151results about How to "Improve distinguishability" patented technology

Improved fluoresced imaging (FI) and other sensor data imaging processes, devices, and systems are provided to enhance display of different secondary types of images and reflected light images together. Reflected light images are converted to a larger color space in a manner that preserves the color information of the reflected light image. FI or secondary images are transformed to a color range within the larger color space, but outside the color area of the reflected light images, allowing the FI or secondary images to be combined with them in a manner with improved distinguishability of color. Hardware designs are provide to enable real-time processing of image streams from medical scopes. The combined images are encoded for an electronic display capable of displaying the larger color space.

The invention discloses a brain tumor segmentation method based on a deep neural network and a multi-modal MRI image. The method includes steps: constructing the deep neural network, wherein the deep convolution neural network includes two three-layer convolution layers, a three-layer full connection, and a classification layer, an input layer corresponds to the multi-modal MRI image, and each node of an output layer corresponds to a tumor classification label; performing MRI image preprocessing; training a network model; and testing the model, performing normalization on a to-be-segmented tumor image sequence by employing image blocks of an MRI image sequence and mean values and standard deviations thereof in a training process, inputting the normalized image sequence to the deep neural network with the optimization network connection weight, obtaining node values of the classification layer, and obtaining the tumor classification of a to-be-segmented brain tumor image. According to the method, tumor abstract topological characteristic information in the multi-modal MRI image is mined and extracted by employing the deep neural network, and high segmentation accuracy and high segmentation precision can be guaranteed in brain tumor segmentation of the multi-modal MRI images.

The invention provides a remote sensing image scene classification method based on multi-scale depth feature fusion and transfer learning in order to solve the classification problem of remote sensingimage scenes. The method comprises the following steps: firstly, obtaining a multi-scale remote sensing image by using a Gaussian pyramid algorithm, inputting the multi-scale remote sensing image into a full convolutional neural network, and extracting multi-scale depth local features; cutting the image to a fixed size required by the CNN; obtaining global features of a full connection layer in an input network, using compact bilinear pooling operation to encode multi-scale depth local features and global features obtained by a CNN, fusing the two depth features to jointly represent a remotesensing image, enhancing the mutual relation between the features, and enabling the obtained features to be more distinctive; and finally, classifying the remote sensing image scenes by utilizing a transfer learning technology and combining the two methods. According to the convolutional neural network, a VGG16-Net is used as a basic network.

The invention relates to a target retrieval method based on a group of randomized visual vocabularies and context semantic information. The target retrieval method includes the following steps of clustering local features of a training image library by an exact Euclidean locality sensitive hash function to obtain a group of dynamically scalable randomized visual vocabularies; selecting an inquired image, bordering an target area with a rectangular frame, extracting SIFT (scale invariant feature transform) features of the inquired image and an image database, and subjecting the SIFT features to S<2>LSH (exact Euclidean locality sensitive hashing) mapping to realize the matching between feature points and the visual vocabularies; utilizing the inquired target area and definition of peripheral vision units to calculate a retrieval score of each visual vocabulary in the inquired image and construct an target model with target context semantic information on the basis of a linguistic model; storing a feature vector of the image library to be an index document, and measuring similarity of a linguistic model of the target and a linguistic model of any image in the image library by introducing a K-L divergence to the index document and obtaining a retrieval result.

The invention discloses a sea-surface ship object detecting and extracting method of an optical remote sensing image, and aims at reducing the false alarm rate effectively, extracting ship objects of different sizes rapidly and accurately, obtaining amount and position information of the objects, and being low in computing complexity. Multi-vision significance is detected on the basis of a frequency-domain model, a hyper complex frequency domain transformation model and a quaternion Fourier transform phase spectral module are fused in a weighted manner to overcome disadvantages of the two models and enhance advantages of the two models, and further sea-surface background interface is inhibited, the integral continuity of detected objects and differentiation performance among the objects are enhanced, and the target area of the sea surface is searched effectively. False alarm against possible heavy cloud layers and islands in the images is reduced, an improved histogram in the gradient direction is used to represent the distribution feature of the gradient structure of the object, the detected objects are discriminated according to established rules and conditions, whether a detected object is a ship is determined, the false alarm rate is reduced greatly, and the detecting accuracy is improved.

