent robustness, higher computational efficiency and the like.

The invention discloses a two-way privacy protective system and method for inquiring a medical diagnostic service, used for mainly solving the problem that privacy protection of inquiry information of medical users and data resources of service providers are not related in the prior art. The system comprises a medical diagnostic server and a medical user terminal; the medical diagnostic server completes system initialization, provides login and distributes a secrete key to the medical user terminal, stores data of a nonlinear SVM model built by using the existing medical database, provides a privacy protective medical diagnostic service for the medical user terminal, and sends an inquiry response result to the user terminal; and the medical user terminal sends a service inquiry request to the medical diagnostic server and performs decryption and polynomial aggregation operation of the inquiry response result returned by the medical diagnostic server, such that a pre-diagnosis result is obtained. According to the invention, privacy protection of the inquiry information of the medical users and the data of the nonlinear SVM model can be realized; and the system and the method can be used for providing an online disease prediction service for the medical users.

The invention discloses a vehicle type recognition method with a deep network model based on spatial pyramid pooling. The method comprises the steps that images in a vehicle type database are input into the deep network model for convolution layer characteristic extraction, so that a convolution layer characteristic diagram is constituted; spatial pyramid convolution computation is carried out to each image in the convolution layer characteristic diagram according to different scales, so that a characteristic diagram of spatial pyramid layers is constituted; all the characteristics of the spatial pyramid layers are pooled to constitute full-connected layers, so that a final characteristic representations of vehicle type images can be obtained; the characteristic representations off the vehicle type images are used in training of a linear support vector machine to obtain a vehicle type recognition system; and the characteristic representation of a to-be-recognized vehicle is also acquired and input into the recognition system, so that the vehicle type can be recognized. A traditional deep network model requires that an input image must have a fixed size, so that operations of large-scale vehicle type image data are limited. The method disclosed by the invention adopts the deep network model based on the spatial pyramid pooling, so that the problem is effectively solved; and the method has high practicability and robustness.

The invention provides a color histogram based vehicle body color identification method. The method comprises three main steps of license plate positioning, vehicle body color feature region positioning and vehicle body color identification. The method particularly comprises: firstly, determining a license plate position through a license plate identification technology, and detecting a coarse region of vehicle body color through the height, width and position coordinate information of a license plate; preprocessing the region of the vehicle body color to reduce external environment interferences; further searching the preprocessed coarse region of the vehicle body color to obtain an accurate feature region of the vehicle body color; and converting the feature region into a Lab color space, extracting a color histogram from the Lab space, and performing vehicle body color training and identification by using a nonlinear SVM. The method is capable of intelligently processing traffic vehicle videos and images and automatically identifying the vehicle body color. Compared with other schemes in the same field, the method has very high identification accuracy and is high in robustness in a complicated environment.

The invention discloses a SAR image target identification method based on authentication non-linear dictionary learning. The SAR image target identification method is mainly used for solving the problem of the prior art of low identification precision. The SAR image target identification method is characterized in that 1, a training set SAR amplitude image random face characteristic is extracted, and is used as a training sample, and is mapped to a projection space in a non-linear way; 2, an authentication code matrix is built according to the category number of the training sample; 3, an authentication characteristic training linear SVM classifier is acquired by using the authentication non-linear dictionary learning; 4, a to-be-tested SAM amplitude image random face characteristic is extracted, and is used as a testing sample, and is mapped to the projection space; 5, trained dictionaries are renormalized, and the sparsity of the testing sample, which is acquired by using a KOMP method, is used to express a vector; 6, the authentication characteristic of the testing sample is extracted, and is input in the trained SVM classifier, and then the category of the target in the to-be-tested SAM amplitude image is acquired. The SAR image target identification method is advantageous in that the precision of the target identification is improved, and the SAR image target identification method is used for the classification identification of the target in the SAR image.

The invention provides a facial micro-expression recognition method which can improve the recognition accuracy of micro-expressions. The method includes: acquiring a first training set from a macro expression database to pre-train a 3D convolutional neural network (CNN) and storing a pre-trained network model; acquiring a second training set from a micro-expression database, adjusting the pre-trained network model and extracting a feature vector from the last fully connected layer of the 3D CNN, inputting the feature vector into a linear SVM classifier for training; inputting a to-be-tested image, extracting a feature vector from the last fully connected layer of the 3D CNN and inputting the feature vector into the linear SVM classifier for classification. The present invention relates tothe technical field of image processing and pattern recognition.

