CN107590472B - Spectrum classifier and classification method based on push-broom spectrum imager - Google Patents
Spectrum classifier and classification method based on push-broom spectrum imager Download PDFInfo
- Publication number
- CN107590472B CN107590472B CN201710844657.XA CN201710844657A CN107590472B CN 107590472 B CN107590472 B CN 107590472B CN 201710844657 A CN201710844657 A CN 201710844657A CN 107590472 B CN107590472 B CN 107590472B
- Authority
- CN
- China
- Prior art keywords
- spectrum
- units
- broom
- push
- classification
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000001228 spectrum Methods 0.000 title claims abstract description 90
- 238000000034 method Methods 0.000 title claims abstract description 21
- 238000012545 processing Methods 0.000 claims abstract description 20
- 239000013598 vector Substances 0.000 claims description 32
- 239000011159 matrix material Substances 0.000 claims description 22
- 230000003595 spectral effect Effects 0.000 claims description 22
- 238000012549 training Methods 0.000 claims description 21
- 230000003287 optical effect Effects 0.000 claims description 20
- 238000000605 extraction Methods 0.000 claims description 14
- 238000004422 calculation algorithm Methods 0.000 claims description 10
- 238000004458 analytical method Methods 0.000 claims description 8
- 238000001514 detection method Methods 0.000 claims description 6
- 238000013139 quantization Methods 0.000 claims description 5
- 241000764238 Isis Species 0.000 claims description 2
- 238000005516 engineering process Methods 0.000 abstract description 7
- 238000003384 imaging method Methods 0.000 abstract description 5
- 230000005540 biological transmission Effects 0.000 abstract description 4
- 238000007635 classification algorithm Methods 0.000 abstract description 3
- 238000000701 chemical imaging Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 239000000126 substance Substances 0.000 description 2
- 238000007476 Maximum Likelihood Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012306 spectroscopic technique Methods 0.000 description 1
- 238000004611 spectroscopical analysis Methods 0.000 description 1
- 230000001131 transforming effect Effects 0.000 description 1
Landscapes
- Image Analysis (AREA)
- Spectrometry And Color Measurement (AREA)
Abstract
The invention provides a spectrum classifier and a classification method based on a push-broom spectrum imager, aiming at solving the technical problem that the load of a data transmission, storage and processing system is greatly increased due to the fact that the reconstructed data volume of a data cube is too large in the existing imaging spectrum technology. The spectrum classifier comprises a push-broom spectrum imager, a spatial light modulator, a controller, a detector and a signal processing module; the spatial light modulator modulates the spectrum of the push-broom spectrum imager after light splitting according to the digital voltage signal output by the controller; the detector is used for carrying out current travelling extracting the characteristics of the target pixel; the signal processing module is used for carrying out classification judgment on the current line target pixels and outputting classification results. The method directly applies the classification algorithm to the spectrum after the light is split in the push-broom spectrum imager, realizes spectrum classification, can directly output classification results without reconstructing a data cube, and can effectively avoid the problem of large data volume faced at present.
Description
Technical Field
The invention belongs to the technical field of optics, and relates to a spectrum classifier and a classification method based on a push-broom spectrum imager.
Background
The hyperspectral imaging technology can detect the two-dimensional geometric space and the one-dimensional spectral information of the target at the same time, and has the characteristic of 'map-in-one'; the hyperspectral imaging technology realizes breakthrough improvement of spectral resolution, the number of wave bands is more, and the spectral information corresponding to each pixel can be drawn into a complete and continuous spectral curve as a function of wavelength, so that diagnostic spectral characteristics capable of distinguishing different substances are reflected. The characteristics determine that the hyperspectral image has stronger ground feature recognition and fine classification capability than other remote sensing images.
The fine classification of objects using hyperspectral data is one of the core content of hyperspectral imaging technology applications, the final objective of which is to assign a unique class identification to each pixel in a hyperspectral image. At present, the hyperspectral classification algorithm takes a data cube obtained by reconstructing data output by a hyperspectral imager as a research object, analyzes spectrum and space information of various substances, performs feature extraction, and performs classification judgment by a proper classification method.
However, hyperspectral imaging acquires the rich spatial, radiation and spectral information of the target, and brings a new challenge while bringing opportunities for fine classification of the target. The huge data volume inevitably brings pressure to a data processing system, and the huge operation volume causes the great increase of the computer load; meanwhile, the large data volume also brings time and storage space bottlenecks to the data transmission system and the storage system.