The invention discloses a face recognition method based on distinguishability feature fusion, comprising the following steps: A, cutting a global image and at least two local images in each training sample image; B, carrying out model training on each intercepted image by adopting a multi-loss function to obtain a corresponding model, wherein the multi-loss function is obtained by combining the a-softmax loss function with the center loss function. C, carrying out fusion and dimension reduction of each model obtain by training by using a ternary loss triplet loss function, and obtaining the final depth feature of the training sample image. The technical proposal disclosed in the application can solve the problem of data fusion, face posture and model fusion in the process of face recognition by using CNN, and achieve better face recognition effect.

One embodiment of the present invention provides a system that detects malicious data in an ad-hoc network. During operation, the system receives data at a node in the ad-hoc network, wherein the data was sensed and redundantly communicated to the node by other nodes in the ad-hoc network. Note that in this ad-hoc network, a given node senses data associated with itself and with proximate nodes in the ad-hoc network. In this way, proximate nodes in the ad-hoc network can redundantly sense data about each other. Next, the system determines at the node if the received data, along with data sensed locally by the node, is consistent. If not, the system uses a model which accounts for malicious nodes to determine an explanation for the inconsistency.

Golf ball 2 has spherical main body 4, mark layer 6 and paint layer 8. The main body 4 has spherical core 10, and cover 12 provided to cover this core 10. The cover 12 does not include titanium oxide. The cover 12 includes a fluorescent colorant. The main body 4 has a chroma saturation of equal to or greater than 25. The paint layer 8 is constituted with a resin composition. This resin composition includes composite particles. The composite particle has a nucleus, and a coat layer provided to coat this nucleus. The nucleus is constituted with mica, while the coat layer is constituted with titanium oxide. The amount of the composite particles is 2 parts by weight or greater and 30 parts by weight or less per 100 parts by weight of the base resin of the paint layer 8.

The invention discloses a whisper feature extraction method which is characterized by comprising the following steps of (1) whisper speech auditory spectrum feature representation; (2) feature dimension reduction and robust performance analysis; the feature dimension reduction and robust performance analysis include three contents: (a) the low-dimensional topological structure features are extracted from the high-dimensional auditory spectrum; (b) topological structure feature time sequence analysis is performed; and (c) topological structure feature stability analysis is performed; and (3) acoustic model optimization is performed; the training mechanism of performing passive learning and then active learning is adopted in acoustic model optimization so as to enhance the adaptive performance of the model. According to the whisper speech feature extraction method, dimension reduction is performed on the auditory perception spectrum features to obtain the topological structure features, and the distinguishability of the whisper speech features is strengthened by strengthening the time sequence weight of the features. The distance between the feature vectors of different meanings is maximized through two target functions, and the distance between the feature vectors of the same meaning is also minimized so as to enhance the robustness of the features.

The invention discloses a pedestrian detection method based on layered kernel sparse representation. The method is characterized by comprising steps as follows: preprocessing acquired traffic videos to obtain required positive and negative samples, acquiring multi-scale eigenvectors through layered sub-block division, and constructing two kinds of dictionary matrixes; preprocessing to-be-detected pedestrian images to obtain test samples, and performing pedestrian feature extraction with a method adopted in the dictionary construction process to form eigenvectors of the test samples; performing kernel sparse decomposition on the eigenvectors of the test samples with histogram cross kernel functions, adopting Gaussian functions for weighting in an iteration solution process, and reconstructing errors to realize pedestrian detection. With the adoption of the method, the higher detection performance can be realized, the pedestrian detection accuracy under the condition of partial shielding is effectively improved, and the robustness of a pedestrian detection system for appearance changes is enhanced.

The invention discloses a multi-soft-constraint stereo matching method based on a cost matrix. The method comprises the following steps: calculating a three-dimensional matching cost matrix at first, then carrying out multiple-dimension downsampling, and forming a cost matrix pyramid; carrying out the corresponding multiple-dimension downsampling of an image, forming an image pyramid, and carrying out segmentation of all image layers; carrying out cost accumulation of self-adaption weight layer by layer and cost accumulation under the voting-type segmentation constraint secondly, and achieving the cost accumulation under the multiple-dimension constraint through transmitting the cost accumulation results of the upper layer to the lower cost matrix; and achieving stereo matching through combining a reliable point cost diffusion method in a bottom layer cost matrix. The method improves the stability and reliability of matching results, solves the matching problems of weak texture and repeated texture regions and parallax error discontinuous places, and can be used for improving modeling, related to the engineering application of stereo matching, based on images.

