CN111967182A - Hyperspectral modeling method based on mixed marks and used for spectral analysis - Google Patents
Hyperspectral modeling method based on mixed marks and used for spectral analysis Download PDFInfo
- Publication number
- CN111967182A CN111967182A CN202010722630.5A CN202010722630A CN111967182A CN 111967182 A CN111967182 A CN 111967182A CN 202010722630 A CN202010722630 A CN 202010722630A CN 111967182 A CN111967182 A CN 111967182A
- Authority
- CN
- China
- Prior art keywords
- hyperspectral
- image data
- model
- black
- primary
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 39
- 238000010183 spectrum analysis Methods 0.000 title claims description 16
- 238000001228 spectrum Methods 0.000 claims abstract description 38
- 230000003595 spectral effect Effects 0.000 claims abstract description 35
- 238000012549 training Methods 0.000 claims abstract description 25
- 238000002372 labelling Methods 0.000 claims abstract description 10
- 238000012216 screening Methods 0.000 claims abstract description 10
- 230000008569 process Effects 0.000 claims abstract description 8
- 238000010801 machine learning Methods 0.000 claims abstract description 6
- 238000012360 testing method Methods 0.000 claims abstract description 5
- 238000010606 normalization Methods 0.000 claims description 7
- 238000012545 processing Methods 0.000 claims description 7
- 239000013589 supplement Substances 0.000 claims description 5
- 239000003550 marker Substances 0.000 claims description 4
- 230000000694 effects Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 238000007500 overflow downdraw method Methods 0.000 description 4
- 230000006872 improvement Effects 0.000 description 3
- 238000000701 chemical imaging Methods 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000013499 data model Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000003912 environmental pollution Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 239000000047 product Substances 0.000 description 1
- 230000001502 supplementing effect Effects 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/17—Systems in which incident light is modified in accordance with the properties of the material investigated
- G01N21/25—Colour; Spectral properties, i.e. comparison of effect of material on the light at two or more different wavelengths or wavelength bands
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- General Health & Medical Sciences (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Investigating Or Analysing Materials By Optical Means (AREA)
Abstract
The invention discloses a method for establishing a primary hyperspectral model and a secondary hyperspectral model based on machine learning respectively; respectively establishing an original spectrum database and a secondary spectrum database; training a primary hyperspectral model by using an original spectral database, and reserving the comprehensive proportion of all classification conditions in the training process instead of directly determining the classification; marking unmarked hyperspectral image data by using the primary hyperspectral model, screening the hyperspectral image data marked by the primary hyperspectral model, and selecting 30-70% of more reliable hyperspectral image data according to the comprehensive proportion of all classification conditions reserved by the primary hyperspectral model to construct a secondary spectral database; and training the secondary hyperspectral model by taking the secondary spectral database as a training set and the original spectral database as a test set. The invention does not need huge spectrum storage space, fully utilizes the unlabeled spectrum and greatly reduces the labeling cost.
Description
Technical Field
The invention relates to the technical field of hyperspectral remote sensing, in particular to a hyperspectral modeling method based on mixed marks and used for spectral analysis.
Background
At present, a hyperspectral remote sensing technology, namely a hyperspectral resolution remote sensing technology, is a brand new remote sensing technology which is rapidly developed in recent years, is firstly applied to the fields of spaceflight, industry, geological mapping and the like, and is also widely applied to the fields of agriculture, environmental pollution detection, medical health and the like nowadays. The hyperspectral remote sensing technology mainly comprises the steps of collecting hyperspectral image data of a target area by using a hyperspectral imager, modeling by using an existing hyperspectral database and a proper modeling method, and predicting the collected hyperspectral image data by using an established data model so as to complete remote sensing of the target.