Many novel imaging spectrum technologies, such as a computational spectrum imaging technology based on a compressed sensing theory, proposed by students at home and abroad in recent years reduce the system overhead to a certain extent by reducing the data acquisition amount. However, these advanced novel imaging spectroscopic techniques, the reconstruction algorithms of which tend to be relatively complex and difficult to implement, the reconstruction effect being dependent on the sparsity of the spectroscopic data cube. Moreover, its ultimate goal is still the reconstruction of the data cube, and thus still faces the large data volume problem with the reconstructed data cube.
Disclosure of Invention
In order to solve the technical problem that the load of a data transmission, storage and processing system is greatly increased due to the fact that the reconstructed data volume of a data cube is too large in the conventional imaging spectrum technology, the invention provides a spectrum classifier based on a push-broom spectrum imager and a classification method.
The technical scheme of the invention is as follows:
the spectrum classifier based on the push-broom spectrum imager is characterized by comprising the push-broom spectrum imager, a spatial light modulator, a controller, a detector and a signal processing module;
the push-broom spectrum imager is used for scanning and dispersing one row of target pixels to form a spectrum after light splitting, which is arranged according to the wavelength sequence and is projected onto the spatial light modulator;
the spatial light modulator modulates the spectrum after light splitting according to the digital voltage signal output by the controller; the digital voltage signal is a quantized value of a feature vector in a projection matrix obtained by projecting a spectrum space of a data cube into an N-1 dimensional feature space through a linear discriminant analysis algorithm, and the quantized value is sequentially output by the controller according to the feature value of the feature vector from large to small; n is the category number of the known spectrum;
the detector is used for receiving the light information modulated by the spatial light modulator and extracting the characteristics of the current line target pixels;
the signal processing module receives the characteristic information output by the detector, and carrying out classification judgment on the current row of target pixels, and outputting classification results.
Further, the spatial light modulator is a two-dimensional digital micro-mirror array, and comprises a plurality of micro-mirror units, wherein each two rows of micro-mirror units correspond to one target pixel; the reflecting mirrors of the micro mirror units are respectively controlled to rotate by the controller to realize independent optical switches, and the optical information is modulated by controlling the optical switching time of each micro mirror unit;
the number of the micro mirror units in each row of the two-dimensional digital micro mirror array is more than or equal to the band number of the spectrum after light splitting, and the number of the micro mirror units in each column is more than or equal to 2 times of the pixel number of one row of target pixels; in the two-dimensional digital micromirror array, the micromirror units in the odd-numbered rows are controlled by the quantized values corresponding to the positive values in the feature vectors, and the micromirror units in the even-numbered rows are controlled by the quantized values corresponding to the negative values in the feature vectors; the modulated light output by each two rows of micro mirror units is received by two corresponding detection units in the detector.
Further, the signal processing module is used for classifying and judging the current line target pixels according to a Bayesian judgment criterion.
The invention also provides a method for carrying out spectrum classification on the target pixel by utilizing the spectrum classifier based on the push-broom spectrum imager, which comprises the following steps:
1) The push-broom spectrum imager scans and disperses a row of target pixels to form a spectrum after light splitting, which is arranged according to the wavelength sequence;
2) The controller outputs a digital voltage signal to control the spatial light modulator to carry out optical modulation on the spectrum after the light splitting generated in the step 1); the digital voltage signal is determined by a linear discriminant analysis algorithm: projecting the spectrum space of the training sample set to an N-1 dimensional feature space to obtain a projection matrix, and quantizing feature vectors in the projection matrix to obtain the digital voltage signal after quantization; the digital voltage signal is controlled by the controller according to the characteristics the characteristic values corresponding to the vectors are sequentially output from large to small; n is the category number of the known spectrum;
3) The detector receives the spectrum information modulated in the step 2) and realizes the feature extraction of the current line target pixels;
4) The signal processing module receives the characteristic information extracted in the step 3), carries out classification judgment on the current line target pixels, and outputs classification results of the current line target pixels;
5) Repeating the steps 1) -4), and completing the spectrum classification of the target pixels of the other rows line by line in sequence.
