614 results about "Feature combination" patented technology

The invention discloses a face recognition method, device, system and apparatus based on a convolutional neural network (CNN). The method comprises the following steps: S1, face detection: using a multilayer CNN feature architecture; S2, key point positioning: obtaining the key point positions of the face by using connecting reference frame regression networks in a cascaded way in deep learning; S3, preprocessing: obtaining a face image with a fixed size; S4, feature extraction: obtaining a feature representative vector by means of a feature extraction model; and S5: feature comparison: determining the similarity according to a threshold or giving a face recognition result according to distance sorting. The method adds a multilayer CNN feature combination to a traditional CNN single-layer feature architecture to treat different imaging conditions, trains a face detection network with good robustness in the monitoring environment from massive image data sets based on a deep convolution neural network algorithm, reduces a false detection rate, and improves detection response speed.

The invention relates to auxiliary medical diagnoses, and aims to provide a method for automatically identifying whether a thyroid nodule is benign or malignant based on a deep convolutional neural network. The method for automatically identifying whether the thyroid nodule is benign or malignant based on the deep convolutional neural network comprises the following steps: reading B ultrasonic data of thyroid nodules; performing preprocessing for thyroid nodule images; selecting images, and obtaining nodule portions and non-nodule portions through segmentations; averagely dividing the extracted ROIs (regions of interest) into p groups, extracting characteristics of the ROIs by utilizing a CNN (convolutional neural network), and performing uniformization; taking p-1 groups of data as a training set, taking the remaining one group to make a test, and obtaining an identification model through training to make the test; and repeating cross validation for p times, and then obtaining an optimum parameter of the identification model. The method can obtain the thyroid nodules through the automatic segmentations by means of the deep convolutional neural network, and makes up for the deficiency that a weak boundary problem cannot be solved based on a movable contour and the like; and the method can automatically lean and extract valuable feature combinations, and prevent the complexity of an artificial feature selection.

The invention discloses an image stitching method based on unmanned aerial vehicle POS information and image SURF feature combination and relates to the digital image processing field, the GIS field,the survey field and other relevant fields. According to the method, first, geometric correction is performed on images; second, geographic coordinates of four corners of each image are calculated; based on the geographic coordinates of the first image, a position relation of homonymy matching target positions is obtained by extracting SURF features of adjacent image overlapping regions, and therefore the geographic coordinates of the following images are sequentially corrected; and last, an adaptive gradual-in-gradual-out fusion algorithm is adopted, a panoramic image with a good visual effect is obtained, and good stitching of the images is completed. Through the method, an image feature extraction algorithm and the geographic coordinates of the images are combined, and stitching efficiency and the visual effect are both improved greatly compared with traditional feature extraction and stitching algorithms; and the image obtained after stitching contains geographic information, so that the image has certain practical value.

The invention discloses an attention mechanism relationship comparison network model method. An attention relationship comparison network model for small sample learning under a small amount of labeled sample data is constructed. Based on a relational network architecture, the model is divided into a feature coding part, a feature combination part and a relational coding part, the feature coding module is used for extracting image feature information, and the feature combination part is used for recombining the extracted query image feature information with the training image feature information of each group to form a new combined feature map. The relation encoding module performs nonlinear metric learning of the network; by introducing an attention mechanism and a spectrum normalizationmethod into an end-to-end deep convolutional neural network model, the model has higher classification accuracy under the condition of small sample learning, the stability of a final training result of the model is improved, and the image classification accuracy of an existing model in small sample learning is improved.

The invention discloses a shopping behavior prediction method, and the method comprises the steps: selecting different target users at different shopping stages, and obtaining sample data from a user behavior log of the selected target users; respectively extracting a first feature set for marking user behavior from the sample data at each shopping stage; respectively training the first feature sets at different stages through a decision tree model, obtaining a feature combination through multiple iteration, and enabling the feature combination to serve as a second feature set; respectively employing the first and second feature sets at all shopping stages to train a pre-built machine learning model, wherein the machine learning model is used for predicting the shopping demand degree of a user; and determining the shopping stage of a to-be-detected user according to the machine learning model at different shopping stages. The invention also provides a corresponding shopping behavior prediction device.

