53 results about "Spatial classification" patented technology

The human neural response is coupled with computer pattern analysis for robust single-event detection of significant non-stationary brain responses triggered upon occurrence of a task-relevant stimulus. Classifier performance is enhanced fusing together the outputs of multiple different classifiers albeit multiple spatial classifiers to extract a temporal pattern as the brain response evolves, time and frequency-based spatio-temporal classifiers, and/or EEG and pupillary classifiers.

The invention discloses a human motion recognition method based on local features. The human motion recognition method includes the first step of carrying out human body detection and regional division, the second step of extracting the features of space-time interest points and corresponding position information from a motion video sequence, the third step of carrying out spatial classification on the space-time interest points according to regional division results, the fourth step of respectively obtaining the space-time interest points included in each specific human body area in a training set and the space-time interest points included in each specific human body area in a test set, the fifth step of clustering the space-time interest points included in each specific human body area in the training set through a clustering algorithm to obtain a corresponding lexicon, the sixth step of independently processing each specific human body area in the training set and each specific human body area in the test set respectively through a word bag model and respectively extracting word bag features of the specific human body areas in the training set and the test set, and the seventh step of using a classifier to carry out modeling on human motion to achieve the purpose of motion recognition. According to the human motion recognition method, the outstanding human body local features are utilized to improve accuracy rate of human motion recognition.

The invention discloses an electrocardiosignal detection device and analysis method based on a joint neural network. The method comprises the following steps: firstly, building a joint neural networkalgorithm on a machine learning server, and training a model; aiming at preprocessed ECG data, enabling the model to extract data spatial features and acquire a spatial classification probability through a residual neural network module; extracting time sequence features of the data on a dimensionality-reduced spatial feature map through a bidirectional long-short-term memory neural network and anattention module, and acquiring a time sequence classification probability; finally, fusing the two classification probabilities to obtain a detection result; acquiring a small amount of ECG data ofa patient from a wearable device, performing manual marking, inputting the ECG data into the machine learning server, performing fine-tuning on the model, and deploying the final model to an intelligent mobile device; and finally, realizing real-time anomaly detection through wireless transmission of the wearable device and the intelligent mobile device. The invention develops the wearable devicefor electrocardiosignal acquisition and the real-time detection, and provides an effective technical means for auxiliary diagnosis of heart diseases.

A graphics system that implements a binning database with an accumulation buffer to perform super sampling. In one embodiment, an application of a host machine passes a geometry of a full scene to a binning database which stores the scene in spatially sorted bins. The contents of the bin are passed to rendering hardware, and rendered multiple times, each time with a stochastic offset applied to the sample points. The results are accumulated in an accumulation buffer and prepared for display.

The invention discloses a method for digital restoration of an urban spatial pattern, which belongs to the field of the restoration of urban spaces. The method comprises the following steps: registration and scanning of a historically old map: registering and correcting through making full use of ArcGIS software and remote sensing data of a city to be restored, and registering the historically old map into the present remote sensing data; restoration of a historical urban space; correction of errors in restoration results of the historical urban space: correcting the errors in the restoration results with the spatial analysis method of the ArcGIS and the mathematical statistics, thereby realizing the digital restoration of the urban spatial pattern. With the adoption of the 3S technology, the invention changes the conventional method for spatial pattern restoration simply depending on paper drawings and verification of historical documents, effectively improves the drawing efficiency and quality of the historical urban map, forms 'a map' of urban spatial patterns in different historic periods based on the unified spatial scale, the unified spatial reference and the unified spatial classification, and guarantees the restoration, management, presentation and application of the urban spatial patterns.

A method for determining insides and outsides of boundaries includes an external data input step of inputting external data constituted by boundary data of objects, a cell division step of dividing the external data into rectangular parallelepiped cells having boundary planes orthogonal to each other, a cell classification step of classifying the cells into a boundary cell that includes the boundary data and a non-boundary cell that does not include the boundary data, and a space classification step of classifying the non-boundary cells into a plurality of spaces that are partitioned by the boundary cells.

