88 results about "Spectral space" patented technology

The invention provides a multi-object tracking method and a device driven by a characteristic trace. A device observing data module converts a physical property and a motion property of the object into data of a spectrum, a space and a time domain; a characteristic trace generating, managing, confirming and deleting module is used for forming the characteristic trace, managing, confirming and deleting the characteristic trace, and finally outputting characteristic trace information; a tracking door forming module is used for receiving the characteristic trace data and generating and outputting the data of the adjacent domains by defining adjacent domains by the characteristic trace; a filtering module is used for receiving the characteristic trace data of the characteristic trace generating, managing, confirming and deleting module, and predicting the state and position of an output object. The method provides the characteristic trace based on the physical basis of the multi-object detection and tracking, gives the physical property and the motion property of using the object in the spectrum, the space and the time domain, fuses the measuring data to generate the characteristic trace, and uses the Markov confidence of the characteristic trace to realize the multi-object tracking method with high-precision.

The present invention discloses a hyperspectral remote sensing image classification method based on the combination of six-layer convolutional neural network and spectral-space information. The method comprises: selecting the hyperspectral remote sensing image data of a certain number of bands; performing space mean-filtering on the two-dimensional image data of each selected band and then converting the format of the multi-band data corresponding to each pixel element; converting the one-dimensional vector into a square matrix, meaning that each pixel elements corresponds to a square matrix data; then, designing a six-layer classifier based on the deep learning template with an input layer, a first convolution layer, a largest pooling layer, a second convolution layer, a full connection layer and an output layer; extracting the square matrix data corresponding to several pixel elements as a training set to be inputted into the classifier and training the classifier; extracting the square matrix data corresponding to several pixel elements as a training set to be inputted into the trained classifier; observing the output classification result of the trained classifier; comparing with the real classification information; and verifying the performances of the trainer. With the method of the invention, higher classification accuracy can be obtained than from the currently available 5-CNN method.

The present invention carries out pre-selection on many LPC codevectors stored in an LSF codebook 101 using a weighted Euclidean distortion as a measure and carries out a full-code selection on the LPC codevectors left after the pre-selection using an amount of distortion in a spectral space as a measure. This makes it possible to improve the quantization performance of the LPC parameter vector quantizer and improve the quality of synthesized speech of the speech coder/decoder.

Generation of a transformed Interim Connection Space for spectral data is provided. A first Interim Connection Space for spectral data in a full spectral space is accessed. A first map, which characterizes a linear transformation from the full spectral space to a first color space, is accessed. A second map, which characterizes a linear transformation from the first Interim Connection Space to the first color space, is determined. The first Interim Connection Space is decomposed into orthogonal subspaces, the orthogonal subspaces including a first subspace that is a null space of the second map and a second subspace that is an orthogonal complement of the null space in the first Interim Connection Space. The transformed Interim Connection Space is generated based on the first subspace and the second subspace.

The invention provides a figure regular hyperspectral image band selection method based on subspace learning. The method comprises the concrete realization steps that (1) a hyperspectral image data matrix is inputted; (2) the data matrix is normalized; (3) a figure regular band similarity matrix is calculated; (4) a figure regular band similarity diagonal matrix is calculated; (5) a reconstruction matrix is constructed; (6) the reconstruction matrix is initialized; (7) the number of times of iteration is set; (8) a subspace band selection matrix is calculated; (9) whether the current number of times of iteration is greater than the maximum number of times of iteration is judged, and the step (10) is performed if the judgment result is yes, or one time of current iteration is added and the step (8) is performed; (10) the subspace band selection matrix is outputted; and (11) a subspace data matrix is constructed. A learning mechanism is provided and spectral space geometric structure information is utilized so that the accuracy of band selection can be enhanced.

Mapping spectral colors in an Interim Connection Space (ICS) of a full spectral space based on an objective function is provided. A spectral color value in the ICS is accessed, and a spectral gamut boundary of the destination gamut is accessed. The spectral color value is mapped into mapped spectral color value based on minimization of an objective function of coordinates of a first subspace of the ICS, by fixing coordinates of a second subspace of the ICS, subject to a constraint that a result is within the spectral gamut boundary. The first subspace is a null space of a transformation from the ICS to a color space, while the second subspace is an orthogonal complement of the first subspace in the ICS. The constraint is determined by a gamut section that is an intersection of the spectral gamut in the ICS and an affine subspace characterized by the fixed coordinates of the second subspace.

A method of removing streaks from multi-band digital images is presented in which a multi-band image is transformed to an advantageous spectral space, in which a streak removal operation is applied to the image in the advantageous spectral space. The streak removal operation is a method of removing columnar streaks from a multi-band digital image of the type in which it is assumed that pixels in a predetermined spatial and spectral region near a given pixel are strongly related to each other and employing gain and offset values to compute streak removal information, a test is performed for a strong relation between the pixels in a predetermined region spatially and spectrally near a given pixel and streak removal information is computed only if such a strong relationship exists, whereby image content that does not extend the full length of the image in the column direction will not be interpreted as a streak.