The existing modeling method mainly comprises multivariate linear regression analysis, minimum deviation-two-times regression and a support vector machine. Therefore, efficient processing of hyperspectral data is required during hyperspectral modeling. The hyperspectral image used by hyperspectral telemetry is a three-dimensional data cube formed by a spectrum dimension and a two-dimensional image, the requirement on the resolution of the hyperspectral spectrum dimension is high, the spatial resolution of the image also has great influence on telemetry precision, and therefore the hyperspectral data volume is extremely large, and the modeling time required by modeling by using various modeling methods is long.
The whole modeling process can be roughly divided into data acquisition, feature engineering, model training, model verification and online operation. Human time and effort tend to be mostly spent in the feature engineering section, where such enormous amounts of data may be disappointing to modelers when selecting and labeling training sets prior to modeling. Therefore, it is necessary to develop a suitable data preprocessing and feature screening method.
Disclosure of Invention
The invention provides a hyperspectral modeling method based on mixed labels for spectral analysis, which is used for solving the technical problems in the prior art.
The technical scheme adopted by the invention for solving the technical problems in the prior art is as follows: a hyperspectral modeling method based on mixed labels for spectral analysis is used for respectively establishing a primary hyperspectral model and a secondary hyperspectral model based on machine learning; respectively establishing an original spectrum database and a secondary spectrum database; training a primary hyperspectral model by using an original spectral database, and reserving the comprehensive proportion of all classification conditions in the training process instead of directly determining the classification; marking unmarked hyperspectral image data by using the primary hyperspectral model, screening the hyperspectral image data marked by the primary hyperspectral model, and selecting 30-70% of more reliable hyperspectral image data according to the comprehensive proportion of all classification conditions reserved by the primary hyperspectral model to construct a secondary spectral database; and training the secondary hyperspectral model by taking the secondary spectral database as a training set and the original spectral database as a test set.
Further, a hyperspectral camera and a black-and-white camera which are fixed in position are adopted, and hyperspectral image data and black-and-white image data are collected simultaneously; extracting specific details of the black-and-white image, finding out a mark with specificity in the black-and-white image, extracting high-frequency information in the black-and-white image, wherein the position of the high-frequency information in the black-and-white image corresponding to the original black-and-white image is the position of the specific mark; performing primary intensity calibration on a hyperspectral image acquired by a hyperspectral camera, performing intensity normalization processing on the positions of pixel points of each wave band, searching a corresponding position of a special mark in the hyperspectral image after the intensity normalization processing, solving the relative offset of the position of the special mark in the hyperspectral image and the position in a black-and-white image, solving a calibration parameter of the hyperspectral image relative to the black-and-white image by combining the lens axis included angles of the hyperspectral camera and the black-and-white camera, and fusing hyperspectral image data and black-and-white image data according to the calibration parameter; and storing the fused image data into an original spectrum database.
Further, residual errors of the spectral dimension and the spatial dimension of the hyperspectral image data are matched with residual errors in the black-and-white image data, details in the black-and-white image data are added into the hyperspectral image data, and the details of the hyperspectral image data are supplemented.
Further, high frequency information in the black-and-white image is extracted by performing two-dimensional fourier transform on the black-and-white image.
Furthermore, a snapshot-type hyperspectral camera based on the optical filter array is adopted to collect hyperspectral image data, and the total pixel number of the black and white camera is more than or equal to that of the hyperspectral camera.
Further, the condition for screening the hyperspectral image data marked by the primary hyperspectral model is as follows: the single classification exceeds t times the total case value, and t takes a value of 2 to 5.
Further, the method comprises the steps of:
step a, setting the accuracy of a primary hyperspectral model and a secondary hyperspectral model; establishing a standby database;
b, dividing the hyperspectral image data of the secondary spectral database into n groups, randomly distributing the hyperspectral image data of each group, performing n-fold modeling on the primary hyperspectral model and the secondary hyperspectral model, selecting a group of hyperspectral image data with the highest comprehensive accuracy of the primary hyperspectral model and the secondary hyperspectral model, and storing the group of hyperspectral image data into a standby database;
step c, judging whether the accuracy of the primary hyperspectral model and the secondary hyperspectral model reaches a set value, if so, turning to the step d, and if not, repeating the step b;
and d, combining the hyperspectral image data in all the standby databases with the hyperspectral image data in the original spectral database to form a training set of the mixed marker.
