CN116306819B - Hyperspectral cross calibration method and device based on spectrum reconstruction and electronic equipment - Google Patents
Hyperspectral cross calibration method and device based on spectrum reconstruction and electronic equipment Download PDFInfo
- Publication number
- CN116306819B CN116306819B CN202310286996.6A CN202310286996A CN116306819B CN 116306819 B CN116306819 B CN 116306819B CN 202310286996 A CN202310286996 A CN 202310286996A CN 116306819 B CN116306819 B CN 116306819B
- Authority
- CN
- China
- Prior art keywords
- hyperspectral
- band
- neural network
- network model
- data
- 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
- 238000000034 method Methods 0.000 title claims abstract description 37
- 238000001228 spectrum Methods 0.000 title claims abstract description 23
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 38
- 238000003062 neural network model Methods 0.000 claims abstract description 32
- 230000007787 long-term memory Effects 0.000 claims abstract description 16
- 230000007246 mechanism Effects 0.000 claims abstract description 16
- 238000012549 training Methods 0.000 claims abstract description 15
- 230000015654 memory Effects 0.000 claims description 40
- 239000011159 matrix material Substances 0.000 claims description 30
- 230000006870 function Effects 0.000 claims description 15
- 238000004422 calculation algorithm Methods 0.000 claims description 11
- 238000012417 linear regression Methods 0.000 claims description 9
- 230000003595 spectral effect Effects 0.000 claims description 9
- 238000013528 artificial neural network Methods 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000000605 extraction Methods 0.000 claims description 6
- 230000009467 reduction Effects 0.000 claims description 6
- 238000012216 screening Methods 0.000 claims description 5
- 238000011022 operating instruction Methods 0.000 claims 1
- 230000006403 short-term memory Effects 0.000 abstract description 7
- 230000008569 process Effects 0.000 description 7
- 230000005855 radiation Effects 0.000 description 4
- 238000004590 computer program Methods 0.000 description 3
- 230000003287 optical effect Effects 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000003068 static effect Effects 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000000802 evaporation-induced self-assembly Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000002310 reflectometry Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
- G01N21/27—Colour; Spectral properties, i.e. comparison of effect of material on the light at two or more different wavelengths or wavelength bands using photo-electric detection ; circuits for computing concentration
- G01N21/274—Calibration, base line adjustment, drift correction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/10—Interfaces, programming languages or software development kits, e.g. for simulating neural networks
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A40/00—Adaptation technologies in agriculture, forestry, livestock or agroalimentary production
- Y02A40/10—Adaptation technologies in agriculture, forestry, livestock or agroalimentary production in agriculture
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- General Physics & Mathematics (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Image Processing (AREA)
Abstract
The invention belongs to the technical field of spectrum calibration, and relates to a hyperspectral cross calibration method and device based on spectrum reconstruction, and electronic equipment, wherein the method comprises the following steps: establishing a convolutional neural network model with a multi-head attention mechanism; selecting hyperspectral wave band data of a target wave band; establishing a long-term and short-term memory neural network model, and training the model; reconstructing hyperspectral band data to obtain hyperspectral data of a target band; cross-scaling is performed using target band hyperspectral data. The method effectively solves the problem that the traditional cross calibration requires similar spectrum bands between the remote sensor to be calibrated and the target remote sensor, trains the convolutional neural network model with a multi-head attention mechanism to enable the convolutional neural network model to have good band selection capacity and screen out a band with stronger representativeness in hyperspectral data, performs spectrum reconstruction by utilizing a long-term and short-term memory network after band selection, fits out required band information, and further meets the requirement of cross calibration.
Description
Technical Field
The invention belongs to the technical field of spectrum calibration, and particularly relates to a method and a device for cross calibration based on spectrum reconstruction and electronic equipment.
Background
The trend of current remote sensing applications gradually goes from qualitative to quantitative, which puts higher demands on the reliability of radiation calibration. Although the traditional site replacement calibration method has higher calibration precision, the traditional site replacement calibration method has a plurality of limitations and is difficult to develop in a large scale and high frequency. Cross scaling is a field-free scaling technique that has developed relatively rapidly in recent years and which solves the problem of a high field scaling threshold.