The invention relates to the field of the identification method, and particularly relates to a personnel reidentification method based on deep learning and distance metric learning. The identificationmethod comprises the steps that (1) a pedestrian target detection method based on the convolutional neural network is adopted to process the video data so as to detect the pedestrian target in the video; (2) the initial characteristics of the pedestrian target are coded by using an unsupervised RBM network through the bottom-up mode so as to obtain a visual dictionary having sparsity and selectivity; (3) supervised fine adjustment is performed on the initial visual dictionary by using error back propagation so as to obtain the new image expression mode of the video image, i.e. the image deeplearning representation vector; and (4) the metric space closer to the real semantics is acquired by using the distance metric learning method of characteristic grouping and characteristic value optimization, and the pedestrian target is identified by using a linear SVM classifier. The essential attributes of the image can be more accurately expressed so as to greatly enhance the accuracy of pedestrian reidentification.

The invention discloses a pedestrian detection method and a device based on a characteristic combination. The method comprises that: training is performed on positive and negative samples of pedestrians, Hog characteristics and LBP characteristics of the samples are extracted and combined to form characteristic vectors of the samples; then a cascaded rejection mechanism and the characteristic vectors of the samples are combined via a cascaded linear SVM classifier which performs classification from simple ones to complicated ones so that a cascaded linear SVM pedestrian classifier is realized. The device comprises: (1) an image acquisition module: images in front of a vehicle are acquired by utilizing a vehicle-mounted pick-up head; (2) an image detection and processing module: de-noising and enhancement processing is performed on the acquired images, and pedestrian detection is performed by utilizing the loaded and trained classifier; and (3) an image display module: the pedestrians in the acquired images are identified by using rectangle frames and prompted on a vehicle-mounted DVD display screen. Technical problems of reducing false detection rate and enhancing accuracy are mainly solved by the pedestrian detection method and the device based on the characteristic combination.

The invention provides a histopathological image classification method based on the convolutional neural network. The method mainly comprises steps that a data set is established, a fine-tuning protocol is adopted and carried out, a pre-trained convolutional neural network is taken as a characteristic extractor and a fine-tuned CNN is taken as a classifier, for the process, firstly, images of different body parts with different texture patterns are manually selected from the scan, patch selection and accuracy calculation are carried out, secondly, a specific single-layer complete connection layer is selected for network training, thirdly, the pre-trained CNN is taken as the characteristic extractor, depth characteristics are utilized to train the linear support vector machine (SVM) for classification, and lastly, the fine-tuned CNN is taken as the classifier, and the acquired network is utilized to classify test patches. The method is advantaged in that the data set is utilized to design and train the deep network, details are richer, the data is more perfect, retrieval and classification accuracy is significantly improved, and classification of histopathological images is effectively realized.

The invention discloses a double-layer-cascade-based facial feature detection method. According to the method, a sparse feature is designed for an image including a face at a first level and a target candidate frame is obtained by using an SVM learning feature; localization of a local feature point is carried out by using a face alignment method at a second level; replacement is carried out by using a face feature point directly according to a feature extraction method using an SIFT; and then a false detection window is rejected by using a linear SVM learning feature to realize facial feature detection, wherein a result of each time is used as a sample fed back to the SVM for learning. According to the invention, because the first-layer candidate window is determined and the result of each time is used as a sample fed back to the SVM for learning, the detection speed increases; with the face alignment method, corresponding model establishment for various attitudes of the face is not required; and with combination of the high-precision SIFT feature detection method, the false detection rate is reduced effectively.

An image analysis and deep learning-based number of people statistical method disclosed by the present invention comprises the following steps of A carrying out the pyramid model calculation on an input image, and generating the images of a plurality of resolutions and sizes; B sliding the windows on each layer of a pyramid, calculating the HOG feature values of the window areas, classifying via a linear support vector machine (SVM) classifier, and determining whether the windows are the head-shoulder areas; C for each head-shoulder area given out in the step B, extracting the corresponding image, normalizing to a set same size, and inputting in a deep neural network to obtain the classification output; D carrying out the non-maximum suppression on the all head-shoulder windows outputted in the step C to merge the overlapped detection results of the adjacent areas and scales. The image analysis and deep learning-based number of people statistical method of the present invention can improve the insufficiency of the prior art, and can realize a higher number of people statistical performance with the faster speed.