The invention has the advantages and positive effects that: the invention provides a hyperspectral modeling method based on mixed labeling for spectral analysis, which is used for respectively establishing a primary hyperspectral model and a secondary hyperspectral model based on machine learning, training the primary hyperspectral model and the secondary hyperspectral model in a self-learning confrontation mode, reserving the comprehensive proportion of all classification conditions, purposefully screening spectra with higher certainty, constructing a secondary spectral database, and repeatedly modeling until the comprehensive accuracy reaches a satisfactory value, thereby obtaining a model with better effect. The invention does not need huge spectrum storage space, fully utilizes the unlabeled spectrum and greatly reduces the labeling cost.
Detailed Description
To further understand the contents, features and effects of the present invention, the following examples are listed, and the following detailed descriptions are made:
a hyperspectral modeling method based on mixed labels for spectral analysis is used for respectively establishing a primary hyperspectral model and a secondary hyperspectral model based on machine learning; respectively establishing an original spectrum database and a secondary spectrum database; training a primary hyperspectral model by using an original spectral database, and reserving the comprehensive proportion of all classification conditions in the training process instead of directly determining the classification; marking unmarked hyperspectral image data by using the primary hyperspectral model, screening the hyperspectral image data marked by the primary hyperspectral model, and selecting 30-70% of more reliable hyperspectral image data according to the comprehensive proportion of all classification conditions reserved by the primary hyperspectral model to construct a secondary spectral database; and training the secondary hyperspectral model by taking the secondary spectral database as a training set and the original spectral database as a test set.
The high spectral image data marked by the primary high spectral model is screened, only the spectrum with high certainty is selected, the specific selection proportion can be freely selected, the general selection range can be 30-70%, the selection condition is the comprehensive proportion of all classification conditions, namely when the single classification exceeds t times of the total condition value, the spectrum is considered to be reliable, and the selection value of t is recommended to be 2-5 according to the input spectrum quality. And after selecting the reliable hyperspectral data, constructing the reliable hyperspectral data into a secondary spectrum database.
The primary hyperspectral model and the secondary hyperspectral model can be established based on various machine learning models suitable for spectral analysis, such as a PLS model, an SVM model, a convolutional neural network model, a cyclic neural network model and the like.
Further, a hyperspectral camera and a black-and-white camera which are fixed in position can be adopted to simultaneously acquire hyperspectral image data and black-and-white image data; the specific detail extraction can be carried out on the black-and-white image to find out the mark with specificity in the black-and-white image, the high-frequency information in the black-and-white image can be extracted, and the position of the high-frequency information in the black-and-white image corresponding to the position in the original black-and-white image is the position of the specific mark; the method comprises the steps of performing primary intensity calibration on a hyperspectral image acquired by a hyperspectral camera, performing intensity normalization processing on the positions of pixel points of each wave band, searching a position corresponding to a special mark in the hyperspectral image after the intensity normalization processing, solving the relative offset of the position of the special mark in the hyperspectral image and the position of the special mark in a black-and-white image, solving a calibration parameter of the hyperspectral image relative to the black-and-white image by combining the lens axis included angles of the hyperspectral camera and the black-and-white camera, and fusing hyperspectral image data and black-and-white image data according to the calibration parameter; and storing the fused image data into an original spectrum database. The invention fuses the hyperspectral image data collected by the snapshot hyperspectral camera of the spectral filtering array with the black and white image data collected by the black and white camera. And supplementing the hyperspectral image data with the black-and-white image data to improve the spatial resolution of the hyperspectral image.
The high frequency information in the black and white image can be extracted by performing a two-dimensional fourier transform on the black and white image.
