249results about How to "Strong discrimination" patented technology

The invention provides an MRI (Magnetic Resonance Imaging) brain tumor localization and intratumoral segmentation method based on a deep cascaded convolution network, which comprises the steps of building a deep cascaded convolution network segmentation model; performing model training and parameter optimization; and carrying out fast localization and intratumoral segmentation on a multi-modal MRIbrain tumor. According to the MRI brain tumor localization and intratumoral segmentation method provided by the invention based on the deep cascaded convolution network, a deep cascaded hybrid neuralnetwork formed by a full convolution neural network and a classified convolution neural network is constructed, the segmentation process is divided into a complete tumor region localization phase andan intratumoral sub-region localization phase, and hierarchical MRI brain tumor fast and accurate localization and intratumoral sub-region segmentation are realized. Firstly, the complete tumor region is localized from an MRI image by adopting a full convolution network method, and then the complete tumor is further divided into an edema region, a non-enhanced tumor region, an enhanced tumor region and a necrosis region by adopting an image classification method, and accurate localization for the multi-modal MRI brain tumor and fast and accurate segmentation for the intratumoral sub-regions are realized.

A receiving antenna circuit is arranged in a reactive near-field of a modulated magnetic induction signal and forms a narrow band-pass filter. The output of the antenna circuit is not subjected to any frequency translation prior to digitising, and the signal may nevertheless be digitised with low resolution, which radically reduces the power consumption of the receiving circuits. The reduction in power consumption is of several orders of magnitude, which allows implementation of such receiving circuits in battery-driven devices such as hearing devices without substantially affecting the battery life.

The invention discloses a track and convolutional neural network feature extraction-based behavior identification method, and mainly solves the problems of computing redundancy and low classification accuracy caused by complex human behavior video contents and sparse features. The method comprises the steps of inputting image video data; down-sampling pixel points in a video frame; deleting uniform region sampling points; extracting a track; extracting convolutional layer features by utilizing a convolutional neural network; extracting track constraint-based convolutional features in combination with the track and the convolutional layer features; extracting stack type local Fisher vector features according to the track constraint-based convolutional features; performing compression transformation on the stack type local Fisher vector features; training a support vector machine model by utilizing final stack type local Fisher vector features; and performing human behavior identification and classification. According to the method, relatively high and stable classification accuracy can be obtained by adopting a method for combining multilevel Fisher vectors with convolutional track feature descriptors; and the method can be widely applied to the fields of man-machine interaction, virtual reality, video monitoring and the like.

The invention discloses a tensor hyperspectral image spectrum-space dimensionality reduction method based on a deep convolutional neural network. The method comprises steps of: in view that it may significantly increase the parameter space of the deep convolutional neural network to directly use high-band tensor data, performing dimensionality reduction on the waveband of a normalized hyperspectral image by introducing a maximum likelihood intrinsic dimensionality estimation algorithm and principal component analysis to obtain a low-band hyperspectral image; converting the low-band hyperspectral image into a tensor low-band hyperspectral image by means of a window field, and keeping the spectrum and space information of each pixel; and performing spectrum-space dimensionality reduction on the tensor low-band hyperspectral image by means of the deep convolutional neural network in order that a characteristic subjected to the dimensionality reduction includes spectrum information and space information. The tensor hyperspectral image spectrum-space dimensionality reduction method may acquire a high overall classification precision and Kappa coefficient by using the spectrum characteristic and space field characteristic of the hyperspectral data.

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 remote sensing image scene classification method based on multi-branch convolutional neural network fusion. The method comprises the following steps: firstly, randomly dividing a scene data set into a training set and a test set in proportion; performing preprocessing and data amplification on the data set; obtaining an object mask graph and an attention graph according to the processed data through an object detection network and an attention network respectively; respectively inputting the original image, the object mask image and the attention image training set into a CNN network for fine adjustment; obtaining three groups of test sets, respectively obtaining optimal classification models, respectively obtaining outputs of Softmax layers through the optimal classification models by taking the three groups of test sets as inputs, and finally obtaining a final prediction result by fusing the outputs of the three groups of Softmax layers through a decision-making level. The classification accuracy and the classification effect can be improved.

The invention discloses a method for realizing classification of scene images. In the method, the scene images are classified by adopting two stages of classifiers, wherein the first-stage classifier obtains candidate classifications by utilizing global structural information characteristics, and judges similar classification pairs according to classification results; and the second-stage classifier distinguishes similar classifications by utilizing local texture information characteristics. By adopting the cascade of the classifiers and comprehensively utilizing the global structural information characteristics and the local texture information characteristics of the scene images, the method can realize robust classification of different scene classifications and effectively distinguish similar scene classifications.

