242 results about "Structure extraction" patented technology

The invention discloses a noise classification method of Gaussian Mixture Model based on a neural network. The recognition rate of noise classification can be improved by using the method. When the method is training, the extracted eigenvector acts as the input of auto-associative neural network (AANN) after being delayed, the structure of the eigenvector is studied by AANN, and the temporal information of the eigenvector sequence is extracted. Then the study result is provided to Gaussian Mixture Models (GMM) in the form of residual eigenvector. The GMM training is carried out by using Expectation Maximization (EM). The weight coefficient of AANN is updated using the method of backward inversion with inertia. The noise classification method makes full use of the advantages of AANN and GMM and greatly improves the recognition rate of the whole noise classification system.

The invention provides a near-infrared imaging device and an identification method based on the palm vein and the palm print. Firstly, a palm image is obtained through a near-infrared imaging device, the central subblock sample needed to be processed is extracted, the subblock is inputted into two feature extraction modules: a genus palm print information code extraction and a vein structure extraction, then the two features are respectively matched, each own similarity of the two features is respectively calculated through using different similarity evaluation methods, the optimized weighted array of the genus palm print and the vein vessel structure is obtained according to a training sample, then the two similarities perform the similarity level fusion, then the similarity level after the fusing performs decision-making and comparing according to a scheduled threshold value, and then the final determination is obtained in reference to the fusioned matching. The near-infrared imaging device and the identification method based on the palm vein and the palm print can overcome the disadvantages of less image features and single processing, and has the advantages of improving the identification rate and the stability of the system.

A low-illumination image processing method and device address the problem of noise amplification in existing contrast enhancement techniques when applied to original low-illumination image. A noise suppression filter is additionally arranged before an operation of contrast enhancement, and smoothing processing is performed on an inverse color image of a low-illumination image by adopting a first filtering coefficient and a second filtering coefficient, so that image contrast is enhanced while random noise is suppressed. Texture and noise level parameter of an image are calculated according to a local characteristic inside block of the image. Weighted averaging is performed on a first smoothing image and a second smoothing image after smoothing processing according to the texture and noise level parameters. The texture image is obtained by performing texture structure extraction on a gradient image of an inverse color image, and the texture image is combined with a weighted image to sharpen the weighted image, to have an effect of enhancing image details. Therefore, the contrast of low-illumination image can be effectively enhanced, various types of noise can be filtered, and the image color and details can be retained at the same time to obtain a clear and vivid restored image.

The invention discloses a single image fine rain removal method based on a depth convolutional neural network. Firstly, through carrying out background texture structure extraction, nonlinear mapping and rain line area restoration on an inputted rain graph by an initial rain removal network, an initial clear rain-free image is finally obtained; the initial clear rain-free image and the original image are inputted to a fine rain removal network with a single convolutional layer at the same time, more details in the background area are thus restored, and a high-definition rain removal image is finally obtained. Through adopting a caffe framework, the initial rain removal network and the fine rain removal network are trained, parameters of each convolutional layer are obtained precisely, fine rain removal processing is carried out on a rain image, and compared with the traditional convolutional neural network rain removal method, the method of the invention can obtain a higher-quality rain-free image, the practicability is strong, and the method can be widely applied to more scenes.

The invention discloses a non-uniform texture small defect detection method based on an improved Faster R-CNN model, and the method comprises the steps: obtaining an image of a to-be-detected object with a to-be-detected defect, and carrying out the feature extraction of the image of the to-be-detected object through the improved Faster R-CNN model; the improved Faster R-CNN model is specificallycharacterized in that features are extracted by embedding a feature pyramid multi-scale fusion structure of a residual complementary attention gate module in the last three stages of a VGG16 network structure. The improved Faster R-CNN model integrates feature map information of different scales and receptive fields of each stage from top to bottom in a feature pyramid manner; the judging capacityfor different types of defects is enhanced, particularly the judging capacity for wrinkles of different degrees is remarkably improved, the overall recall rate is remarkably increased, and the industrial precision requirement is met. The residual complementary attention gate module can guide multi-scale feature fusion to suppress complex background information, and then context information is captured globally to locate small defects more accurately.