The residual errors of the spectral dimension and the spatial dimension of the hyperspectral image data can be matched with the residual errors in the black-and-white image data respectively, and the details in the black-and-white image data can be added into the hyperspectral image data correspondingly to supplement the details of the hyperspectral image data. The method can uniformly process two image data, namely black-white image data and hyperspectral image data, by utilizing a multilevel residual error fusion method, can comprehensively consider residual errors of a spectral dimension and a spatial dimension, and adds details in the black-white image data to match and supplement the details of the hyperspectral image data.
The snapshot-type hyperspectral camera based on the optical filter array can be used for collecting hyperspectral image data, and the total pixel number of the black-and-white camera can be more than or equal to that of the hyperspectral camera.
The snapshot hyperspectral camera and the black and white camera can adopt applicable products in the prior art. The Snapshot-type hyperspectral camera can adopt a Mosaic Snapshot type Snapshot-type hyperspectral imaging camera produced by IMEC or a Snapshot-type multispectral hyperspectral imaging camera of CMV2000 series produced based on CMOSIS and the like.
Further, when the hyperspectral image data marked by the primary hyperspectral model is screened, the screening conditions can be as follows: the single classification exceeds t times the total case value, and t takes a value of 2 to 5. the higher the t value, the higher the classification accuracy, but the fewer the number of spectral classifications.
Further, a hyperspectral modeling method based on mixed labels for spectral analysis may include the steps of:
step a, setting the accuracy of a primary hyperspectral model and a secondary hyperspectral model; establishing a standby database;
b, dividing the hyperspectral image data of the secondary spectral database into n groups, randomly distributing the hyperspectral image data of each group, performing n-fold modeling on the primary hyperspectral model and the secondary hyperspectral model, selecting a group of hyperspectral image data with the highest comprehensive accuracy of the primary hyperspectral model and the secondary hyperspectral model, and storing the group of hyperspectral image data into a standby database;
step c, judging whether the accuracy of the primary hyperspectral model and the secondary hyperspectral model reaches a set value, if so, turning to the step d, and if not, repeating the step b;
and d, combining the hyperspectral image data in all the standby databases with the hyperspectral image data in the original spectral database to form a training set of the mixed marker.
The working process and working principle of the present invention are further described below by taking a preferred embodiment of the present invention as an example:
firstly, the spatial resolution of the hyperspectral data needs to be improved. The method adopts a snapshot type hyperspectral camera based on a light filter array to acquire hyperspectral image data, adopts a black-and-white camera to acquire black-and-white image data, has a relative position regulation between a lens of the black-and-white camera and a lens of the hyperspectral camera, has an angle between the axes of the two lenses, and can ensure that the total pixel count of the black-and-white camera is not less than that of the hyperspectral camera, and adopts the black-and-white image data to supplement the hyperspectral image data.
The spectral filter array unit cell of the snapshot type hyperspectral camera based on the filter array is 4 x 4 or 5 x 5. The total pixel points of the black and white camera can be equal to or higher than the total pixel points of the snapshot-type hyperspectral camera, and the spatial resolution of the demosaic hyperspectral image finally obtained is mainly determined by the spatial resolution of the black and white camera.
The invention fuses the high spectrum image data collected by the snapshot high spectrum camera of the spectrum filtering array and the black and white image data collected by the black and white camera, and the fusion method is as follows: because the system is a fixing device, namely the angle between the black-and-white camera and the hyperspectral camera is fixed, the distance position of the shot object is not fixed, and therefore image alignment cannot be directly realized in a calculation mode. The method comprises the steps of firstly, extracting specific details of a black-and-white picture to find a mark with specificity in the picture, wherein the specific implementation mode is that two-dimensional Fourier transform is carried out on the picture to extract high-frequency information in the picture, and the position of the high-frequency information in the picture, which corresponds to an original picture, is the position of the special mark; and performing primary intensity calibration on the image acquired by the hyperspectral camera, performing intensity normalization on the positions of the pixel points of each wave band, searching the corresponding positions of the special marks in the normalized image, calculating image alignment parameters according to the offset distance and the included angle of the two cameras, and aligning the hyperspectral image and the black and white image.