The invention discloses a method for assisting in learning by pedestrian re-identification feature fusion. The method comprises the following steps that: adopting a global feature extraction model obtained by adding local feature training for extracting the global feature of a pedestrian image, and carrying out pedestrian re-identification by the global feature. The training of the global featureextraction model comprises the following steps that: collecting a whole body image training set, detecting a local image in the whole body image training set, and obtaining a local image training set;independently utilizing the whole body image training set and the local image training set to train a whole body convolutional neural network and a local convolutional neural network to obtain a whole body model and a local model; and independently utilizing the whole body model and the local model to extract the global feature and the local feature of the whole body image training set and the local image training set, and utilizing the global feature which carries out local feature fusion to train the whole body model to obtain a global feature extraction model. By use of the method, duringtraining, the local feature and the global feature are subjected to fusion, and pedestrian re-identification accuracy is improved.

The invention discloses an undesirable image detecting method based on connotative theme analysis, which is substantially used for solving the problem of wrong judgment on normal images resulting from semantic information consideration failure in the present undesirable information detecting method. The scheme is as follows: extracting a skin region of an image by a double-blending Gaussian model; generating a codebook base containing distinguishing features in the skin region by a word bag model, and representing each training image to a group of word co-occurrence vectors with weights via aword frequency-inverse identification file frequency method; forming all co-occurrence vectors to a co-occurrence matrix, performing LDA model creation on the co-occurrence matrix to obtain the themeof the image; inputting the mixed theme of the training image in a BP neural network to train an undesirable image classifier; and obtaining the theme of an image to be measured, inputting the theme to the undesirable image classifier, and judging whether the theme is an undesirable image so as to finish the undesirable image detection. As shown in the test, the invention can be used for better distinguish the undesirable images and the normal images, so that the invention can be used for filtering the erotic information in the images.

The invention provides an Alzheimer's disease and mild cognitive impairment identification method based on two-dimension features and three-dimension features. The method particularly comprises a step of performing pretreatment of a medical image, wherein the pretreatment comprises pre-segmentation, registration and other processes; a step of performing two-dimension textural feature extraction of the medical image, wherein features comprise the quadratic statistic of a gray-level co-occurrence matrix and a multiscale and multidirectional feature value of Gabor wavelet transformation; a step of performing three-dimension morphological feature extraction of the medical image, i.e., extracting volume features of an area of interest; a step of performing feature fusion of three-dimension morphological features and two-dimension textural features; and a step of constructing a support vector machine to achieve identification of Alzheimer's disease and mild cognitive impairment. According to the method provided by the invention, the three-dimension morphological features and the two-dimension textural features are combined, so that the content of the medical image can be expressed in a comprehensive and accurate manner. The method can improve identification of Alzheimer's disease and mild cognitive impairment, thereby providing a more effective clinic assistant diagnosis.

The invention discloses a vehicle brand model identification method based on deep learning. The method designs a firm deep convolution network structure by aiming at vehicle brand model identification, adopts the deep convolution network to extract the image characteristics of a vehicle face area under different scales and combines with a support vector machine to classify the image characteristics so as to automatically finish the vehicle brand model identification. The method is high in identification accuracy.

The invention discloses a hyperspectral image classification method based on linear regression Fisher discrimination dictionary learning (LRFDDL). The method includes the following steps that the textural features and the spectral features of each hyperspectral image are fused and normalized; a dictionary and a linear classifier are obtained by learning through an LRFDDL algorithm; model prediction is conducted. According to the hyperspectral image classification method based on LRFDDL, the LRFDDL algorithm is applied to hyperspectral image classification, so that the classification accuracy of the hyperspectral images is improved; in addition, the textural features of the hyperspectral images are added into the classification, and thus the classification effect of the hyperspectral images is further improved.