The invention discloses a cell detection method based on a sliding window and depth structure extraction features. The cell detection method is used for automatically detecting cells by utilizing depth model extraction features and then applying a sliding window technology to a pathological section image. The cell detection method comprises the following steps: section image blocking, training of stacked and sparse self-coding of a feature extraction model, detector training, scanning of a large image by the sliding window and cell position labeling. According to the cell detection method, the large section image is used as a search object, the positions of cells in the image can be found more accurately, faster and completely by adopting a new method of combining a detector and the sliding window, and a good detection effect can be achieved for some unobvious cells in the image. The automatic cell detection method disclosed by the invention can be used for assisting a clinical doctor in carrying out quantitative evaluation on digital pathological sections and accurately and rapidly carrying out clinical diagnosis, so that the diagnosis difference of different observers or one observer at different time periods is reduced.

The invention discloses a text detecting method based on extraction of linear structures. A text connection region can regard as a combination of the linear structures which are in different shapes. The natural scene text detecting method comprises the following steps of roughly positioning text regions by extracting the linear structures; performing color clustering and region growth based on color distances to obtain a complete text connection region from the initial text regions; and further analyzing geometric and spatial position features of the connection region to obtain a candidate text character string. Extraction of the linear structures is based on a multi-scale Hessian matrix filtering method, a channel R, a channel G and a channel B of an original colorful image are operated respectively, and influences of illumination on the image are overcome to a certain degree by the text detecting method; the text detecting method can be used for well detecting texts with different sizes and different stroke widths by color clustering of Gaussian mixed models, region growth of color distances and a two-step color analysis method; and the robustness of the method is further improved.

The invention provides a method for extracting a phytocoenosium spatial structure. The method comprises: performing multi-resolution segmentation of a to-be-tested remote-sensing image in a target area to obtain remote-sensing image objects with different resolutions; establishing a corresponding relation between an image resolution of the to-be-tested remote-sensing image and an ecological organization resolution to obtain an image resolution of each plant type in the to-be-tested remote-sensing image, wherein the plant types include a meadow, a shrub, an arbor, a population and a group, wherein the meadow, the shrub, and the arbor are plant individuals; performing vegetation classification of a pre-selected sample of the to-be-tested remote-sensing image in plant individual and population image resolution according to the plant individuals and the population image resolution; summing the classification result of each resolution to a grouped data layer; and calculating plant individuals and parameters of a population spatial structure in a group resolution object boundary. The method for extracting the phytocoenosium spatial structure is relatively accurate, and is low in monitoring cost and high in objectivity.

The invention provides a low-illumination image processing method and device. A noise suppression filter is additionally arranged before operation of contrast enhancement by aiming at a problem of noise amplification existing in an original low-illumination image contrast enhancement technology, and smoothing is performed on the inverse color image of a low-illumination image by adopting a first filtering coefficient and a second filtering coefficient so that image contrast is enhanced and random noise is suppressed simultaneously. Texture and noise level parameters of the image are calculated according to the local block interior characteristics of the image. Weighted averaging is performed on a first smooth image and a second smooth image after smoothing according to the texture and the noise level parameters. A texture image is obtained by performing texture structure extraction on the gradient image of the inverse color image, and the texture image is combined with the weighted image and the weighted image is sharpened so that the effect of enhancing image details can be realized. Therefore, contrast of the low-illumination image can be effectively enhanced, various types of noise can be filtered, the image color and details can be retained and thus a clear and lifelike restored image can be obtained.

A structure extraction unit extracts a heart region from a three-dimensional image of a chest, an image display control unit displays a volume rendering of the extracted heart region, an information adding unit adds additional information such as text and voice to a VR image displayed according to desired display conditions which include an image orientation. The display conditions when the additional information is added are stored with the additional information as designated display conditions. An allowable range setting unit sets a predetermined allowable range for the designated display conditions. During subsequent display of the VR image, an information reproducing unit enables the added information to be reproduced when display conditions are within the allowable range of the designated display conditions.