The idea of cross calibration is that when the in-orbit satellite remote sensor to be calibrated and the satellite remote sensor with better calibration result observe the same area at the same time and similar angles, the calibration of the satellite remote sensor to be calibrated can be realized by comparing the measured values of the in-orbit satellite remote sensor to be calibrated and the satellite remote sensor with better calibration result. Compared with the field replacement calibration technology, the method can calibrate the satellite data of multiple remote sensors without establishing a ground correction field. Its advantages are low cost and high-frequency radiation calibration between multiple remote sensors. Liu Jiaxin et al used MODIS satellites in 2021 to cross-scale the thermal infrared channels of China's wind cloud satellite No. 4. While the constraints of cross-scaling are reduced compared to site scaling, cross-scaling requires that the remote sensor to be scaled have a similar spectral range as the target remote sensor, which results in some remote sensors to be scaled lacking available target remote sensors with similar spectral ranges for reference.
Disclosure of Invention
In order to solve the technical problems, the invention provides a hyperspectral cross calibration method and device based on spectrum reconstruction and electronic equipment.
In a first aspect, the present invention provides a hyperspectral cross-scaling method based on spectral reconstruction, comprising:
Establishing a convolutional neural network model with a multi-head attention mechanism;
selecting hyperspectral wave band data of a target wave band by using the convolutional neural network model;
Establishing a long-term memory neural network model, and training the long-term memory neural network model by adopting a back propagation algorithm;
reconstructing the hyperspectral band data of the selected target band by using the trained long-short-term memory neural network model to obtain reconstructed hyperspectral data of the target band;
and performing cross scaling by using the target band hyperspectral data.
In a second aspect, the invention provides a hyperspectral cross scaling device based on spectrum reconstruction, which comprises a model building unit, a selecting unit, a model training unit, a reconstruction unit and a cross scaling unit;
The model building unit is used for building a convolutional neural network model with a multi-head attention mechanism;
the selection unit is used for selecting hyperspectral wave band data of a target wave band by utilizing the convolutional neural network model;
the model training unit is used for building a long-period memory neural network model and training the long-period memory neural network model by adopting a back propagation algorithm;
The reconstruction unit is used for reconstructing the hyperspectral band data of the selected target band by using the trained long-short-term memory neural network model to obtain reconstructed hyperspectral data of the target band;
The cross scaling unit is used for cross scaling by using the target band hyperspectral data.
In a third aspect, the present invention provides an electronic device, comprising:
A processor and a memory;
the memory is used for storing computer operation instructions;
the processor is used for executing the hyperspectral cross scaling method based on spectrum reconstruction by calling the computer operation instruction.
The beneficial effects of the invention are as follows: the method effectively solves the problem that the traditional cross calibration requires similar spectrum bands between the remote sensor to be calibrated and the target remote sensor, trains the convolutional neural network model with a multi-head attention mechanism to enable the convolutional neural network model to have good band selection capacity and screen out a band with stronger representativeness in hyperspectral data, performs spectrum reconstruction by utilizing a long-term and short-term memory network after band selection, fits out required band information, and further meets the requirement of cross calibration.
On the basis of the technical scheme, the invention can be improved as follows.
Furthermore, a plurality of multi-head attention modules and a full connection layer are adopted as the basic structure of the long-short-period memory neural network model, and a step function is added as a filter during output.
Further, selecting hyperspectral band data of a target band using the convolutional neural network model includes: performing feature extraction on the hyperspectral wave band data to obtain a weight matrix of a target wave band; and screening the weight matrix of the target wave band, and reserving the hyperspectral wave band with the largest weight by using a step function to obtain the hyperspectral wave band data of the target wave band.
And when the trained long-short-term memory neural network model is used for reconstructing the hyperspectral wave band data of the selected target wave band, the radiance of each wave band is used as one-dimensional characteristics, and the filtered wave band set is used as a sequence to perform long-short-term memory neural network calculation.
Further, updating the parameters of the convolutional neural network model by adopting a back propagation algorithm, and calculating the mean square error between an original matrix and a matrix reconstructed by using the convolutional neural network model; and determining a wave band meeting the reconstruction of the original full-wave band information according to the mean square error.