The invention discloses a high-resolution accurate two-dimensional direction-of-arrival estimation method based on planar co-prime array virtual domain tensor spatial spectrum search, and mainly solves the problems of signal multi-dimensional information loss and spatial spectrum resolution and accuracy limitation in the existing method. The method comprises the following implementation steps: constructing a planar co-prime array; modeling a receiving signal tensor of the planar co-prime array; deriving a virtual domain equivalent signal based on a planar co-prime array second-order mutual correlation tensor; constructing an equivalent received signal of a virtual domain uniform area array; deriving a fourth-order autocorrelation tensor of the virtual domain smooth signal; realizing signaland noise subspace classification based on multi-dimensional feature extraction of a virtual domain autocorrelation tensor; and high-resolution accurate two-dimensional direction-of-arrival estimation based on virtual domain tensor spatial spectrum search. According to the method, high-resolution accurate two-dimensional direction-of-arrival estimation based on tensor spatial spectrum search is realized on the basis of multi-dimensional feature extraction of virtual domain tensor statistics of the planar co-prime array, and the method can be used for passive detection and target positioning.

There is provided a hologram recording and reproducing apparatus for a hologram recording medium that stores optical interference fringes therein as a diffraction grating generated by coherent reference light and signal light. The hologram recording and reproducing apparatus includes a control circuit that is connected to a spatial light modulator and controls each pixel in such a way that the reference light is modulated according to information data to produce the signal light. The control circuit spatially classifies a plurality of pixels in the spatial light modulator into a central modulation area disposed on the optical axis and at least one annular modulation area sequentially disposed around the central modulation area in a concentric manner, and controls the pixels in the central modulation area and the pixels in the annular modulation area using respective different recording modulation methods to deliver the signal light through the central modulation area and the annular modulation area.

The invention discloses an indoor scene monocular vision space recognition method in a terrestrial gravity field environment. The method is characterized by comprising the following steps that first, ultra-pixel image segmentation based on pixel colors and spatial positions is carried out on an image; second, further clustering is carried out on the ultra-pixel image through a color space spectral clustering method based on human vision multi-scale perception characteristics, spectral clustering based on profiles and forms is carried out on color classification image blocks, initial space classification of an indoor scene is carried out through an outdoor gravity field blurred vision distribution density function, the image blocks with strong facade profile characteristics are classified into facades through the Manhattan strength characteristic, the boundaries between a ceiling and the facades and boundaries between the ground and the facades are searched for on the basis of the indoor scene perspective principle, indoor scene image space recognition is carried out through the indoor gravity field blurred vision distribution density function, and indoor space is marked on the basis of the ground and ceiling perspective principle to generate a depth graph. The indoor scene monocular vision space recognition method in the terrestrial gravity field environment is high in practicability and robustness.

The invention relates to a skull sex determining method based on spatial analysis. The method comprises the following steps: carrying out data standardization on skull data of a training sample; carrying out sub-spatial analysis on a standardized training sample data set so as to obtain a sub-spatial projection matrix; projecting the training sample to a low dimensional sub-space from a high dimensional original feature space, wherein a classifier is designed in the low dimensional sub-space; projecting standardized unknown skull data to the low dimensional sub-space by using a sub-spatial projection matrix obtained from the training; and classifying in the low dimensional sub-space by using the designed classifier. With the adoption of the method, complex manual measurement or expert knowledge is not needed, and the accuracy can be more than 94%, so that the method has important application values in criminal investigation fields.

The present invention provides a method for image deblurring based on correctly specifying the lack of smoothness in the unknown desired sharp image, in terms of Lipschitz (Besov) space classification. The method makes essential use of Singular Integrals and Fast Fourier Transform (FFT) algorithms to estimate the image's Lipschitz exponent a. Such Singular Integrals and Fast Fourier Transforms are then used in a method for regularizing the ill-posed deblurring problem, resulting in a fast, direct, (i.e., non-iterative) computationally effective deblurring procedure.