The invention discloses an underwater sound field numerical simulation method and system based on a Chebyshev polynomial spectrum and a medium. The method comprises the following steps: establishing asimplified control equation of an acoustic propagation Helmholtz equation under a cylindrical coordinate system; completing independent variable transformation from a z coordinate to an x coordinateby using the simplified control equation; implementing Chebyshev forward transformation on the simplified control equation, and mapping a physical space to a spectral space to form a discretized linear equation set in the spectral space; solving the discretized linear equation set in the spectral space to obtain a solution in the spectral space; implementing Chebyshev inverse transformation on thesolution in the spectral space, and mapping the solution in the spectral space back to the physical space; completing independent variable inverse transformation from an x coordinate to a z coordinate to obtain u (r, z), and solving a sound pressure field p; calculating propagation losses. The method is suitable for obtaining higher calculation precision under the condition that few discrete gridpoints are used, the calculation performance of a hardware platform can be brought into full play, and the speed of numerical simulation is remarkably increased.

The invention discloses a multi-source heterogeneous data fusion method based on deep subspace switching integrated learning, which is carried out according to the following steps: S1, respectively extracting features of each source data set and composing spatial spectrum features; S2, carrying out recombination to form a spatial spectrum characteristic data set; S3, processing the sample data inthe spatial spectrum characteristic data set based on a deep sample learn algorithm to form an original sample, a first-order sample and a second-order sample; S4, respectively constructing a classifier model for the three groups of sample sets; S5, classifying the three groups of models by appropriate classification methods to obtain three groups of classification results, and then getting the final classification results by classification voting. The invention combines spectral space information and incorporates layered subspace switching ensemble learning algorithm, utilizes multi-scale spatial spectrum samples, increases useful samples, improves training quality, retains edge information, and improves diversity and separability of samples.

The invention relates to a hyperspectral image classification method, and concretely relates to a hyperspectral image classification method based on spectral space attention fusion and a deformable convolutional residual network. The objective of the invention is to solve the problem of low classification accuracy of hyperspectral images caused by insufficient spectrum and spatial feature extraction and overfitting under small samples due to the fact that the hyperspectral images contain abundant information in the existing hyperspectral image classification. The method comprises the following steps: 1, acquiring a hyperspectral image data set and a corresponding label vector data set; 2, establishing an SSAF-DCR network based on spectrum space attention fusion and a deformable convolution residual error; 3, inputting the x1, the x2, the Y1 and the Y2 into a network SSAF-DCR, and performing iterative optimization by adopting an Adam algorithm to obtain an optimal network; and 4, inputting x3 into the optimal network to carry out classification result prediction. The method is applied to the field of hyperspectral image classification.

The invention discloses a hyperspectral image classification method of a multi-scale spectral space convolutional neural network, and mainly solves the problems that in the prior art, only single-scale features are extracted during inter-spectral feature extraction and spatial feature extraction, and the classification effect on ground object categories with inconcentrated sample distribution or small sample size is poor. According to the implementation scheme, the method comprises the following steps: 1) inputting a hyperspectral image to generate a training sample set and a test sample set with different sample numbers; 2) constructing a multi-scale spectral space convolutional neural network; 3) inputting the training set into a multi-scale spectral space convolutional neural network toobtain a prediction category, calculating hinge cross entropy loss by using the prediction category and a real label, and training the network by using a stochastic gradient descent method until thehinge cross entropy loss converges; and 4) inputting the test sample into the trained multi-scale spectral space convolutional neural network to obtain a classification result. The method can obtain high-accuracy classification under the condition of few training samples, and can be used for ground object type detection of a hyperspectral image.

The invention discloses a medium-low resolution remote sensing image-based enteromorpha coverage information fine extraction method. The method comprises the basic steps of screening and preprocessinga data source; calculating four vegetation indexes of the preprocessed image to obtain a grey-scale map of each vegetation index, and performing local adaptive threshold segmentation on each grey-scale map; calculating an initial enteromorpha coverage range; extracting an enteromorpha distribution rough range and spectral information thereof; constructing and training an enteromorpha prolifera extraction model; and extracting an enteromorpha prolifera coverage fine range. The method provided by the invention is scientific and reasonable, comprehensively considers the accuracy of extracting enteromorpha coverage information by a plurality of vegetation indexes, neural network sample balance and spectral space-time characteristics of enteromorpha pixels in medium-low resolution remote sensing images, and can improve the universality and accuracy of the enteromorpha coverage information extraction method to a certain extent.

The invention provides an intelligent spectrum management and control framework based on a spectrum knowledge graph. The framework comprises a map layer, a device layer and a scene layer. And the graph layer is a driving kernel of the intelligent spectrum management and control framework, namely a multi-domain associated spectrum knowledge graph. And the equipment layer is an execution unit of the intelligent frequency spectrum management and control framework and mainly refers to intelligent frequency equipment for configuring a frequency spectrum knowledge graph. And the scene layer is the application presentation of the intelligent spectrum management and control framework. And the intelligent frequency management center issues a frequency management task to the intelligent frequency equipment in the scene, and the intelligent frequency equipment realizes a given frequency spectrum management and control target and reports information to the intelligent frequency management center at the same time. And the intelligent frequency management center expands and perfects the spectrum knowledge graph according to diversified scenes and tasks. Modeling characterization of the multivariate relation in the spectrum space can be achieved, the intelligent level of spectrum management and control under the complex environment is improved, and the method becomes a new tool in the field of spectrum management and control in the future.