The invention provides a method and a device for online detection of foreign fiber in cotton. The device comprises a cotton opener, channels, windows, a CCD (Charge Coupled Device) camera, an illumination light source, a background plate, a nozzle array, a foreign fiber channel, a foreign fiber collecting unit, a computer, an electrical equipment cabinet, operating buttons, a frequency conversionfan and a cotton conveyance pipe. The method comprises the following steps: opening lint cotton; feeding the lint cotton into the detection area of the CCD camera through the channels; performing theillumination and CCD camera sampling in three steps, and further collecting the detection results of the three steps for identifying the foreign fiber; calculating the time for nozzles in the corresponding positions to the foreign fiber to open and close by combining the action delay time of each nozzle; controlling the corresponding nozzles to operate, so as to remove the foreign fiber; and meanwhile, outputting clean cotton with the foreign fiber being removed from the clean cotton output channel, wherein, by combining the performance parameters of the CCD camera, the cotton flow rate in the corresponding area of each nozzle can be calculated from the data obtained through infrared LED illumination and sampling, and the average cotton flow rate can be further calculated; and then, the cotton flow rate in the channels can be stabilized within a per-determined range by controlling the frequency conversion fan.

The invention discloses a PoLSAR image classification method based on FCN and sparse-low rank subspace expression integration; the method fully utilizes the powerful ability of the FCN which can automatically learn non-linear deep multiple dimensional space characteristics from PoLsar data, can utilize the advantages that a linear dimension reduction algorithm based on sparse-low rank map embedment can simultaneously capture local and global structure information of the PoLsar data, and can effectively integrate the FCN and sparse-low rank subspace expression, thus solving the PoLSAR image non-linear deep space feature extraction problem, the dimension reduction problem, the valid integration problem between polarized information and space information, and effectively solving the PoLSAR image classification problem. The method can obtain integrated multi-layer subspace characteristic information containing various types; the information contains linear and non-linear types, shallow anddeep types, local and global types, and polarized and multiple dimensioned space types, thus providing strong discrimination ability, and greatly improving PoLSAR image classification accuracy.

The invention discloses an image stitching tampering detection method, relates to the field of image analysis, and is an image stitching tampering detection method based on a mixed domain attention mechanism and a cavity space pyramid pooling module, and the method comprises the steps of extracting a depth feature map F of an input image; obtaining a feature map Ffinal of the tampered area by adopting a mixed domain attention mechanism; adopting a cavity space pyramid pooling module to obtain a final positioning mask M; training a stitching tampering detection method based on the mixed domainattention mechanism and the cavity space pyramid pooling module. According to the measurement of image stitching tampering detection based on the mixed domain attention mechanism and the cavity spacepyramid pooling module, the defects that in the prior art, the tampering area of a stitched image cannot be accurately positioned based on a certain specific hypothesis, and a tampering target with asmall area is easily ignored in detection are overcome.

The invention discloses a cross-age face recognition and model training method and device, a computer readable storage medium and an electronic device. The cross-age face recognition model training method comprises the steps of extracting feature vectors of face images in a cross-age face database through a convolutional neural network, wherein the cross-age face database comprises a plurality offace images classified according to age features and classification features of faces; obtaining a norm and a normalization vector of the feature vector, updating age loss corresponding to the norm based on the age feature of the face image, and updating classification loss corresponding to the normalization vector based on the classification feature of the face image; and training the convolutional neural network based on a joint loss of the age loss and the classification loss. Based on the scheme of the embodiment, the cross-age face recognition performance of the model can be improved.

The invention provides a pedestrian re-identification method based on pedestrian identity and attributive character combined identification verification, which makes full use of complementary information of pedestrian identity and attributive character, and performs multi-task learning on a deep convolutional neural network in two modes of combined identification and verification to obtain more discriminative pedestrian characters. According to the method, pedestrian identity characters and pedestrian attributive characters are learned at the same time, so that a character layer of the neuralnetwork can learn overall identity characters of a pedestrian high layer and can also grab semantic characters of a middle layer, the two characters are effectively fused in the same neural network, and therefore, the method has higher robustness and discrimination. Besides, the deep convolutional neural network is trained in a supervised manner by combining two modes of pedestrian recognition andpedestrian verification so that different types of pedestrian pictures can be distinguished by the learned pedestrian characters, the character distance of the same pedestrian can be enabled to be quite short, and the character distance of different pedestrians can be enabled to be quite long.

The invention discloses a target tracking method based on local feature learning. A target object and background are resolved into a large number of local units having dimension and shape invariance which serve as training samples of a target and background classification model, and a mode of deep learning is adopted to learn local expression of the target object and background from the training samples. The degree of confidence that each specific area in an image belongs to the target object is judged, thereby realizing accurate positioning of the target object. Since local expression obtained by learning of a large number of samples has a high target identification capability, the tracking method has relatively high adaptability to situations such as target deformation and target occlusion. When an object appearance model is updated, local areas with a relatively high target confidence degree are extracted as training samples of the model, and the model is updated. In the process of target tracking, the method continuously updates the appearance model, learns key characteristics of the target object, and can obtain a relatively good tracking effect under the circumstance that an appearance change is relatively big.