The invention discloses a high-voltage alternating-current circuit breaker arc burning time measuring system and method. A measuring collection system composed of a voltage divider, a current transformer, a transient waveform recorder and the like is utilized, post optimization processing is carried out on the waveform data through data analysis software, the error interference level caused by a ratio error, a zero drift and electronic digital noise is eliminated or substantially reduced, and the distinguishability of an arc voltage component of only tens of volts is substantially improved; in addition, the step characteristic of the arc voltage component before and after arc burning is utilized, starting and ending moments of the jumping of the arc voltage component are calibrated on a voltage waveform, and the purpose of precisely measuring the arc burning time is realized.

The invention discloses a video sequence registering method based on combination of motion information and background information. The method mainly solves the problem that in the prior art, video sequences in different view points cannot be accurately registered. The method includes the steps that (1) separation of background images and motion targets is carried out on two input video sequences; (2) feature point matching pairs of the background images are obtained, and basis matrixes between the background images are calculated; (3) four sets of matching point pairs corresponding to four spatial points not on the same plane in a three-dimensional space are selected from the background images; (4) intersection points of projection lines, in another video sequence, of points in motion tracks and bipolar lines are obtained; (5) motion track points are matched and a point pair set corresponding to candidate time is obtained; (6) a time line is fitted and time transformation parameters are recovered. The time relation between video sequences can be accurately recovered, registering accuracy is improved and the method can be used for registering video sequences with static backgrounds.

The invention discloses an insulator category detection method based on deep transfer learning. The insulator category detection method comprises the following steps: 1, preprocessing an aerial insulator image; 2, expanding the preprocessed aerial insulator images and classifying different types of aerial insulator images; 3, utilizing a YOLO algorithm to perform initial positioning on an aerial insulator image with a complex background, and performing normalization processing on the positioned insulator; 4, constructing an Inception deep learning network of a multi-level difference adaptive architecture; 5, constructing a classification result and a semantic error entropy of the test sample set; and 6, constructing an insulator state cognitive feedback adjustment mechanism based on semantic error entropy. According to the invention, through a deep transfer learning method, self-optimization adjustment and reconstruction of the insulator state multi-level differential feature space andthe classification criteria thereof can be realized, so that the self-explosion state detection rate of the aerial image of the insulator under different backgrounds is improved, and the actual demand of accuracy and rapidity is met.

The conference device (EMCU) according to the invention has several monaural HRTF filters (HRTF1, . . . , HRTFN), each of which is to be allocated to a conference participant. The abbreviation HRTF stands for “Head Related Transfer Function”. A corresponding HRTF filter (HRTF1, . . . , HRTFN) is used for filtering a monaural audio signal coming from the conference participant to whom it is allocated. An individual, monaural HRTF filter coefficient set is responsible in this case for defining the filter characteristics of a particular HRTF filter. The conference device (EMCU) also has a conference mixing device (MP) coupled to the HRTF filters (HRTF1, . . . , HRTFN), for mixing the individually filtered audio signals from various conference participants and for transferring the mixed audio signals to conference participants.

The invention discloses a sequence deeply convinced network-based pedestrian identifying method. The method comprises the following steps of preprocessing a training image in a pedestrian database to obtain a training sample image, extracting an HOG (Histograms of Oriented Gradients) feature from the obtained training sample image, building and training a sequence restricted Boltzmann machine-based sequence deeply convinced network, using the sequence deeply convinced network to further extract features from the obtained HOG feature to form a feature vector of the training sample, inputting the obtained feature data into a support vector machine classifier, and finishing training; preprocessing a to-be-tested pedestrian image to obtain a test sample; using an HOG and the sequence deeply convinced network to extract pedestrian features from the test sample to form a feature vector of the test sample; inputting the feature vector of the test sample into the support vector machine classifier, and identifying whether the test image is a pedestrian or not. According to the method, better classification performance can be obtained, the accuracy of pedestrian identification is improved, and the robustness of a pedestrian identifying algorithm is enhanced.

The invention discloses a hot topic detection method based on weighted LDA and improved Single-Pass clustering algorithm. The hot topic detection method of the Pass clustering algorithm comprises thefollowing steps: preprocessing text data, including Chinese word segmentation, stop word removal and feature word weighting; modeling the text data by utilizing a weighted LDA topic model, realizing feature dimension reduction by mining hidden topic information in the text data, and filtering and denoising a vectorized result; subjecting text vectorization result processed by LDA topic model weighted by feature words to improved Single-Pass clustering algorithm to carry out clustering; and calculating a hot value of the topic cluster by utilizing the topic cluster scale and the topic cluster compactness, and identifying the hot topic. The detection method has the advantages of being low in algorithm complexity, low in dependency on text input time sequence and the like.