The invention provides a magnetic resonance image feature extraction and classification method based on deep learning, comprising: S1, taking a magnetic resonance image, and performing pretreatment operation and feature mapping operation on the magnetic resonance image; S2, constructing a multilayer convolutional neural network including an input layer, a plurality of convolutional layers, at least one pooling layer/lower sampling layer and a fully connected layer, wherein the convolutional layers and the pooling layer/lower sampling layer are successively alternatively arranged between the input layer and the fully connected layer, and the convolutional layers are one more than the pooling layer/lower sampling layer; S3, employing the multilayer convolutional neural network constructed in Step 2 to extract features of the magnetic resonance image; and S4, inputting feature vectors outputted in Step 3 into a Softmax classifier, and determining the disease attribute of the magnetic resonance image. The magnetic resonance image feature extraction and classification method can automatically obtain highly distinguishable features/feature combinations based on the nonlinear mapping of the multilayer convolutional neural network, and continuously optimize a network structure to obtain better classification effects.

The invention discloses an advertisement click-through rate prediction method based on multi-dimensional feature combination logical regression. The method comprises the first step that feature information of a hierarchical structure of the user hierarchy, feature information of a hierarchical structure of the media hierarchy and feature information of a hierarchical structure of the advertisement hierarchy are extracted from the obtained click-through rate data respectively; the second step that multi-dimensional combination is carried out on the feature information of the hierarchical structure of the user hierarchy, the feature information of the hierarchical structure of the media hierarchy and the feature information of the hierarchical structure of the advertisement hierarchy, three-to-three combination is carried out on one-dimensional feature information in the feature information to obtain a three-dimensional feature combination, and a feature vector combined by the three-dimensional feature information is formed to represent a user cluster; the third step that the second step is carried out repeatedly and a learning set of the feature vector combined by the three-dimensional feature information is obtained; the fourth step that the learning set obtained in the third step is used for training and testing a logical regression model, and the logical regression model is used for predicting the advertisement click-through rate.

This invention relates to a method and a system for generating a discriminatory signal for a neurological condition, where at least one probe compound that has a neurophysiologic effect is provided. Biosignal data are obtained from a subject based on biosignal measurements obtained from biosignal measuring device adapted for placement on a subject, wherein said biosignal data are obtained posterior to the administering of said probe compound to the subject. Analogous biosignal reference data are provided for reference subjects in at least one reference group posterior to the administering of the probe compound, wherein the reference data are utilized for defining reference features having common characteristics between the reference subjects in the at least one reference group, wherein the reference data are processed for defining reference posterior probability vectors for each respective reference subject, wherein each respective posterior probability vector comprises particular feature or a feature combination elements with probability values associated to said elements, the posterior probability vectors resulting in a distribution of said features or feature combinations for said reference subjects. Subsequently, the biosignal data obtained from the subject are used for calculating analogues posterior probability vector for said subject. The discriminatory signal is then generated based on comparison between said posterior probability vector for said subject and the distribution of said features or feature combinations.

The invention relates to a vehicle type identification method based on a support vector machine and used for an earth inductor. The vehicle type identification method includes the following steps: vehicle type waveform data which require to be identified are collected by the earth inductor; a plurality of numeralization features are extracted from waveforms, effective data are screened out, and the features are normalized; multilayer feature selection is performed according to the extracted features, and an optimal feature combination is picked out; a vehicle type classification algorithm based on the clustering support vector machine is built, and parameters in a classification function are optimized by adopting a particle swarm optimization algorithm; a binary tree classification mode is built, classifiers on all classification nodes are trained, and a complete classification decision tree is built; and earth induction waveforms of a vehicle type to be identified are input to obtain identification results of the vehicle type. The vehicle type identification method builds a waveform feature extraction and selection mode, simultaneously adopts the classification algorithm based on the support vector machine and the particle swarm optimization algorithm, greatly improves machine learning efficiency, and enables a machine to identify vehicle types rapidly and accurately.