Further, the optical modulation in the step 2) is implemented by using a two-dimensional digital micromirror array, where the two-dimensional digital micromirror array includes a plurality of micromirror units, and each two rows of micromirror units corresponds to one target pixel; the reflecting mirrors of the micro mirror units are respectively controlled to rotate by the controller to realize independent optical switches, and the optical information is modulated by controlling the optical switching time of each micro mirror unit;
the number of the micro mirror units in each row of the two-dimensional digital micro mirror array is more than or equal to the band number of the spectrum after light splitting, and the number of the micro mirror units in each column is more than or equal to 2 times of the pixel number of one row of target pixels; in the two-dimensional digital micromirror array, the micromirror units in the odd-numbered rows are controlled by the quantized values corresponding to the positive values in the feature vectors, and the micromirror units in the even-numbered rows are controlled by the quantized values corresponding to the negative values in the feature vectors; the modulated light output by each two rows of micro mirror units is received by two corresponding detection units in the detector.
Further, in the step 4), classification judgment is performed on the current target pixel according to a bayesian judgment criterion, wherein the judgment criterion is as follows:
in the above formula, x refers to an N-1 dimensional vector obtained by extracting features of a current sample, and c refers to each spectrum category existing in a training sample set; p (c) is the prior probability of each spectral class c, estimated by the frequency of occurrence of each class sample in the training sample set; p (x|c) is the class conditional probability of x for each spectral class c; mu (mu) c Sum sigma c Respectively the mean value and the variance of each class of samples in the training sample set; x is x i Refers to the ith dimension value of x; n is the number of known spectral categories; sigma (sigma) c,i Sum mu c,i Respectively sigma c Sum mu c Is the i-th dimensional value of (2);
the judging method comprises the following steps: and according to the formula, calculating the P 'of each category in the training sample set corresponding to the current target pixel, wherein the spectrum category corresponding to the maximum value of the P' is judged to be the spectrum category to which the current target pixel belongs.
The invention has the advantages that:
based on the compression coding principle, the method directly applies a classification algorithm (linear discriminant analysis and Bayesian judgment criterion) to the spectrum after light splitting in the push-broom spectrum imager to realize spectrum classification; the invention can directly output the classification result without reconstructing the data cube, can effectively avoid the problem of large data volume faced at present, and greatly lightens the load of a data transmission, storage and processing system.
Drawings
Fig. 1 is a schematic block diagram of the spectral classifier of the present invention.
FIG. 2 is a diagram of a classification process of a spectrum classifier according to an embodiment of the invention.
Detailed Description
The present invention will be described in detail with reference to the accompanying drawings.
FIG. 1 is a schematic block diagram of a spectrum classifier of the invention, which structurally comprises a push-broom spectrum imager, a spatial light modulator, a controller, a detector and a signal processing module.
Firstly, scanning a row of target pixels by a push-broom type spectrum imager, dispersing, and forming a spectrum after light splitting, which is arranged according to a wavelength sequence; then, the spatial light modulator modulates the spectrum after light splitting according to the digital voltage signal (namely the conversion coefficient, which is determined by a linear discriminant analysis algorithm) provided by the controller, and then the detector receives the modulated light information to realize the feature extraction of the current line target pixel; finally, the signal processing module receives the data output by the detector, namely the extracted characteristic information, carries out classification judgment (determined by Bayesian judgment criteria) on the target pixels, and outputs classification results.
Referring to fig. 2, the spatial light modulator of this embodiment employs a two-dimensional array of digital micromirrors. The two-dimensional digital micro-mirror array is an electrically addressed spatial light modulator, and is formed by arranging a plurality of micro-mirror units into a two-dimensional array, wherein each two rows of micro-mirror units correspond to one target pixel; the reflector of each micro mirror unit is independently controlled to rotate by a digital voltage signal output by the controller, so that an independent optical switch is realized, and the modulation of optical information is realized by adjusting the time of the optical switch.