The invention provides a multilevel residual error fusion method for improving the super-resolution of a hyperspectral image and a black-and-white image. The common residual fusion method directly uses hyperspectrum as a set of two-dimensional gray level images, super-resolution improvement is respectively carried out, the correlation of the spectral dimensions of the hyperspectral images is ignored, and the effect is limited. The method simultaneously considers super-resolution improvement of the black-white image and the hyperspectral image, uniformly processes databases of the black-white image and the hyperspectral image by utilizing a multi-level method, and the super-resolution improvement of the hyperspectral image mainly aims to enhance the matching rate with the black-white image instead of improving the effect, comprehensively considers the residual error between the spectral dimension and the spatial dimension, supplements details and matches the details added in the black-white image.
The method of the invention can be divided into two parts, namely spectral marking and secondary modeling. Firstly, a primary hyperspectral model and a secondary hyperspectral model are established, a large amount of unlabeled hyperspectral data are labeled by using the primary hyperspectral model, the method is different from the common modeling discrimination classification, and the step requires that the comprehensive proportion of all classification conditions is kept instead of directly determining the classification. Secondly, screening a large amount of unlabeled hyperspectral data, wherein all spectra in a database cannot be used in secondary classification, in order to ensure the reliability of the model, the method only selects spectra with high certainty, the specific selection proportion can be freely selected, the general selection range can be 30-70%, the selection condition is the comprehensive proportion of all classification conditions, namely when the single classification exceeds t times of the total condition value, the spectrum is considered to be reliable, and the selection value of t is recommended to be 2-5 according to the input spectrum quality. After selecting more reliable hyperspectral data, constructing the more reliable hyperspectral data into a secondary spectrum database, dividing the secondary spectrum database into n parts, wherein n is a self-set parameter, and recommending that the number of spectra in each part is approximately the same as the hyperspectral data in an original database used during the training of a primary hyperspectral model. And taking the secondary spectrum database as a training set and the original database as a test set, performing n-fold cross validation on the primary hyperspectral model and the secondary hyperspectral model, selecting a group of hyperspectral data with the highest model accuracy, recording the group of hyperspectral data, and storing the hyperspectral data in a standby database. And randomly distributing the spectrums in the secondary spectrum database, dividing the spectrums into n parts again, repeating the previous steps, storing the spectrum with the highest modeling accuracy each time until the comprehensive accuracy reaches a satisfactory value, combining all the hyperspectral data stored in the standby database with the original spectrum database to form a training set for modeling finally, wherein the model is the model based on the hybrid marker hyperspectral modeling method.
The above-mentioned embodiments are only for illustrating the technical ideas and features of the present invention, and the purpose thereof is to enable those skilled in the art to understand the contents of the present invention and to carry out the same, and the present invention shall not be limited to the embodiments, i.e. the equivalent changes or modifications made within the spirit of the present invention shall fall within the scope of the present invention.
Claims (7)
1. A hyperspectral modeling method based on mixed labeling for spectral analysis is characterized in that a primary hyperspectral model and a secondary hyperspectral model based on machine learning are respectively established; respectively establishing an original spectrum database and a secondary spectrum database; training a primary hyperspectral model by using an original spectral database, and reserving the comprehensive proportion of all classification conditions in the training process instead of directly determining the classification; marking unmarked hyperspectral image data by using the primary hyperspectral model, screening the hyperspectral image data marked by the primary hyperspectral model, and selecting 30-70% of more reliable hyperspectral image data according to the comprehensive proportion of all classification conditions reserved by the primary hyperspectral model to construct a secondary spectral database; and training the secondary hyperspectral model by taking the secondary spectral database as a training set and the original spectral database as a test set.