The invention relates to the field of hyperspectral image classification, and particularly discloses an enhancement type sparse representation hyperspectral image classifying device and method based on space information constraint. The method comprises the steps that the similarity between a reconstruction pixel and an original pixel are calculated through a normalized Euclidean distance, category feature information is introduced when pixel categories are calculated, space information of eight neighborhoods around the pixel to be measured is extracted, the space information weight is set, and the space classification weight is combined with the spectrum similarity on which the normalized Euclidean distance calculation is carried out to form a classification model based on space information and spectrum information; repeated planning is carried out on pixel categories to be measured, and a residual value and the space information weight are calculated again. The method can effectively improve hyperspectral image classifying precision and has the strong generalization ability.

The present invention provides a classified extraction method of tobacco field remote sensing data. Based on the monthly data set of a land resource satellite, the sequence data of a normalized difference vegetation index (NDVI), a normalized difference water index (NDWI) and a modified soil adjusted vegetation index (MSAVI) is obtained through band operation, thus the mathematical method operations of a mean square deviation, a half square deviation and a mean value are carried out, an indicator factor with a clear physical meaning and phenological information is obtained, a multi-dimensional feature space classification data set is constructed, and a support vector machine (SVM) is operated to automatically extract the farmland type of a paddy field and a dry land in a research area. The method has a certain practical significance for the rational use of the agricultural resources of the area, and a certain reference is provided for making the policies and plans of future agricultural development in the area by related administrative departments.

The invention provides an adaptive recognition method for a wide area power system load model. The method comprises the steps of: calculation of a load time-variable classification threshold, reset start judgment of a load model, judgment on validity of the load model based on a system dominant oscillation mode recognition observation object, spatial classification of wide area power system loads and unified recognition of various loads of the system. The adaptive recognition method continuously monitors the load feature change, and if a load has great feature change, the load classes are uniformly recognized according to the proportional feature formed by the electrical property, and the load model of the simulation system is reset with the recognition result, so that the accuracy and the practical engineering value of the load model are improved, and then the security and the economical efficiency of the system are improved.

The invention relates to an adaptive spatial interpolation method, which belongs to the technical field of soil digital mapping. In order to overcome the difficulty of unreliability of spatial interpolation of soil property in the digital soil mapping, the invention provides an interpolation method, which comprises the following steps of: in a geographic space, separating a soil space from a landscape element in a vertical direction, wherein the bottom layer is a soil layer and the upper layer is a landscape element characteristic vector layer; gridding the soil layer according to a certain scale, extracting landscape element characteristic vectors corresponding to any grid units in the grid soil layer, and constructing a landscape element characteristic vector set; classifying the soil space according to the landscape element characteristic vector set so as to obtain a soil space classification map; and establishing an interpolation operator on the basis of the soil space classification map, and interpolating any unknown units in the soil space according to the interpolation operator. By using the technical scheme, the accuracy and the reliability of the space interpolation of the soil property content under a complicated landscape environment are improved and ensured.

The invention discloses a human face data sortability feature extraction method based on weighted maximum spacing criterion. The problems of poor generalization ability of features extracted by MMC and affected classification accuracy, which are caused by the fact that the existing maximum spacing criterion MMC feature extraction uses limited training samples and cannot accurately estimate true distribution of ultra-high-dimensional space samples, are solved. The method comprises the steps that 1) the inter-class distribution matrix Sb and the intra-class distribution matrix Sw of original data are calculated; 2) inter-class and intra-class distribution matrixes are weighted to acquire the weighted maximum spacing criterion WMMC function; 3) the WMMC criterion function is maximized to acquire a mapping matrix; and 4) the original data are mapped to a WMMC subspace; and 5) classifying is carried out in the WMMC subspace. According to the invention, the method can extract the sortability feature with strong generalization ability under the condition of the ultra-high-dimensional small samples; the human face recognition rate is improved; and the method can be used for the sortability feature extraction of the ultra-high-dimensional small sample data.

An implicit function field of a nonmanifold is held in a form of volume data; a value of an implicit function at a point between lattice points is decided by interpolation; and if a difference in code distances between two adjacent voxels to be interpolated is larger than a fixed width, no surface is formed between the voxels. Furthermore, an entered curved surface is broken down into curved surface patches which enable determination of a front and a back; numbers are given to the front and the back, respectively, to be distinguished from each other; and a space is classified into a plurality of regions by using the number of a surface of a nearest point.