The number of micro mirror units of each row of a two-dimensional digital micro mirror array (DMD) is more than or equal to the band number of the spectrum after light splitting, and the number of micro mirror units of each column is more than or equal to 2 times of the pixel number of a row of target pixels; in other embodiments, the number of micromirrors in each column of a two-dimensional digital micromirror array (DMD) is greater than or equal to the number of bands of the spectrum after light splitting, and the number of micromirrors in each row is greater than or equal to 2 times the number of pixels of a row of target pixels, where the specific structural form is determined according to the direction of light dispersion during light splitting; since the signal controlling the two-dimensional digital micromirror array is a positive digital voltage signal, the eigenvectors in the projection matrix W are analog and have positive and negative values. Firstly, converting a feature vector in a projection matrix W into a digital signal through quantization, then, projecting a spectrum corresponding to each target pixel onto two rows of micro-mirror units of a two-dimensional digital micro-mirror array through light path adjustment, wherein one row of micro-mirror units is controlled by a quantization value corresponding to a positive numerical value in the feature vector, one row of micro-mirror units is controlled by a quantization value corresponding to a negative numerical value in the feature vector, and modulated light output by each two rows of micro-mirror units is respectively received by two corresponding detection units on a detector; and subtracting the data received by the two detection units corresponding to the current target pixel to obtain the characteristics of the current target pixel. In this embodiment, the micromirror units in the odd-numbered rows are controlled by the quantized values corresponding to the positive values in the feature vectors, and the micromirror units in the even-numbered rows are controlled by the quantized values corresponding to the negative values in the feature vectors.
And the signal processing module receives the characteristics of the target pixels acquired and output by the detector, performs classification judgment and outputs classification results. The specific process comprises two major parts of characteristic extraction and classification judgment of target pixels (namely unknown pixels). In view of the fact that hyperspectral data have a large number of bands, the correlation among the bands is strong, and the data redundancy is high, data dimension reduction, namely feature extraction, is needed first, high-dimension data information is effectively expressed by utilizing low-dimension data, and features with the largest difference in different categories are obtained through feature extraction while the data size is compressed. And then, processing the extracted features by using a proper classification judgment algorithm to realize classification.
1. Feature extraction
The feature extraction process corresponds to a process of mathematically transforming the spectral space of the data cube. While spatial light modulators are one of the key devices in the modern optical field, they contain a number of individual cells arranged in a spatial array, each of which is independently controllable to modulate light in real time. Therefore, the spectrum after being split by the push-broom spectrum imager can be directly modulated by using the spatial light modulator, and the spectrum space band coding process of the characteristic extraction algorithm is realized.
The Linear Discriminant Analysis (LDA) algorithm is a supervision dimension reduction method, and the idea is that: given a training sample set, a projection direction is sought to be found based on a multidimensional normal distribution. In this direction, the projections of the different classes of samples can be best separated, i.e. with maximum inter-class distance and minimum intra-class discreteness. In the multi-classification task, the projection operator is obtained by the following formulas (1) to (3).
S b W＝λS w W (1)
Wherein S is b Is an inter-class divergence matrix; s is S W Is an intra-class divergence matrix; n is the number of known spectral categories; m is m i Is the number of training samples for each category; mu (mu) i Is the mean vector of each class training sample; μ is the mean vector of all training samples; x is an N-1 dimensional vector obtained by extracting features of a current sample; lambda isIs a characteristic value of (2); w is a projection matrix and is composed of eigenvectors corresponding to N-1 maximum eigenvalues.
The spectrum space of the training sample set can be projected to the N-1 dimension feature space through LDA, so that the data dimension reduction is realized. The larger the eigenvalue λ, the greater the class separability in the corresponding eigenvector direction. Here, the feature vector in the projection matrix W is a transform coefficient in the feature extraction process.
Therefore, by analyzing the known samples, the projection matrix W can be calculated in advance, and then the controller in fig. 1 sequentially outputs the transform coefficients (feature vectors) in the projection matrix W in order of the corresponding feature values from the large to the small to control the spatial light modulator, thereby realizing feature extraction.
2. Classification judgment
The invention realizes the classification judgment of the target pixel based on Bayes judgment criteria (Bayes). The Bayes criterion is that, to minimize the overall risk, only that spectral class label c that minimizes the conditional risk R (c|x) needs to be selected on each X, as shown in equation (4).
R(c|x)＝1-P(c|x) (4)
P (c|x) is the posterior probability of x, which can be written as equation (5); thus, the Bayes criterion may be interpreted as selecting the class label that maximizes the posterior probability. While for a given x, P (x) is a fixed value, independent of the class mark. Thus, the Bayes criterion can be converted to equation (6).