The invention provides an image change detection method based on a depth-separable convolution network, which is used for solving the technical problem of low detection accuracy existing in the conventional image change detection methods. The implementation steps include: constructing a training sample set, a verifying sample set and a test sample set; using a variant U-Net of a full convolution network as a basic network to establish a depth-separable convolution network; constructing a loss function of the trained depth-separable convolution network; training, testing and verifying the depth-separable convolution network; and using the verified finally-trained depth-separable convolution network to perform test to obtain a change detection result graph. The invention has rich image feature semantics and structure information extracted by the depth-separable convolution network, has strong image expression ability and discrimination, can improve the change detection accuracy, and canbe used in the technical fields of land cover detection, disaster assessment, video monitoring and the like.

The invention discloses a method for processing multi-mode brain nerve image characteristics, and relates to image preprocessing for extracting image characteristics or features of a recognition figure. The method comprises the following steps of: firstly, performing characteristic selection on multi-mode data by adopting a multi-mode nerve image characteristic selection method with sample weightand low rank constraint to obtain a low-dimensional characteristic matrix, calculating a nuclear matrix of each mode to obtain a low-dimensional characteristic matrix, calculating a nuclear matrix ofeach mode, fusing the nuclear matrices of different modes into a nuclear matrix to select more discriminative biomarker characteristics. New Alzheimer disease sample cases are predicted and classifiedby using a multi-core support vector machine. The defects that, in the prior art, a subject can be injured by using biomarker characteristics, an illness brain area of a patient cannot be found by only using one kind of brain image characteristic data or not fully utilizing the multi-mode brain nerve image characteristics, and the characteristics in the used brain image are free of medical explanation are solved.

The invention discloses a vehicle searching method and device, electronic equipment and a computer readable storage medium. According to the scheme, the method comprises the following steps: acquiringthe image of a vehicle to be searched; extracting features of the image of the vehicle to be searched through a convolutional neural network to obtain a feature vector; separately carrying out attributive feature extraction and extraction of a deep feature between opposite positive and negative samples on the feature vector; merging the attributive feature and the deep feature to obtain a vehiclevisual feature of the image of the vehicle to be searched; and carrying out matching calculation on a vehicle visual feature between the image in a vehicle search database and the image of the vehicle to be searched and acquiring a target image matched with the image of the vehicle to be searched. By the technical scheme, the accuracy of vehicle search is high, and because artificial query is notrequired, the vehicle search efficiency is high.

The invention discloses a vehicle key point detection method based on a gradient regression tree. The method comprises the following steps of acquiring an initial position of each key point of a picture to be detected, inputting the initial position of each key point into a trained classifier and detecting so as to acquire a final position of each key point; extracting a characteristic of each key point in each training sample; using the extracted characteristic to carry out classifier training. The invention also discloses an apparatus corresponding to the method. The method and the apparatus of the invention have an advantage that a gradient-regression-tree point-to-point regression strategy is used so that interferences of surrounding points are reduced.

The invention provides a pedestrian re-recognition method based on general similarity measurement learning. The method comprises the steps that dimensionality reduction is carried out on the feature expression vector of a training sample through principal component analysis; the dimensionally reduced feature expression vector is projected into an intra-class subspace; a general similarity function defined by an inner product similarity measurement matrix and a Mahalanobis distance measurement matrix together is learned in the intra-class subspace; and finally pedestrian re-recognition is carried out on a test sample. According to the invention, the method makes full use of distance information and angle information among training samples; two acquired measurement matrices are mutually complementary; and through verification on a VIPeR dataset, joint learning of the inner product similarity measurement and the Mahalanobis distance measurement can effectively improve the performance of pedestrian re-recognition.

The invention discloses a method for fusing a significant structure and a relevant structure of the characteristics of an image. The method for fusing the significant structure and the relevant structure of the characteristics of the image comprises the steps of extracting the HOG characteristic and the LBP characteristic of the image, measuring the significant structure of the image characteristics inside sample sets, measuring the relevant structure of the image characteristics between the sample sets, and conducting fusion and mapping of the significant structure and the relevant structure. According to the method for fusing the significant structure and the relevant structure of the characteristics of the image, the HOG characteristic and the LBP characteristic of the image are extracted firstly, the significant structure of the image characteristics inside the sample sets is measured through x<2> measurement, the relevant structure of the image characteristics between the sample sets is measured through canonical correlation, and finally the structures are fused through a matrix spectrum optimization solution method, so that a fused characteristic set is obtained. By the adoption of the method for fusing the significant structure and the relevant structure of the characteristics of the image, the problem that in the prior art, the significant structure and the relevant structure of the multiple characteristics of the image can not be fused is solved, and structural fusion characteristics high in discrimination capacity are obtained.