The invention discloses a Chinese detection method in a natural scene image based on a connected domain. The method comprises the steps: firstly, obtaining the natural scene image, and carrying out preprocessing of stroke width transformation on the natural scene image, wherein an output of the stroke width transformation is an image, and each pixel value of the image represents a maximum possible stroke width of a pixel of a corresponding position of an original image; demarcating connected domains in the image pixels; extracting various features of a connected component after the connected component is obtained, wherein the feature combinations can well express the connected component; from a Chinese structure, firstly, carrying out within character mergence, then carrying out inter-character mergence, wherein a within character mergence method is used for detecting single Chinese character, a inter-character mergence method is used for detecting text lines, and a text area is demarcated by a rectangular frame. According to the Chinese detection method, a complex structure of Chinese serves as a starting point, pertinence in a Chinese detection aspect is good, and therefore the Chinese detection method in the natural scene image based on the connected domain has high initiative and high accuracy.

The invention relates to a rice disease recognition method based on principal component analysis and a neural network. The method comprises the steps that rice disease image data are acquired and image preprocessing is performed; visual saliency detection is performed, and rice disease images of ideal disease spot outlines are searched from salient map sequences; features are extracted from the rice disease images from the aspects of color, shape and texture, and difference analysis and principal component analysis are performed so that different feature combinations are found; and construction of a machine learning model is performed on different feature combinations and a prediction result is fed back to a client side. The invention also discloses a rice disease recognition system based on principal component analysis and the neural network. Image information is acquired and the images are transmitted to a server side through the network. Preprocessing and disease spot detection are performed on the acquired tissue culturing images through the server side, and management personnel are prompted through a mobile phone short message and a signal lamp and a PC side according to the detection result.

A method for halting malware includes: monitoring plural file system events with a system driver to detect an occurrence of a file system event having a predetermined file type and log event type; triggering a listening engine for file system event stream data of a file associated with the detection of the file system event, the file system event stream data indicating data manipulation associated with the file due to execution of a process; obtaining one or more feature values for each of plural different feature combinations of plural features of the file based on the file system event stream data; inputting one or more feature values into a data analytics model to predict a target label value based on the one or more feature values of the plural different feature combinations and agnostic to the process; and performing a predetermined operation based on the target label value.

A system and method for registering combinations of physical and/or virtual functions for configuring an endpoint are provided. With the system and method, a mechanism informs a management component of a multifunction endpoint's functional combinations. The management component may then use this information to select the functional combinations that are to be made visible to each system image. The informing of the management component may be performed by writing values to various predefined fields in a configuration space for a physical or virtual function that provides information regarding the number of virtual functions supported, the combination of functions supported, a nominal combination of virtual functions to be used by the physical function, and a group identifier for identifying which functions are linked in a combination grouping.

The present invention discloses a sample classification determination method, including the following steps: collecting a sample assembly to form sample databases separately; submitting assemblies in the sample databases to the virtual sandbox environment for operation and then generating corresponding sample analysis reports; analyzing the sample analysis reports, and extracting the unique feature combination information to generate a feature vector set; inputting the feature vector set into classifiers for training to obtain an optimum model; and inputting a program to be tested to the optimum model to obtain a determination result that the program to be tested is a malicious program or a normal program. The present invention improves the efficiency and accuracy of malware detection, avoids the complicated operation and high energy consumption in the dynamic detection technique, and greatly improves the detection speed on the basis of ensuring the accuracy.

The invention provides an image registration method and a device based on depth learning. The image registration model is constructed by utilizing a capsule network. The scalar representation and pooling mechanism in the traditional depth learning convolution network are replaced by a vector feature representation and routing mechanism, and capsules of different levels are connected step by step to carry out feature combination. An image fusion network based on capsule vector is constructed, and the fusion image with the same dimension size as the reference image is output as the registrationoutput. The loss function based on similarity measure between images is constructed to optimize the unsupervised learning registration network by feedback training network parameters. The image registration operation with high accuracy and high robustness is realized.

The present disclosure provides an artificial neural network system for identifying a lesion in a retinal fundus image that comprises a pre-processing module configured to separately pre-process a target retinal fundus image and a reference retinal fundus image taken from a same person; a first neural network (12) configured to generate a first advanced feature set from the target retinal fundus image; a second neural network (22) configured to generate a second advanced feature set from the reference retinal fundus image; a feature combination module (13) configured to combine the first advanced feature set and the second advanced feature set to form a feature combination set; and a third neural network (14) configured to generate, according to the feature combination set, a diagnosis result. By using a target retinal fundus image and a reference retinal fundus image as independent input information, the artificial neural network may simulate a doctor, determining lesions on the target retinal fundus image using other retinal fundus images from the same person as a reference, thereby enhancing the diagnosis accuracy.