P(c|x)＝P(c)P(x|c)/P(x) (5)
P′＝P(c)P(x|c) (6)
In the above formula, c refers to each spectrum class existing in the training sample set; p (c) is the prior probability for each spectral class c, which can be estimated by the frequency of occurrence of each class sample in the training sample set; p (x|c) is the class conditional probability of x for each spectral class c, one classical approach to estimate P (x|c) is maximum likelihood estimation, as shown in equation (7) in the case of continuous properties, where μ c Sum sigma c The mean and variance of each class of samples in the training sample set are respectively.
The dimension of the hyperspectral data after being extracted by the LDA features is N-1, so that for the case that the category number N is more than 2, the extracted features are multidimensional, according to naive Bayes,
wherein x is i Refers to the ith dimension value of x in the N-1 dimension vector after feature extraction; sigma (sigma) c,i Sum mu c,i Respectively sigma c Sum mu c Is the i-th dimensional value of (c).
In summary, when the present invention is used for spectrum classification, the projection matrix W is first calculated in advance according to the formulas (1) - (3), where the size of the projection matrix W is p× (N-1), where P is the spectral band number of the original data (i.e. the data before the dimension reduction, i.e. the data before the feature extraction). I.e. W consists of N-1 eigenvectors, each column being a P-dimensional eigenvector, each eigenvector corresponding to an eigenvalue. The controller sequentially outputs the feature vectors in the quantized projection matrix W according to the order of the corresponding feature values from large to small, thereby sequentially extracting each feature, and sequentially receiving each extracted feature (the feature=positive feature-negative feature) by the detector. After all the features are extracted, calculating P 'of each pixel for each spectrum class c by a signal processing module according to the formula (8), selecting the spectrum class corresponding to the maximum value of P', judging the spectrum class as the class to which the pixel belongs, and outputting a classification result.
Claims (6)
1. The spectrum classifier based on the push-broom spectrum imager is characterized by comprising the push-broom spectrum imager, a spatial light modulator, a controller, a detector and a signal processing module;
the push-broom spectrum imager is used for scanning and dispersing one row of target pixels to form a spectrum after light splitting, which is arranged according to the wavelength sequence and is projected onto the spatial light modulator;
the spatial light modulator modulates the spectrum after light splitting according to the digital voltage signal output by the controller; the digital voltage signal is a quantized value of a feature vector in a projection matrix obtained by projecting a spectrum space of a data cube into an N-1 dimensional feature space through a linear discriminant analysis algorithm, and the quantized value is sequentially output by the controller according to the feature value of the feature vector from large to small; n is the category number of the known spectrum;
the method for obtaining the projection matrix by projecting the spectrum space of the data cube to the N-1 dimension characteristic space through the linear discriminant analysis algorithm comprises the following specific steps: decomposing the characteristic value by using a formula (1) to obtain a projection matrix W andis a characteristic value lambda of (1);
S b W＝λS w W (1)
wherein the inter-class divergence matrix S b And an intra-class divergence matrix S w The method is calculated by a formula (2) and a formula (3) respectively:
wherein m is i Is the number of training samples for each category; mu (mu) i Is the mean vector of each class training sample; μ is the mean vector of all training samples; x is an N-1 dimensional vector obtained by extracting features of a current sample; lambda isIs a characteristic value of (2); w is a projection matrix and is composed of N-1 eigenvectors corresponding to the maximum eigenvalues;
the detector is used for receiving the light information modulated by the spatial light modulator and extracting the characteristics of the current line target pixels;
and the signal processing module receives the characteristic information output by the detector, performs classification judgment on the current row of target pixels and outputs classification results.
2. The spectrum classifier based on the push-broom spectrum imager according to claim 1, wherein the spatial light modulator is a two-dimensional digital micromirror array, and comprises a plurality of micromirror units, wherein each two rows of micromirror units correspond to one target pixel; the reflecting mirrors of the micro mirror units are respectively controlled to rotate by the controller to realize independent optical switches, and the optical information is modulated by controlling the optical switching time of each micro mirror unit;
the number of the micro mirror units of each row of the two-dimensional digital micro mirror array is more than or equal to the band number of the spectrum after light splitting, and the number of the micro mirror units of each column is more than or equal to 2 times of the pixel number of a row of target pixels; in the two-dimensional digital micromirror array, the micromirror units in the odd-numbered rows are controlled by the quantized values corresponding to the positive values in the feature vectors, and the micromirror units in the even-numbered rows are controlled by the quantized values corresponding to the negative values in the feature vectors; the modulated light output by each two rows of micro mirror units is received by two corresponding detection units in the detector.