2. The hyperspectral modeling method based on hybrid labeling for spectral analysis according to claim 1, characterized by employing a hyperspectral camera and a black-and-white camera whose positions are fixed to each other, and simultaneously acquiring hyperspectral image data and black-and-white image data; extracting specific details of the black-and-white image, finding out a mark with specificity in the black-and-white image, extracting high-frequency information in the black-and-white image, wherein the position of the high-frequency information in the black-and-white image corresponding to the original black-and-white image is the position of the specific mark; performing primary intensity calibration on a hyperspectral image acquired by a hyperspectral camera, performing intensity normalization processing on the positions of pixel points of each wave band, searching a corresponding position of a special mark in the hyperspectral image after the intensity normalization processing, solving the relative offset of the position of the special mark in the hyperspectral image and the position in a black-and-white image, solving a calibration parameter of the hyperspectral image relative to the black-and-white image by combining the lens axis included angles of the hyperspectral camera and the black-and-white camera, and fusing hyperspectral image data and black-and-white image data according to the calibration parameter; and storing the fused image data into an original spectrum database.
3. The hyperspectral modeling method based on hybrid labeling for spectral analysis according to claim 2, wherein the residual errors in the spectral dimension and the spatial dimension of the hyperspectral image data are matched with the residual errors in the black-and-white image data, and the details in the black-and-white image data are added to the hyperspectral image data to supplement the details of the hyperspectral image data.
4. The hybrid label based hyperspectral modeling method for spectral analysis according to claim 2, wherein the high frequency information in the black-and-white image is extracted by performing a two-dimensional fourier transform on the black-and-white image.
5. The hyperspectral modeling method based on hybrid labeling for spectral analysis according to claim 2, characterized in that a snapshot hyperspectral camera based on a light filter array is used to collect hyperspectral image data, and the total number of pixels of a black and white camera is greater than or equal to the total number of pixels of the hyperspectral camera.
6. The hyperspectral modeling method based on hybrid labeling for spectral analysis according to claim 1, wherein the conditions for screening the hyperspectral image data after labeling the primary hyperspectral model are as follows: the single classification exceeds t times the total case value, and t takes a value of 2 to 5.
7. The hyperspectral modeling method based on mixed labels for spectral analysis according to claim 1, characterized by comprising the steps of:
step a, setting the accuracy of a primary hyperspectral model and a secondary hyperspectral model; establishing a standby database;
b, dividing the hyperspectral image data of the secondary spectral database into n groups, randomly distributing the hyperspectral image data of each group, performing n-fold modeling on the primary hyperspectral model and the secondary hyperspectral model, selecting a group of hyperspectral image data with the highest comprehensive accuracy of the primary hyperspectral model and the secondary hyperspectral model, and storing the group of hyperspectral image data into a standby database;
step c, judging whether the accuracy of the primary hyperspectral model and the secondary hyperspectral model reaches a set value, if so, turning to the step d, and if not, repeating the step b;
and d, combining the hyperspectral image data in all the standby databases with the hyperspectral image data in the original spectral database to form a training set of the mixed marker.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010722630.5A CN111967182B (en) | 2020-07-24 | 2020-07-24 | Mixed-label-based hyperspectral modeling method for spectrum analysis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010722630.5A CN111967182B (en) | 2020-07-24 | 2020-07-24 | Mixed-label-based hyperspectral modeling method for spectrum analysis |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111967182A true CN111967182A (en) | 2020-11-20 |
| CN111967182B CN111967182B (en) | 2024-04-02 |
Family
ID=73362472