A continuous-region-extraction unit extracts a continuous-region having a voxel value corresponding to an air-region in bronchi from a three-dimensional-medical-image. A tree-structure-construction unit constructs a tree-structure corresponding to the continuous-region. A linear-structure-extraction unit extracts plural linear-structures representing fragments of small-bronchi by analyzing a local density structure in a neighborhood of each point in the three-dimensional-medical-image. A tree-structure-reconstruction unit reconstructs a graph-structure representing the whole bronchi by connecting a node constituting the graph-structure of large-bronchi and a node representing the linear-structures of small-bronchi. At this time, different cost functions are used for a segment connecting the node of the large-bronchi and the node representing the linear-structures of the small-bronchi and a segment connecting the nodes representing the linear-structures to each other. The cost function in the former segment is defined in such a manner that the segment is more likely to be connected as a change in voxel values is smaller.

The invention discloses a three-dimensional image visualization method based on a two-dimensional transfer function, which comprises the steps of: firstly, establishing a generalized model of the two-dimensional transfer function, and designing four classifiers; secondly, designing a concrete model of the two-dimensional transfer function; and thirdly, drawing a three-dimensional image according to optical parameters obtained by the two-dimensional transfer function. The method is simple, has popularization on a volume rendering algorithm based on graphics hardware and a volume rendering algorithm based on software, is visual in results, and can realize optical parameter set, user interest structure extraction, independence structure inhibition and three-dimensional image semitransparent rendering in volume data.

The invention discloses an MODIS data-based regional large-scale crop planting structure extraction method, belongs to the technical field of agricultural planting structures, and aims to solve the problem of large classification error of an existing extraction method for a crop planting structure in agricultural remote sensing monitoring. The method comprises the steps of firstly collecting whole year image data of a large-scale monitoring region, and performing preprocessing to obtain multiple complete images of the large-scale monitoring regions; secondly performing cropping on the complete images, establishing and extracting a complete time sequence document of years, and obtaining an NDVI time sequence curve of each pixel; thirdly extracting 11 pieces of phenological data of the whole year, and performing floating point processing; fourthly extracting key information of each piece of the phenological data; fifthly obtaining multiple phonological partitions; sixthly performing multi-scale segmentation to obtain planting structure units; and finally performing sample-based classification extraction by adopting a nearest neighbor classification method to obtain the crop planting structure of the large-scale monitoring region. The method is used for extracting the crop planting structure.

The invention relates to a park structure extraction method based on LiDAR data and ortho-images. The method comprises the following steps of: dividing a park into a vacancy region and a non-vacancy region by virtue of the LiDAR data, generating a central axis of a vehicle surface patch in the non-vacancy region from the LiDAR data, and acquiring parking lines of the park in the vacancy region from ortho-image data; registering the vehicle center axis in the non-vacancy region and the parking lines in the vacancy region, and dividing parking lanes along a maximum intersection direction of the vehicle center axis and the parking lines; and calculating park structure parameters, and generating dividing lines of the parking lanes, thus finishing the extraction of the park structure. The park structure extraction method can be used for solving the problems of illumination variation, shadow effect, perspective distortion, vehicle covering and the like in the park structure extracting process, and extracting the park structure accurately and precisely.

A structure extraction element obtains three-dimensional structure extraction data by extracting a tubular structure from volume data obtained by imaging a predetermined subject. A relating element relates positions on paths of a two-dimensional template representing a schematic diagram of a three-dimensionally branching tubular structure obtained by imaging a predetermined subject and positions on paths of the three-dimensional structure extraction data to each other. A projection image generation element generates a two-dimensional projection image by projecting voxel values of the tubular structure that is present on the paths of the three-dimensional structure extraction data onto corresponding positions on the paths of the two-dimensional template.

The invention discloses a wireless energy and data synchronous transmission system based on inductive coupling and FSK modulation and a parameter design method thereof, and belongs to the technical field of the wireless electric energy transmission. The problem that the data transmission of the wireless energy and data synchronous transmission system based on multi-carrier communication is high inbit error rate and complex in data loading and circuit structure extraction under a high magnetic field is solved. The data carrier is generated by adopting a FSK modulation way; compared with the ASK modulation way, the interference resistance of the FSK modulation way is strong, and the anti-attenuation performance is good. The way of firstly adopting the inductive coupling comprises loading amodulated signal on a to-be-transmitted energy signal, and the way of inductive coupling is used for extracting the modulated signal which has been transmitted to a superposed signal of a secondary side circuit. Compared with the existing circuits of the data carrier signal loading circuit and the data carrier signal extraction signal based on the capacitance coupling, the circuit structures of the signal loading unit and the signal extraction units are relatively simple.