3. The spectrum classifier based on the push-broom spectrum imager according to claim 1 or 2, wherein the signal processing module is a signal processing module for classifying and judging the current line of target pixels according to a bayesian judgment criterion.
4. A method for spectrally classifying a target pixel using a push-broom spectral imager-based spectral classifier according to any one of claims 1-3, comprising the steps of:
1) The push-broom spectrum imager scans and disperses a row of target pixels to form a spectrum after light splitting, which is arranged according to the wavelength sequence;
2) The controller outputs a digital voltage signal to control the spatial light modulator to carry out optical modulation on the spectrum after the light splitting generated in the step 1); the digital voltage signal is determined by a linear discriminant analysis algorithm: projecting the spectrum space of the training sample set to an N-1 dimensional feature space to obtain a projection matrix, and quantizing feature vectors in the projection matrix to obtain the digital voltage signal after quantization; the digital voltage signals are sequentially output by the controller from large to small according to the characteristic values corresponding to the characteristic vectors; n is the category number of the known spectrum;
3) The detector receives the spectrum information modulated in the step 2) and realizes the feature extraction of the current line target pixels;
4) The signal processing module receives the characteristic information extracted in the step 3), carries out classification judgment on the current line target pixels, and outputs classification results of the current line target pixels;
5) Repeating the steps 1) -4), and completing the spectrum classification of the target pixels of the other rows line by line in sequence.
5. The method according to claim 4, wherein the optical modulation in the step 2) is implemented by using a two-dimensional digital micromirror array, the two-dimensional digital micromirror array comprising a plurality of micromirror units, each two rows of micromirror units corresponding to one target pixel; the reflecting mirrors of the micro mirror units are respectively controlled to rotate by the controller to realize independent optical switches, and the optical information is modulated by controlling the optical switching time of each micro mirror unit;
the number of the micro mirror units in each row of the two-dimensional digital micro mirror array is more than or equal to the band number of the spectrum after light splitting, and the number of the micro mirror units in each column is more than or equal to 2 times of the pixel number of one row of target pixels; in the two-dimensional digital micromirror array, the micromirror units in the odd-numbered rows are controlled by the quantized values corresponding to the positive values in the feature vectors, and the micromirror units in the even-numbered rows are controlled by the quantized values corresponding to the negative values in the feature vectors; the modulated light output by each two rows of micro mirror units is received by two corresponding detection units in the detector.
6. The method according to claim 4, wherein in the step 4), classification judgment is performed on the current target pixel according to a bayesian judgment criterion, and the judgment basis is:
in the above formula, x refers to N-1 dimension obtained by extracting features from the current sampleVector, c, refers to each spectral class present in the training sample set; p (c) is the prior probability of each spectral class c, estimated by the frequency of occurrence of each class sample in the training sample set; p (x|c) is the class conditional probability of x for each spectral class c; mu (mu) c Sum sigma c Respectively the mean value and the variance of each class of samples in the training sample set; x is x i Refers to the ith dimension value of x; n is the number of known spectral categories; sigma (sigma) c,i Sum mu c,i Respectively sigma c Sum mu c Is the i-th dimensional value of (2);