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010722630.5A Active CN111967182B (en) | 2020-07-24 | 2020-07-24 | Mixed-label-based hyperspectral modeling method for spectrum analysis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111967182B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080046217A1 (en) * | 2006-02-16 | 2008-02-21 | Clean Earth Technologies, Llc | Method for Spectral Data Classification and Detection in Diverse Lighting Conditions |
| WO2018045626A1 (en) * | 2016-09-07 | 2018-03-15 | 深圳大学 | Super-pixel level information fusion-based hyperspectral image classification method and system |
| WO2019090509A1 (en) * | 2017-11-08 | 2019-05-16 | 深圳大学 | Hyperspectral image classification method and system |
| CN110298396A (en) * | 2019-06-25 | 2019-10-01 | 北京工业大学 | Hyperspectral image classification method based on deep learning multiple features fusion |
| CN110414438A (en) * | 2019-07-30 | 2019-11-05 | 西南交通大学 | High spectrum image recognition methods based on space clustering Information revision |
-
2020
- 2020-07-24 CN CN202010722630.5A patent/CN111967182B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080046217A1 (en) * | 2006-02-16 | 2008-02-21 | Clean Earth Technologies, Llc | Method for Spectral Data Classification and Detection in Diverse Lighting Conditions |
| WO2018045626A1 (en) * | 2016-09-07 | 2018-03-15 | 深圳大学 | Super-pixel level information fusion-based hyperspectral image classification method and system |
| WO2019090509A1 (en) * | 2017-11-08 | 2019-05-16 | 深圳大学 | Hyperspectral image classification method and system |
| CN110298396A (en) * | 2019-06-25 | 2019-10-01 | 北京工业大学 | Hyperspectral image classification method based on deep learning multiple features fusion |
| CN110414438A (en) * | 2019-07-30 | 2019-11-05 | 西南交通大学 | High spectrum image recognition methods based on space clustering Information revision |
Non-Patent Citations (1)
| Title |
|---|
| 刘舒扬;张晨;赵安娜;李奇峰;贾晓东;: "基于光谱特性的高判别准确度激光近场探测传感器研究", 红外与激光工程, no. 04 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111967182B (en) | 2024-04-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Liu et al. | YOLOv5-Tassel: Detecting tassels in RGB UAV imagery with improved YOLOv5 based on transfer learning | |
| CN108537742B (en) | Remote sensing image panchromatic sharpening method based on generation countermeasure network | |
| Tuia et al. | Domain adaptation for the classification of remote sensing data: An overview of recent advances | |
| Fu et al. | Joint camera spectral sensitivity selection and hyperspectral image recovery | |
| Karami et al. | Automatic plant counting and location based on a few-shot learning technique | |
| CN109558823B (en) | Vehicle identification method and system for searching images by images | |
| CN109583324A (en) | A kind of pointer meters reading automatic identifying method based on the more box detectors of single-point | |
| Tuia et al. | Recent advances in domain adaptation for the classification of remote sensing data | |
| CN112712049B (en) | Satellite image ship model identification method under small sample condition | |
| CN103559499B (en) | RGB vector matching rapid-recognition system and method | |
| CN107341538A (en) | A kind of statistical magnitude method of view-based access control model | |
| CN103591940A (en) | Method of evaluating confidence of matching signature of hyperspectral image | |
| Yuan et al. | Translation, scale and rotation: cross-modal alignment meets RGB-infrared vehicle detection | |
| Zheng et al. | Multisource-domain generalization-based oil palm tree detection using very-high-resolution (vhr) satellite images | |
| Jiang et al. | Local and global structure for urban ALS point cloud semantic segmentation with ground-aware attention | |
| Liu et al. | Deep domain adaptation for pavement crack detection | |
| CN115188066A (en) | Moving target detection system and method based on cooperative attention and multi-scale fusion | |
| He et al. | CSiT: A multiscale vision transformer for hyperspectral image classification | |
| Jaffe et al. | Remote sensor design for visual recognition with convolutional neural networks | |
| CN112418262A (en) | Vehicle re-identification method, client and system | |
| CN111967182A (en) | Hyperspectral modeling method based on mixed marks and used for spectral analysis | |
| Fu et al. | Complementarity-aware Local-global Feature Fusion Network for Building Extraction in Remote Sensing Images | |
| CN116597246A (en) | Model training method, target detection method, electronic device and storage medium | |
| CN113792609A (en) | Method for detecting image abnormity through aligning multi-spectrum cross-domain features in automatic driving | |
| Wang et al. | Stereo Rectification Based on Epipolar Constrained Neural Network |
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 |