The invention aims to solve the problems that the remote sensing image water area segmentation extraction method in the prior art is poor in self-adaptation due to the fact that a segmentation critical value is manually set, a large number of non-water-area land types exist in a result, and a large number of impulse noise exists in the result. The invention provides a remote sensing image water area segmentation and extraction method for super-pixel classification and identification. In combination with an improved linear clustering super-pixel segmentation method, a remote sensing image is divided into a plurality of super-pixels which are good in homogeneity, compact in layout and capable of well keeping edge information; superpixels are used as a feature extraction unit, water area features in a remote sensing image are extracted from three perspectives of spectrum, texture and terrain, the features of a water area and non-water areas are described more accurately, a typical learning sample library is constructed, and a nonlinear support vector machine is used for supervised classification. Experimental results show that the method can overcome the defects of the prior art and remarkably improve the water area segmentation and extraction precision and speed of the remote sensing image.

The invention provides a traffic light detection method and system based on an adaptive background suppression filter and a combined direction gradient histogram, which is used for detecting and identifying various traffic signal lights appearing in a real road traffic environment. In this method, the color of the image is normalized by Gamma correction method, and the RGB space image is convertedinto gray image by gray processing. By extracting HOG and local RGB color histogram, the HOG-RGB combinatorial feature is used as the basic feature descriptor for detecting traffic lights, The adaptive background suppression filtering method proposed by the invention is used for detecting candidate areas of traffic lights. After a series of candidate windows are obtained, the candidate areas arefurther verified and recognized as different semantic types by using a linear SVM classifier obtained by training, and the specific types of traffic lights are confirmed. The invention utilizes the adaptive background filtering method to extract and detect the traffic signal lamp features, and has good robustness to different weather and illumination in actual driving environment.

The invention discloses an image semantic extraction method applied in an electronic guidance system. Image semantics can be rapidly and accurately extracted, and requirements of the electronic guidance system based on computer vision for image understanding can be satisfied. The image semantic extraction method comprises: 1) a training stage: building a training image library T, and extracting ascale invariant feature transform (SIFT) features of images in T to form a set F; constructing a dictionary V through a multi-stage close relationship propagation algorithm; mapping F onto Fv throughthe dictionary V and sparse codes; and training a linear support vector machine (LSVM) through Fv; and 2) a use stage: dividing a collected image Iq into 10 equal sub-blocks overlapped partially; extracting features of each sub-block through the feature extraction method; classifying the features of each sub-block through the LSVM to obtain corresponding semantic information; and determining a semantic tag of the image Iq according to the semantic information of the sub-blocks, and converting the semantic tag into voice output.

The invention discloses a human face identity recognition method based on a GB(2D)2PCANet depth convolution model.A model training method includes the following steps that preprocessed human face samples are sequentially fed into a first feature extraction layer, multiple sub-blocks are scanned from obtained Gabor feature images, mean removal is conducted, an optimal projection axis is extracted through (2D)2PCA and convoluted with a training set original sample, and a first layer of feature map is obtained; the first layer of feature map is fed into a second feature extraction layer, the steps are repeated, and a second layer of feature map is obtained; a feature map is output in a binarized mode, and local area histograms are calculated and spiced to serve as final features; the final features are fed into a linear SVM classifier, and an optimized human face identity recognition model is obtained.Effective feature expression can be automatically learnt, good locality is achieved, good robustness is achieved for illumination, expressions, noise and the like, and the recognition performance of human face identities is improved.

The present invention discloses a linear SVM (support vector machine) classification service query system and method with two-way privacy protection, for mainly solving the privacy protection problems of user query information and SVM model data which are not involved in the prior art. The system comprises a trusted authentication center, a classification query server and a user terminal. The trusted authentication center completes system initialization to provide registration and key distribution for the user terminal and the classification query server. The classification query server stores linear SVM classifier model data that have been establised, provides a classification query service of the privacy protection for the user terminal, and sends a query result to the user terminal. The user terminal sends a query service request to the classification query server, performs decryption and polynomial aggregation operation to the query result returned by the classification query server, and obtains final classification information. According to the system and the method provided by the present invention, the privacy protection of the user query information and the linear SVM model data is implemented. The system and the method can be used to provide a user with an online DNA query and a disease prediction service.

The invention provides methods and systems for reliably detecting objects in a received video stream from a camera. Objects are selected and a bound around selected objects is calculated and displayed. Bounded objects can be tracked. Bounding is performed by using Histogram of Oriented Gradients and linear Support Vector Machine classifiers.

The invention discloses a human body behavior recognition method based on a convolutional neural network and a recurrent neural network, and the method comprises the steps: tracking a human body behavior through a sensor, and collecting a three-dimensional coordinate vector group and an RGB video of a human body joint in a time period; training the three-dimensional coordinates of the human jointby using a recurrent neural network RNN to obtain a time feature vector; and training the RGB video by using a convolutional neural network (CNN) to obtain a space-time feature vector, finally combining the time feature vector and the space-time feature vector and normalizing, feeding the normalized space-time feature vector and the normalized space-time feature vector to a classifier of the linear SVM, finding a parameter C of the linear support vector machine (SVM) by using the verification data set, and finally obtaining a comprehensive identification model. According to the method, the problem of overfitting of action classification of the model in the model training process can be solved, and meanwhile, the human body behavior recognition efficiency and accuracy can be effectively improved.