the judging method comprises the following steps: and according to the formula, calculating the P 'of each category in the training sample set corresponding to the current target pixel, wherein the spectrum category corresponding to the maximum value of the P' is judged to be the spectrum category to which the current target pixel belongs.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710844657.XA CN107590472B (en) | 2017-09-19 | 2017-09-19 | Spectrum classifier and classification method based on push-broom spectrum imager |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710844657.XA CN107590472B (en) | 2017-09-19 | 2017-09-19 | Spectrum classifier and classification method based on push-broom spectrum imager |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107590472A CN107590472A (en) | 2018-01-16 |
| CN107590472B true CN107590472B (en) | 2024-04-05 |
Family
ID=61047355
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710844657.XA Active CN107590472B (en) | 2017-09-19 | 2017-09-19 | Spectrum classifier and classification method based on push-broom spectrum imager |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107590472B (en) |
Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE10019652A1 (en) * | 2000-02-20 | 2001-09-06 | Deutsch Zentr Luft & Raumfahrt | Colorimetric evaluation of remote reconnaissance sensors forming multi- and hyperspectral data involves integration over spectral range by summing over sensor-specific detection channels |
| CN1789933A (en) * | 2005-11-10 | 2006-06-21 | 中国科学院上海技术物理研究所 | Infrared static type high luminous flux Fourier transform imaging spectrometer |
| CN101770584A (en) * | 2009-12-30 | 2010-07-07 | 重庆大学 | Extraction method for identification characteristic of high spectrum remote sensing data |
| CN101916359A (en) * | 2005-01-27 | 2010-12-15 | 剑桥研究和仪器设备股份有限公司 | Methods and apparatus for classifying different parts of a sample into respective classes |
| CN103308466A (en) * | 2013-06-04 | 2013-09-18 | 沈阳仪表科学研究院有限公司 | Portable multispectral imaging system with light filter color wheel and spectral image processing method of multispectral imaging system |
| CN103400350A (en) * | 2013-07-22 | 2013-11-20 | 中国科学院西安光学精密机械研究所 | Spectral image restoration method of coded aperture spectral imager |
| KR101332717B1 (en) * | 2012-09-06 | 2013-11-25 | 인하대학교 산학협력단 | A method of spectral dimensionality reduction based on intergrated bispectrum phase for hyperspectral image analysis |
| CN104021396A (en) * | 2014-06-23 | 2014-09-03 | 哈尔滨工业大学 | Hyperspectral remote sensing data classification method based on ensemble learning |
| CN106228186A (en) * | 2016-07-20 | 2016-12-14 | 湖南大学 | Classification hyperspectral imagery apparatus and method |
| CN106447630A (en) * | 2016-09-18 | 2017-02-22 | 清华大学 | High-spectral image sharpening method based on probability matrix decomposition |
| CN106557784A (en) * | 2016-11-23 | 2017-04-05 | 上海航天控制技术研究所 | Fast target recognition methodss and system based on compressed sensing |
| CN106845517A (en) * | 2016-12-14 | 2017-06-13 | 西北大学 | Spectral tensor dimensionality reduction and classification method based on Tucker decomposition |
| CN107003785A (en) * | 2014-12-09 | 2017-08-01 | 巴斯夫欧洲公司 | Fluorescence detector |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8699889B2 (en) * | 2010-03-24 | 2014-04-15 | University Of Central Florida Research Foundation, Inc. | Polarization demultiplexing using independent component analysis |
| CN101893552B (en) * | 2010-07-06 | 2012-06-27 | 西安电子科技大学 | Hyperspectral imager and imaging method based on compressive sensing |
| CN102353449B (en) * | 2011-06-20 | 2014-11-12 | 中国科学院空间科学与应用研究中心 | Ultra-weak light multispectral imaging method and system |
| US9476824B2 (en) * | 2012-10-12 | 2016-10-25 | Purdue Research Foundation | Optical chemical classification |
-
2017
- 2017-09-19 CN CN201710844657.XA patent/CN107590472B/en active Active
Patent Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE10019652A1 (en) * | 2000-02-20 | 2001-09-06 | Deutsch Zentr Luft & Raumfahrt | Colorimetric evaluation of remote reconnaissance sensors forming multi- and hyperspectral data involves integration over spectral range by summing over sensor-specific detection channels |
| CN101916359A (en) * | 2005-01-27 | 2010-12-15 | 剑桥研究和仪器设备股份有限公司 | Methods and apparatus for classifying different parts of a sample into respective classes |
| CN1789933A (en) * | 2005-11-10 | 2006-06-21 | 中国科学院上海技术物理研究所 | Infrared static type high luminous flux Fourier transform imaging spectrometer |
| CN101770584A (en) * | 2009-12-30 | 2010-07-07 | 重庆大学 | Extraction method for identification characteristic of high spectrum remote sensing data |
| KR101332717B1 (en) * | 2012-09-06 | 2013-11-25 | 인하대학교 산학협력단 | A method of spectral dimensionality reduction based on intergrated bispectrum phase for hyperspectral image analysis |