Further, cross-scaling using the target band hyperspectral data, comprising:
After the reconstructed target band hyperspectral data are obtained, one-dimensional radiance information is expanded and calculated to obtain a multidimensional tensor;
Performing dimension reduction operation on the tensor by using two linear regression networks, and fitting the reconstructed wave band information to obtain radiance;
Inputting data into the trained long-short-period memory neural network model, and fitting out the radiance of a missing wave band;
and taking the fitted wave band radiance as a target value, and calculating a scaling coefficient by combining with the DN value of the remote sensor to be scaled to finish cross scaling.
Drawings
FIG. 1 is a flow chart of a hyperspectral cross scaling method based on spectral reconstruction provided in example 1 of the present invention;
FIG. 2 is a schematic diagram of a hyperspectral cross-scaling apparatus based on spectral reconstruction in accordance with the present invention;
Fig. 3 is a schematic diagram of an electronic device in embodiment 3 of the present invention.
Icon: 30-an electronic device; 310-a processor; 320-bus; 330-memory; 340-transceiver.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments of the present invention. The components of the embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations.
Example 1
As an embodiment, as shown in fig. 1, to solve the above technical problem, the present embodiment provides a hyperspectral cross scaling method based on spectrum reconstruction, including:
Establishing a convolutional neural network model with a multi-head attention mechanism;
selecting hyperspectral wave band data of a target wave band by using a convolutional neural network model;
establishing a long-term memory neural network model, and training the long-term memory neural network model by adopting a back propagation algorithm;
reconstructing hyperspectral band data of the selected target band by using the trained long-term and short-term memory neural network model to obtain reconstructed hyperspectral band data of the target band;
cross-scaling is performed using target band hyperspectral data.
In the practical application process, a convolutional neural network model with a multi-head attention mechanism is established, and the function of band selection is realized through model training.
Optionally, a plurality of multi-head attention modules and a full connection layer are adopted as the basic structure of the long-short-period memory neural network model, and a step function is added as a filter during output.
In the practical application process, when a convolutional neural network model with a multi-head attention mechanism is established, the method specifically comprises the following steps:
s11: adopting 64 multi-head attention modules and a full connection layer as basic structures of a model to be trained, and adding a step function as a filter during output;
s12: constructing a convolutional neural network structure, and reconstructing an original matrix by using the filtered low-rank matrix by using multi-layer convolution;
S13: and updating network parameters by adopting a back propagation algorithm, calculating the mean square error between the original matrix and the matrix reconstructed by using the convolutional neural network, and analyzing the difference between the original matrix and the reconstructed matrix, wherein the smaller the mean square error is, the smaller the difference is. ; and selecting hyperspectral wave band data of a target wave band by using a convolutional neural network model with a multi-head attention mechanism, wherein the smaller the difference is, the more successful the wave band is selected, otherwise, the larger the difference is, the less representative the selected wave band is and the redundancy is high.
The characteristic extraction capacity of the convolutional neural network is enhanced by adopting an attention mechanism, a large number of redundant hyperspectral wave bands are subjected to characteristic extraction, a weight matrix of the wave bands is obtained, the wave band with the most representativeness is selected, and a ij is set as attention distribution; softmax is the activation function; s (k j,qi) is a scoring function for attention; exp (s (k j,qi)) is an exponential mapping of the attention scoring function; context i is input data; the attention mechanism formula is as follows:
When the attention network is used to weight the band matrix, the size of the input data X is not changed, and only the selected band value is 1 in the tensor X with the size of h×w×s, and the other unselected band values are 0. Let h i be the ith attention head; w i (q) is a learnable parameter of query q; w i (k) is a learnable parameter for key k; w i (v) is a learnable parameter of value v; r Pv is the domain of action of attention; the multi-head attention algorithm formula is as follows:
hi=f(Wi (q)q,Wi (k)k,Wi (v)v)∈RPv;
H i is the ith attention head, q is the query, k is the key, v is the value, linear conversion is needed after the output of a plurality of attention heads is obtained, and the results of the plurality of attention heads are spliced according to the following formula:
Optionally, selecting hyperspectral band data of the target band by using the convolutional neural network model includes: performing feature extraction on hyperspectral wave band data to obtain a weight matrix of a target wave band; screening the weight matrix of the target wave band, and reserving the hyperspectral wave band with the largest weight by using a step function to obtain hyperspectral wave band data of the target wave band.