Disclosed is an apparatus and method for color-editing capable of simply carrying out color-editing spectral values, such as spectral reflectance, spectral transmittance, spectral radiance and spectral radiant intensity. When color-editing a spectral value of spectral image, the spectral color is transformed into a low-dimensional color space easy for the human to handle in an HLS space transforming section and color-editing and color-correction are carried out in the low-dimensional color space in on HLS color-editing section thereby generating a proper spectral value from a transformed low-dimensional color in a spectral-space inverse-transforming section. Also, a spectral color editing section receives a spectral editing instruction signal and transforms a wavelength width of a spectral color peak to a narrow one, thereby improving saturation.

Illumination estimation device includes: target object area extraction unit which extracts an object area from a multispectral image, the object area being an area, including specular reflection, of an object; distribution characteristics estimation unit which estimates information representing a subspace that approximates a distribution of observation values in a spectral space, as distribution characteristics, the distribution characteristics being characteristics of a spreading manner of the observation values in the object area within the spectral space; dimension reduction unit which selects a distribution characteristic to be used from among the distribution characteristics estimated by the distribution characteristics estimation unit depending on the number of dimensions to be reduced, the number of dimensions to be reduced being determined in advance by a light environment; and illumination estimation unit which estimates spectral characteristics of illumination based on the subspace information to be represented by the distribution characteristic selected by the dimension reduction unit.

The invention discloses an image multispectral analysis method used for a video monitoring or machine vision field, belongs to the fields of image processing, video monitoring and intelligent transportation, and particularly relates to a video image histogram spectrum segmentation method applied to the video monitoring or machine vision field, and a method for local spectrums to extend and stretch to whole spectral space. A gray scale histogram spectrum of an integral video image is segmented according to requirements, and then, an interested spectrum is stretched to expand to the whole spectral space. Only the interested gray scale spectrum is stretched to the whole spectral space, or each spectrum is stretched to the whole spectral space one by one, so that the dynamic range of the corresponding image of a corresponding spectrum is expanded, the visibility of image details is improved, and the perception and identification problems of vision information under a condition of a wide dynamic range can be favorably solved, and uninterested spectrums can be removed so as to save data bandwidth.

The invention discloses a hyperspectral image spectral space classification method based on class characteristic iterative random sampling. The method comprises the following steps: calculating classcharacteristic criteria of each class according to hyperspectral image data and ground object class label information; iteratively calculating the number of training samples which should be allocatedto each category during classification according to a category feature criterion; calculating an average spectral characteristic and an autocorrelation matrix of a target ground object according to the number of the distributed training samples of each type and the parameter information of the target ground object sample set; calculating abundance information corresponding to each pixel in the hyperspectral image for the ground object through a constraint energy minimization method according to the average spectral characteristic of each ground object and the inverse matrix of the autocorrelation matrix; and calculating a similarity coefficient of two adjacent iterative classification results according to the classification image information, judging whether the similarity coefficient meets an iterative classification cut-off condition, and carrying out iterative classification on the fused and updated data.

The invention discloses a hyperspectral image dimension reduction method, which comprises the following steps: firstly, dividing an original hyperspectral image into non-overlapped superpixels by using an over-segmentation method; next, as the pixel point in one super-pixel usually belongs to the same kind of objects, describing the spatial information by using an intra-class graph in the invention; and finally, introducing the intra-class graph based on the super-pixel level into an LGDE model as a regular item. In addition, in order to effectively capture the nonlinear characteristics of thehyperspectral image, the invention expands the linear LGDE into a kernel version. An original pixel point classification method (RAW), a principal component analysis (PCA), a linear discriminant analysis (LDA) method, a spectral space linear discriminant analysis (SSLDA) method, a local reservation projection (LPP) method, a collaborative graph-based discriminant analysis (CGDE) method, a sparsegraph-based discriminant analysis (SGDE) method, and a local graph-based discriminant analysis (LGDE) method are compared. Under the same experiment conditions, the classification result of the methodis more accurate.

The invention belongs to the technical field of hyperspectral image classification, and particularly relates to a hyperspectral image classification method and device combined with a spectral space multi-layer perceptron, and the method comprises the steps: firstly extracting global spectral features from a hyperspectral image through a spectral multi-layer perceptron; then extracting local spatial features from the hyperspectral image by using a spatial multilayer sensor; and finally, fusing the global spectral features and the local spatial features by using a multi-layer perceptron to perform joint classification on the hyperspectral image. The spectral features and the spatial features can be extracted from the hyperspectral image, the features can be effectively fused and then combined classification is carried out, and the invention has high classification precision.