The invention discloses a hyperspectral image classification method based on a parallel attention mechanism residual network. The method comprises the steps that firstly, a residual block is constructed, two parallel attention branch network branches are embedded into the residual block, and the two parallel attention branch network branches conduct recognition learning on spatial feature information and spectral feature information of input data after applying a spectral attention mechanism and a spatial attention mechanism respectively; secondly, training an input training data set by utilizing a hyperspectral image classification network formed by a plurality of constructed residual blocks which are connected in sequence; wherein the hyperspectral image classification network further comprises a 3D average pooling layer and a full connection layer which are connected in sequence, and the 3D average pooling layer is connected to the residual block adjacent to the 3D average pooling layer and used for carrying out spatial dimension adjustment on data output by the current residual block so as to reduce the calculation overhead of the whole network; and finally, inputting the feature information into a full connection layer of the hyperspectral image classification network to obtain an image classification result.

The invention discloses a text emotion analysis method based on a bi-directional interactive neural network, comprising the following steps: collecting entities; Text emotion corpus set, which is divided into training set and test set; Preprocessing the entities and texts in the corpus; Relative position information and adopting global word vector information to construct word and sentence representations. The entity and text word vectors of the training corpus are inputted into the neural network, training and the affective classification model. Inputting The test set entity and text word vector into the neural network model, and calculating the prediction probability of each sample. Using quantum-heuristic multi-modal decision fusion method, the weighted fusion of text prediction probability and image prediction probability is used to obtain more accurate and intelligent multi-modal emotion classification results.

The invention discloses a pedestrian re-identification method based on double constraint metric learning and sample reordering. The method comprises two stages of training and testing; the training stage comprises the following steps: establishing a cross-camera correlation constraint; establishing a same-camera correlation constraint; and solving a metric matrix; the testing stage comprises the following steps: using the metric matrix to perform feature space projection; calculating the Euclidean distance of query pictures and candidate pictures in a feature space; calculating the initial ordering of the candidate pictures; selecting the first K candidate pictures in a ordering queue; constructing a probabilistic hypergraph by using the relevance of the first K candidate pictures in the feature space; calculating a reordering result based on the probabilistic hypergraph; and returning the final ordering of the candidate pictures. The pedestrian re-identification method based on the double constraint metric learning and sample reordering provided by the invention considers two correlation constraints of training samples simultaneously, so that a feature space obtained by learning is more suitable for pedestrian re-identification, and at the same time, the relevance of the candidate pictures is used to reorder, so that a more accurate pedestrian re-identification result is obtained.

The invention relates to a face recognition method based on adaptive weighting and local characteristic fusion. The method comprises the main steps of firstly performing blocking on an image, calculating information entropy of each subblock and obtaining a weighting coefficient of the subblock, then extracting a texture characteristic of the image in a local binary pattern, extracting edge information of the image by means of a histogram of gradient, performing adaptively weighted partial characteristic fusion on a sample image, and finally performing classified recognition by means of a nearest neighbor method. According to the face recognition method, robustness of the local texture characteristic and the edge characteristic of the image to image local shadow and illumination change is utilized; a finally output characteristic has relatively high interference resistance and relatively high discriminating capability; and a good recognition effect is realized on the condition of illumination and gesture change.

The embodiment of the invention provides a face identification method and device based on the Gabor binary mode. The device comprises a threshold value determination module, a filtering processing module, a determination module, an obtaining module and an identification module, wherein identification factors of all first filtering response images in a training image set are obtained by the threshold value determination module with the Fisher criterion, and pixel point threshold values under scales in all directions are determined according to the identification factors; the filtering processing module conducts gamma wave filtering on the images to be processed, and preset second filtering response images under scales in all directions are obtained; the determination module determines LGBP binary system graphs of the second filtering response images according to the pixel point threshold values of the second filtering response images; feature vectors of the images to be processed are obtained by the obtaining module according to the LGBP binary system graphs; the similarity between the images to be processed and trained images is obtained by the identification module according to the feature vectors and the feature vector of any trained image in the trained image set, and identification results are obtained according to the similarity threshold values. By means of the face identification method and device, the capacity for identifying the face can be improved.