Disclosed is a manufacturing method for a fine powder exhibiting improved solubility, little impurity contamination, and a high recovery rate. Material to be ground and a grinding medium are suspended and stirred in a liquefied inert gas dispersion medium such as dried ice, and the material to be ground is made into a sub-micron or nano-sized fine powder. A uniform fine powder can be obtained when the material to be ground is a mixture having two or more components. Impurity contamination can be reduced by using granular dry ice as the grinding medium.

The invention discloses a stochastic gradient Bayesian SAR image segmentation method based on a sketch structure, mainly used for solving the problem that SAR image segmentation in the prior art is inaccurate. The stochastic gradient Bayesian SAR image segmentation method comprises the following implementation steps of: (1), sketching an SAR image to obtain a sketch image of the SAR image; (2), according to an area chart of the SAR image, and dividing a pixel subspace of the SAR image; (3), performing hybrid aggregation structured surface feature pixel subspace segmentation through a method based on a stochastic gradient variational Bayesian network model; (4), performing independent target segmentation based on the sketch line aggregation feature; (5), performing line target segmentation based on a visual semantic rule; (6), performing segmentation of a pixel subspace in a homogeneous area by adopting a polynomial-based logistic regression prior model; and (7), combining segmentation results to obtain a segmentation result of the SAR image. By means of the stochastic gradient Bayesian SAR image segmentation method based on the sketch structure disclosed by the invention, the good segmentation effect of the SAR image is obtained; and the stochastic gradient Bayesian SAR image segmentation method can be used for semantic segmentation of the SAR image.

The invention discloses an anti-counterfeit paper. The anti-counterfeit paper comprises a paper body and an anti-counterfeit material loaded on the paper body, the anti-counterfeit material comprises a fiber, and at least two photochromic particles are distributed on the fiber along the length direction of the fiber at intervals. The anti-counterfeit material in the anti-counterfeit paper is different from common anti-counterfeit fiber materials in structures, can be easily distinguished, is difficult to imitate, and has very good recognizability only through a single anti-counterfeit ink.

The invention discloses a multi-source domain adaptive model and method based on partial feature alignment, and the method comprises the steps: enabling a feature selection module for partial featureextraction to generate a selection vector of a feature level according to the similarity of each feature dimension of a source domain and a target domain on the basis of a conventional convolutional neural network or residual neural network feature extractor, wherein after the selection vector acts on the initial feature map, part of features highly related to the target domain in the source domain can be screened out. On the basis, the invention further provides three partial feature alignment loss functions for intra-class, inter-domain and inter-class features, so that the purified featuremap has better distinguishability for a classifier, and partial features related to a source domain and a target domain are highlighted. The method is used for the multi-source field self-adaptive classification data set, and compared with an existing multi-source field self-adaptive model, the classification accuracy is higher, and the feature selection effect is better.

The invention relates to a folded bill identification method and device. The device comprises a bill input port, a signal collection module, a signal identification module and a receiving / rejecting module, wherein the signal identification module further comprises a first high-pass filtering unit, a first low-pass filtering unit, a second high-pass filtering unit, a second low-pass filtering unit, a difference filtering picture unit, a first feature extraction unit, a second feature extraction unit, a third feature extraction unit and an identification decision unit. According to the folded bill identification method and device, features are classified effectively in a high / low pass filtering manner, so that the distinguishability of the features are improved; especially, different types of features correspond to different classifiers; furthermore, the classifiers have similar functions with an Adaboost classifier and can ensure the identification confidence level, so that an identification system is relatively robustly compatible with complex situations, such as external environment interference and bill body pollution; the whole folded bill solving method and device can effectively identify folded bills.

The invention discloses a scrambling face recognition method based on a variable cascade information bottleneck. The method comprises the following steps: 1) preprocessing a training data set; 2) defining and initializing a model; 3) forward propagation of the data in batches; 4) calculating a loss function; 5) calculating a gradient through back propagation, and updating the weight; 6) performingiterative training, and repeatedly executing the steps 3) to 5) until the whole model converges; And 7) testing a model result, inputting the face image in the test set, and obtaining a final classification result through the whole model. According to the method, training is carried out under three data sets of ORL, CMUPIE and PUBFIG, testing is carried out on a test set, results are displayed under different scrambling operations, and the method can achieve the highest recognition accuracy.