A speech processing apparatus includes a spectrum envelope extracting unit which extracts the spectrum envelope of an input speech signal, a spectrum envelope deforming unit which applies deformation to the spectrum envelope to generate a deformed spectrum envelope, a spectrum fine structure extracting unit which extracts the spectrum fine structure of the input speech signal, a deformed spectrum generating unit which generates a deformed spectrum by combining the deformed spectrum envelope with the spectrum fine structure, and a speech generating unit which generates an output speech signal on the basis of the deformed spectrum. This apparatus emits a disrupting sound based on the output speech signal to prevent a third party from eavesdropping on a conversation.

The invention discloses an unmanned aerial vehicle weak and small target detection method based on a space-time attention mechanism. The method comprises the following steps: S1, performing cascade transmission and multi-scale fusion target feature extraction based on space attention are carried out; s2, inputting an image feature sequence based on time attention: in combination with the time continuity of the spatial position and attitude features of the target in the continuous detection process, using a section of continuous feature sequence as input, and inputting the continuous feature sequence into a time attention structure built by a time sequence network to enhance the target information; s3, decision output based on space-time attention: outputting the target position and the category confidence according to the image features extracted by the space-time attention structure. The invention has the advantages of high efficiency, low cost, accurate data, flexible operation and the like.

The invention provides a 3D vascular structure extraction method based on a convolutional cycle network. Main contents comprise experiment data, integration data, a convolutional neural network (CNN) framework and a convolutional long-term and short-period memory unit (ConvLSTM). The method comprises the following steps that clinic data and synthesis data are used to carry out experiment; actual data adopts a high-resolution fluorescence multiphoton microscope; in a mice, abdominal cavity model imaging is performed to acquire an image so as to describe a vascular system; a noise is added to simulate an actual condition; a CNN network and a convolutional LSTM network are combined; the convolutional long-term and short-period memory unit uses a deep configuration and shallow configuration combination and a training network and uses a weighted binary system cross entropy loss to test a training result. In the invention, from a complex microscope image, a real 3D vascular structure can be extracted, convolution complexity in three dimensions is reduced, a number of model parameters is greatly decreased, a wide view is possessed, and abundant characteristics can be extracted.

The embodiment of the invention discloses a method and device for generating structured information. The method comprises the specific steps of capturing all the page content of an enterprise web site in a predetermined field; categorizing the page content to the category of the pre-structured enterprise information and other categories according to the feature of the page content, and obtaining the categorized page; dividing the categorized page into a content page and a form page, using the content page and the form page as tags, and labelling the categorized page; extracting at least one of the categorized pages, and obtaining the extracted information: body block extraction, body content structured extraction, image text block extraction, list block extraction and structuring of predetermined location content; constructing the structured information based on the extracted information. The implementation method ensures the diversity of information sources, so that the content feature presented by the generated structured information is rich and redundant information is reduced.

A label adding unit adds labels to structures such as a body surface region, a lung region, bronchi, and pulmonary nodules of a human extracted by a structure extraction unit from a three-dimensional image of a chest. An image display control unit displays the three-dimensional image by volume rendering on a display unit. At this time, a label display determination unit determines at least one label to be displayed with the volume rendering image to be displayed based on the opacity during the volume rendering display. A label display control unit displays the determined label with the volume rendering image on the display unit.

The invention discloses a structural extraction method and system of positional information based on a semantic position model. The method includes establishing a positional basic concept ontology, a positional physical concept ontology based on the semantic position model and filling ontology instance; classifying the positional basic concept ontology and the positional physical concept ontology to establish corresponding positional structure concept ontologies; automatically mapping the location structuring concept ontologies to generate rule classes, each rule class corresponds to one positional physical concept ontology and is inheriting and greedy; instantiating all the rule classes to generate rule objects, and to finally generate rule sets; classifying the rule sets, re-classifying them into a plurality of rule groups to form a rule retrieval dependency graph; further establishing a reverse retrival between the concept ontology and the rules of each rule group; matching positional descriptive graphs by using a rule matching algorithm to match and finish structuring extraction of positional information.