| CN103308466A (en) * | 2013-06-04 | 2013-09-18 | 沈阳仪表科学研究院有限公司 | Portable multispectral imaging system with light filter color wheel and spectral image processing method of multispectral imaging system |
| CN103400350A (en) * | 2013-07-22 | 2013-11-20 | 中国科学院西安光学精密机械研究所 | Spectral image restoration method of coded aperture spectral imager |
| CN104021396A (en) * | 2014-06-23 | 2014-09-03 | 哈尔滨工业大学 | Hyperspectral remote sensing data classification method based on ensemble learning |
| CN107003785A (en) * | 2014-12-09 | 2017-08-01 | 巴斯夫欧洲公司 | Fluorescence detector |
| CN106228186A (en) * | 2016-07-20 | 2016-12-14 | 湖南大学 | Classification hyperspectral imagery apparatus and method |
| CN106447630A (en) * | 2016-09-18 | 2017-02-22 | 清华大学 | High-spectral image sharpening method based on probability matrix decomposition |
| CN106557784A (en) * | 2016-11-23 | 2017-04-05 | 上海航天控制技术研究所 | Fast target recognition methodss and system based on compressed sensing |
| CN106845517A (en) * | 2016-12-14 | 2017-06-13 | 西北大学 | Spectral tensor dimensionality reduction and classification method based on Tucker decomposition |
Non-Patent Citations (2)
| Title |
|---|
| Saadat Nasehi et al.A New Feature Dimensionally Reduction Approach Based on General Tensor Discriminant Analysis in EEG Signal Classification.《2011 International Conference on Intelligent Computation and Bio-Medical Instrumentation》.2012,188-191页. * |
| 刘敬.核Direct LDA子空间高光谱影像地物分类.《计算机科学》.2012,第39卷(第6期),274-277页. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107590472A (en) | 2018-01-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Kuo et al. | On data-driven saak transform | |
| US10776659B2 (en) | Systems and methods for compressing data | |
| Mahoor et al. | Facial action unit recognition with sparse representation | |
| US9160914B2 (en) | User control of the visual performance of a compressive imaging system | |
| US20170284867A1 (en) | Imaging method and apparatus | |
| US9025883B2 (en) | Adaptive quality image reconstruction via a compressed sensing framework | |
| Das | Hyperspectral image, video compression using sparse tucker tensor decomposition | |
| CN110650340A (en) | Space-time multiplexing compressed video imaging method | |
| US20240037709A1 (en) | Optical Method | |
| CN108387316A (en) | Video type self-adaptive target recognition device and method | |
| CN107590472B (en) | Spectrum classifier and classification method based on push-broom spectrum imager | |
| Wang et al. | Sparse representation-based hyperspectral data processing: Lossy compression | |
| JP2023092185A (en) | Image processing apparatus, learning method, and program | |
| Proskura et al. | Classification of Compressed Multichannel Images and Its Improvement | |
| CN114662315B (en) | Design and/or optimization method of nano-photonics device | |
| Gurram et al. | Ensemble learning based on multiple kernel learning for hyperspectral chemical plume detection | |
| Ahanonu et al. | Clustering Regression Wavelet Analysis for Lossless Compression of Hyperspectral Imagery. | |
| CN111243043B (en) | Hyperspectral compressed sensing method, hyperspectral compressed sensing equipment and hyperspectral compressed sensing system based on modified linear mixed model | |
| CN208270075U (en) | Video type self-adaptation target recognition device | |
| CN207302084U (en) | Spectrum classifier based on push-broom type spectrum imager | |
| Lv et al. | Hyperspectral compressive sensing | |
| Zhu et al. | Lossless Compression for Hyperspectral Images Using Cascaded Prediction | |
| CN112816068B (en) | Infrared spectrum imaging system and spectrum image reconstruction method thereof | |
| Nicolo | Homogeneous and heterogeneous face recognition: enhancing, encoding and matching for practical applications | |
| EP4430370A1 (en) | Hyperspectral imaging apparatus and methods |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20240801 Address after: Room A205-1, 2nd Floor, Building A, Collaborative Innovation Port R&D Office Building, Hongguang Avenue, Fengdong New City, Xixian New Area, Xi'an City, Shaanxi Province 710086 Patentee after: Shaanxi Zhuoke Aviation Micro Technology Co.,Ltd. Country or region after: China Address before: 710119, No. 17, information Avenue, new industrial park, hi tech Zone, Shaanxi, Xi'an Patentee before: XI'AN INSTITUTE OF OPTICS AND PRECISION MECHANICS OF CAS Country or region before: China |
|
| TR01 | Transfer of patent right |