The invention discloses an anti-robustness image feature extraction method based on variational spherical projection. The method comprises the following steps: 1) initializing a model; 2) preprocessing a data set; 3) performing forward broadcasting on the variational spherical projection; 4) calculating the loss function; 5) performing the adversarial training regularization; 6) calculating the gradient in the reverse direction, and updating the weight value; 7) repeating the step 2) to the step 6) until convergence is carried out to obtain a depth feature extraction model, wherein a high distinguishable characteristic can be obtained by taking the mean parameter of the parameter coding process as the characteristic during application. According to the method, training is carried out on aCASIA-webface data set, and a test is carried out on an LFW data set, the anti-robustness of the model can be guaranteed, and meanwhile, the feature has high separability.

The invention discloses a remote sensing image sea surface ship detection method based on visual attention mechanism and information entropy, including the steps of collecting an optical remote sensing image, carrying out the wavelet decomposition, reconstructing a feature map, calculating a saliency map, counting thesaliency degree of corresponding position pixels in all feature maps based on themultivariable Gaussian probability density function, calculating a global saliency map of the inputted images, reserving the ship target area, carrying out the threshold segmentation and the slice extraction which are used for extracting the slices of the suspected ship target area; obtaining the discriminant entropy which is used to improve the deficiency of traditional entropy definition which depends on the spatial structure information of the image and to obtain the discriminant entropy which can better describe the content of the target slice; identifying the ship targets and false alarms: using the improved identification entropy to characterize the slice features of ship targets, thereby effectively distinguishing the ship targets and the false alarms. The method of the invention can detect and confirm the ship target on the sea surface of the optical remote sensing image from coarse to fine.

A rubber molded article which has a good visibility, prevent medical errors effectively by easily and certainly identifying injector or other containers containing medicinal agents which are used together with colored rubber molded articles, and prevent problems caused by dispersion of coloring agents into medicinal agents without harming the inherent properties of a rubber molded article like sealing performance.A colored film layer 4 is positioned in the interior portion and a transparent resin portion 5 extends from at least one surface of the colored film layer 4 to the outermost layer.

The embodiment of the invention discloses a high-resolution remote sensing image weak and small target detection method and device based on deep learning. The method comprises the steps of obtaining ato-be-processed remote sensing image; inputting the remote sensing image to be processed into a pre-trained convolutional neural network, carrying out 4-time downsampling, 8-time downsampling and 16-time downsampling respectively on the remote sensing image to be processed through the convolutional neural network; obtaining the priori boxes of different sizes corresponding to the to-be-processedremote sensing image, identifying the target priori boxes of which the target category confidence is greater than a preset threshold, and determining the coordinate information of a target included inthe to-be-processed remote sensing image through a preset clustering algorithm according to the coordinate information of each target priori box, wherein the first layer of the convolutional neural network comprises a residual component, the second layer, the third layer and the fourth layer of the convolutional neural network each comprise four residual components, and each residual component comprises two convolutional layers and a fast link. By applying the scheme provided by the embodiment of the invention, the weak and small target detection precision can be improved.

The invention provides a supervised linear dimensionality reduction method with a separation probability of a minimax probability machine and belongs to the technical field of computer machine learning and statistical learning. The method comprises the steps that: a supervised linear dimensionality reduction model with the separation probability of the minimax probability machine is established, an input of the model is a sample set with multiple dimensions and categories, and an output is a projection matrix, when the dimensions are reduced to 1 dimension, an object belongs to a single projection vector object, when the dimensions are reduced to multiple dimensions, objects belong to multiple projection vector objects. According to the method, a separation probability between samples is used as a distance measurement between the categories, a conjugate gradient method is used for optimization, and finally, each category pair has a projection matrix of a maximum separation probabilityas far as possible. According to the method, the distinguishability of the data and the accuracy and efficiency of the subsequent classification can be improved, and a good application effect can be achieved in the problems of multiple types of dimension reduction.

The invention discloses an automatic merchandise classifying method on the basis of binary word segmentation and a support vector machine. The method mainly includes: subjecting all merchandise titles in a training set to binary word segmentation processing to construct a feature word library; constructing merchandise classification sets, expressing the merchandise titles as specific vectors according to the feature word library, generating training data by the aid of the specific vectors and the merchandise classification sets, and performing parameter optimization on the training data by a sequential dual method to obtain optimal classification vectors; calculating inner products of the optimal classification vectors and the specific vectors expressed by titles of merchandises to be classified, and selecting the classification corresponding to the maximum inner product as classification which the merchandises belong to. The automatic merchandise classifying method solves the problems that a product feature information base is hard to construct, and an automatic merchandise classifying method is long in training time and unsatisfactory in effect due to a feature space construction in the prior art.