The invention discloses a method for intelligent recommendation of a 3D picture based on deep learning and transfer learning. At first, a public big sample scene picture dataset is used to obtain a universal scene image classifier based on the deep learning; then, a 3D picture scene picture dataset collected by a user is used for the transfer learning, and the universal image classifier is converted into a 3D picture scene space classifier; then, a Hash perception algorithm is used to establish an information fingerprint base of a 3D picture design scheme recommendation picture base; and finally, a scene picture shot by the user is matched with the 3D picture design scheme base for screening, a matched candidate subset is obtained, an information fingerprint Hamming distance between each picture in the subset and the user's picture is computed, and the 3D picture with the minimum distance is recommended to the user. According to the invention, based on the deep learning and transfer learning, 3D picture design is achieved in a specific environment and a specific space structure; and design duration of the 3D picture is also shortened.

Model compression is combined with model compensation. Model compression is needed in embedded ASR to reduce the size and the computational complexity of compressed models. Model-compensation is used to adapt in real-time to changing noise environments. The present invention allows for the design of smaller ASR engines (memory consumption reduced to up to one-sixth) with reduced impact on recognition accuracy and/or robustness to noises.

The invention discloses an automatic image annotation system oriented to unknown pre-training annotation data, and the system comprises an acquisition module, an image annotation module and a model migration module, wherein the model migration module is connected with the acquisition module and the image annotation module respectively; the acquisition module is used for acquiring a group of image annotation tasks and images to be annotated; the model migration module is used for migrating a group of determined pre-trained image processing models to an updated image processing model adaptive to a to-be-labeled image domain in an unsupervised manner, and comprises a model splitting unit, an information maximization loss constraint unit, a clustering unit, a label space classification unit, a label distribution unit, a distribution updating unit and a convergence unit which are connected in sequence. And the image annotation module is used for generating annotation information matched with the to-be-annotated image and performing visual annotation.

Embodiments of the invention disclose an information processing method and device, and a computer readable storage medium. The method comprises the steps: obtaining target to-be-processed text information; extracting corresponding text vector information from the target to-be-processed text information through the trained neural network model, wherein the trained model is obtained by performing spatial classification training on to-be-trained text vector information formed by the to-be-trained text information with the label according to the label type; clustering the text vector information according to the spatial distance, and determining text information of different clusters. Therefore, larger text vector information can be extracted, classified and defined from the target to-be-processed text information through the trained model for spatial classification training according to the label type; according to the method, clustering is carried out according to the text vector information and the spatial distance, the clustering effect of the obtained text information of different clusters is more accurate, the labor cost is saved, and the information processing efficiency is greatly improved.

The invention discloses a city identification and extraction method of full-polarimetric SAR image based on clustering algorithm, comprising the following steps: 1) preprocessing four single-channel images of full-polarimetric SAR system; 2) perform discrete wavelet data fusion on four single channel images; 3) use density clustering algorithm to extract urban area; 4) merging different urban areaclusters into the same cluster to obtain the merged urban area image; 5) carry out error processing on that combine town area image to obtain the final town area image, The invention provides a fastand low-cost urban area extraction method based on remote sensing technology, which provides a theoretical basis for understanding the urban spatial classification, the utilization of resources such as land and water, and the influence of human activities, thereby providing a theoretical basis for more rational utilization of urban resources.

The invention discloses an urban three-student space identification and pattern analysis method based on multi-source data, and the method comprises the steps: determining a three-student space classification system in a city, and constructing a calculation index; obtaining remote sensing image data, POI quantity data and AOI area data, and carrying out data processing and index calculation; carrying out three-life space classification and identification on the processed data, and training and generating a three-life space classification model; predicting three-generation types of other space units in the whole research area according to the trained model; and carrying out three-generation spatial pattern analysis on the research area. Multi-source data is utilized, and a random forest method is adopted, so that the recognition is more accurate and rapid, and the credibility is high.