The invention discloses an extremum connected domain based character detection method in a natural scene image. The method comprises the steps of acquiring the natural scene image firstly, and extracting a maximally stable extremal region separated from the natural scene image, wherein the output of the separated maximally stable extremal region is a series of non-overlapped regions, and each region is a connected component; after obtaining the connected components, extracting various features of the connected components, wherein the feature combination can well express the connected components; from the character structure, performing intra-character combination firstly, and then performing inter-character combination, wherein the intra-character combination method is for detecting a single character, and the inter-character combination method is for detecting text lines; and finally, analyzing angular point distribution of the text lines, screening to obtain a text region, and demarcating the text region by a rectangular frame. The method starts with character edge features, and has good pertinence in character detection, thereby having higher initiative and accuracy.

The invention discloses a device for classifying dynamic electrocardio data. The device comprises an electrocardio data collecting device, an electrocardio information database, an electrocardio index obtaining device, an electrocardio index screening device, a feature combination obtaining device, a classifier screening device and a classification result output device, wherein the electrocardio index screening device is used for carrying out difference analysis on electrocardio indexes to screen out electrocardio indexes with significance differences, the feature combination obtaining device is used for carrying out feature combination on at least two items in the electrocardio indexes, with the significance differences, screened by the electrocardio index screening device to obtain a plurality of feature combinations, the classifier screening device adopts a plurality of classifiers for testing the feature combinations obtained by the feature combination obtaining device so as to screen out the optimum classifier and the optimum feature combination, the classification result output device is used for receiving the personnel information of patients and pathological data relevant to the heart activity state, classifying the pathological data, relevant to the heart activities, of the patients according to the optimum feature combination and the optimum classifier screened out by the classifier screening device, and outputting the classification results.

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

The invention discloses a method and a device for human face recognition. The method comprises steps: through an already-trained first convolutional neural network, a first feature group for a first human face area in a picture is extracted, wherein the first feature group presents human face features in the picture; through an already-trained second convolutional neural network, a second feature group for a second human face area in the picture is extracted, wherein the second human face area is determined by a second area where the human face in the picture is, and the second feature group presents clothes features in the picture; the first feature group and the second feature group are combined, dimension reduction processing is carried out on the feature combination after the combination, and a third feature group is obtained; and according to the cosine distance between the third feature group and already-extracted reference human face features, whether the human face in the picture and the human face corresponding to the reference human face features are the same human face is determined. According to the technical scheme of the invention, the peripheral clothes and ornaments of the user face area can be combined with the user face features for human face recognition, and the human face recognition accuracy is greatly improved.

The invention provides a multi-target tracking method based on multiple feature combination and a Mean Shift algorithm. The method comprises the following steps S100) a background model is initialized, a background in a video image is updated in a frame difference method, the background is differentiated in a background difference method, and the video image is binarized; S200) denoising and multi-target segmentation are carried out on the binary image sequentially to obtain segmented images with contours of movement targets; and S300) and a multi-target tracking algorithm is carried out on the segmented images on the basis of multiple feature combination and the Mean Shift algorithm. According to the method, features of R, G and B components in a color feature RGB model, H and S components of an HIS model and a gray-scale histogram are tracked comprehensively, the robustness and instantaneity of a target tracking result are improved, and the target sharpness is extracted.

A system and method for improving the efficiency of an object-level instruction stream in a computer processor. Translation logic for generating translated instructions from an object-level instruction stream in a RISC-architected computer processor, and an execution unit which executes the translated instructions, are integrated into the processor. The translation logic combines the functions of a plurality of the object-level instructions into a single translated instruction which can be dispatched to a single execution unit as compared with the untranslated instructions, which would otherwise be serially dispatched to separate execution units. Processor throughput is thereby increased since the number of instructions which can be dispatched per cycle is extended.