The invention discloses an image scene classification method based on multi-characteristic fusion; the method comprises the following steps: firstly extracting image GIST characteristics, SIFT characteristics and PHOG characteristics; carrying out local constraint linearity coding for image SIFT characteristics, and pooling SIFT characteristic sparse coding so as to obtain a sample image SIFT characteristic sparse vector; cascading the sample image GIST characteristics, SIFT characteristics and PHOG characteristics so as to form the sample image final characteristic expression, and inputting the sample image final characteristic expression into a linear classifier for training; using said method to extract image expressions from to-be-classified images, and inputting the image expression into a trained linear SVM classifier so as to finish the classification work. The method can improve the classification precision, and can enhance the system robustness.

The invention discloses a linear SVM model training algorithm for privacy protection based on vector homomorphic encryption and belongs to the field of information technology security. The method comprises the following steps: step 1, a user encrypts a training data set by using a vector-based homomorphic encryption scheme VHE and sends the encryption result to a server; step 2, the server calculates the encryption result, obtains a ciphertext linear kernel function matrix and returns the ciphertext linear kernel function matrix to the user; step 3, the user decrypts the ciphertext linear kernel function matrix to obtain a plaintext linear kernel function matrix and sends the plaintext linear kernel function matrix to the server; step 4, the server adopts a ciphertext SMO algorithm to train the plaintext linear kernel function matrix and returns the training result to the user.

The invention relates to an eye positioning method applied to face recognition. The method includes the following steps that: a light reflecting area of an image is detected and removed; a face is detected with the Viola-Jones method of the Ada-Boost algorithm; the normalized gradient vector of a face area is calculated, binarization is carried out, a black frame of eyes is detected, the average gray value of adjacent pixels which are not located in a black frame area is adopted to replace the gray value of pixels which are in the black frame area; an eye training set and a non-eye training set are constructed, the non-linear SVM (Support Vector Machine) of a quadratic kernel function is trained, calculation and evaluation are carried out on areas which adopt pixels around the eyes as centers, a pixel with the maximum value is evaluated and is adopted as an eye position, the evaluation is named as confidence; and if the confidence is greater than a set threshold value, the eye position is a final positioning result, otherwise PCA (Principal Component Analysis) is adopted to evaluate the eye position, rotation and zooming transformation is performed on the face area, the Gabor coefficients of transformed images are calculated, and the confidence of face detection is calculated, an image with the largest confidence is selected, and the average position of eyes in the image is adopted as the eye position of the original image.

The invention discloses an extreme learning machine and color feature fusion based pedestrian gender identifying method. The method includes extracting extreme learning machine features of training images whose gender attributes are not marked; extracting HSV color features of input training image whose gender attributes are not marked, combining the extreme learning machine features with the color features and thus obtaining fusion features, and training a pedestrian gender classifier by utilizing a linear SVM (Support Vector Machine) according to the fusion features and training image labels; extracting image features of a to-be-tested image by utilizing a model obtained through training and extracting HSV color features of the to-be-tested image, fusing the two kinds of features and thus obtaining the fusion features of the to-be-tested image, and classifying the fusion features by utilizing the pedestrian gender classifier of the linear SVM obtained in the training process. According to the invention, extreme learning features and color features of the input images are extracted and fused effectively, mutual complementation of the two kinds of features is realized, and pedestrian gender attributes are captured more effectively, so that pedestrian gender identification rate is improved.

The invention discloses a pedestrian detection method based on binary image improved HOG characteristics. The method comprises steps that a pedestrian training sample database is established; binary processing on sample images is carried out; improved HOG characteristic vectors of the binary images are extracted; Gauss normalization for the improved HOG characteristic vectors is carried out, positive and negative samples are utilized for training to acquire various parameters of an SVM model, and a linear SVM model is established; to-be-detected video frame images are pre-processed to acquire binary images; the present improved HOG characteristic vectors are acquired through calculation; the improved HOG characteristic vectors are inputted to the linear SVM model, if model output is determined to be a positive sample, a target is detected, and the target position is outputted; whether each traversal window has the target is detected in a window traversal mode. The method is advantaged in that problems of large memory consumption and slow detection speed in a pedestrian detection process can be effectively solved.