In the practical application process, a convolutional neural network model with a multi-head attention mechanism is used for weighting a wave band matrix, a 1 value is given to a wave band with strong representativeness, a 0 value is given to a redundant wave band, then the selected wave band matrix is used for restoring a complete original matrix, when the selected wave band matrix can approximately express the original matrix, the model can be considered to be converged, and the wave band selection result is effective and strong representativeness. And screening the obtained weight matrix, and reserving hyperspectral wave bands with larger weight and stronger representing capacity by using a step function. Only selected bands are used in the process of spectral reconstruction using a long and short-term memory network to avoid redundancy and excessive time complexity.
Optionally, when the trained long-short-term memory neural network model is used for reconstructing hyperspectral band data of the selected target band, the radiance of each band is used as one-dimensional characteristics, and the filtered band set is used as a sequence to perform long-short-term memory neural network calculation.
In the actual application process, I t is set as an input gate at the time t; f t is a forgetting door at the moment t; o t is an output gate at time t; x t is the input value of the network at the current moment; w xi is a learnable parameter of the current moment of the input door; w xf is a learnable parameter of the current moment of the forgetting door; w xo is a learnable parameter of the current moment of the output door; h t-1 is the output value of LSTM at the previous time; w hi is a learnable parameter input by the input gate once; w hf is a learnable parameter input by the forget gate once; w ho is a learnable parameter input by the forget gate once; b i is the bias term of the input gate; b f is a bias term of the forget gate; b o is the bias term of the output gate; sigma is; the long-term memory network formula is as follows:
It=σ(XtWxi+Ht-1Whi+bi);
Ft=σ(XtWxf+Ht-1Whf+bf);
Ot=σ(XtWxo+Ht-1Who+bo);
Optionally, updating the parameters of the convolutional neural network model, and calculating the mean square error between the original matrix and the matrix reconstructed by using the convolutional neural network model; the band satisfying the original full band information is determined according to the mean square error.
Optionally, cross-scaling using target band hyperspectral data includes:
After the reconstructed target band hyperspectral data are obtained, one-dimensional radiance information is expanded and calculated to obtain a multidimensional tensor;
performing dimension reduction operation on tensors by using two linear regression networks, and fitting the reconstructed wave band information to obtain radiance;
inputting data into a trained long-short-period memory neural network model, and fitting out the radiance of a missing wave band;
And taking the fitted wave band radiance as a target value, and calculating a calibration coefficient by combining a DN value (Digital Number) of the remote sensing device to be calibrated, so as to finish cross calibration.
In the practical application process, after the calculation of the long-term and short-term memory neural network, the characteristic dimension of the data can be changed, and linear regression is needed, and optionally, the specific process of the linear regression is as follows:
after the reconstructed target band hyperspectral data are obtained, setting the dimension number of output features as 64 and setting a hidden layer as 3;
Reducing the dimension of the 64-dimensional features by using a linear regression mode, wherein the number of the feature dimensions after the dimension reduction is 8, and then reducing the feature dimensions to 1 by using a linear regression mode again;
The reduced tensor format is (n, 1), where n is the number of input bands, and the tensor is deformed into (1, 1) using a linear connection;
And finally obtaining the radiance value of the target wave band through three times of calculation and using the radiance value for cross calibration.
The following will simulate hyperspectral datasets using two sets of datasets for training a neural network:
JPL (Jet Propulsion Laboratory) hyperspectral dataset
The spectrum characteristics of the ground features are the interaction results of electromagnetic waves and the surface of the ground features, and are often expressed as reflection, heat radiation, microwave radiation and scattering characteristics of the ground features in different wave bands, the ground feature types are screened from a spectrum database, then the ground feature reflectivity is converted into ground feature emissivity, atmospheric profile data are obtained from a thermodynamic database, and finally the screened data are calculated through MODTRA simulation software to obtain corresponding radiance.
Tigr (The Thermodynamic Initial Guess Retrieval) atmospheric profile dataset
The TIGR atmospheric profile data set is an atmospheric profile database established by French dynamic weather laboratories and is mainly used as an initial estimate of atmospheric profile inversion, and comprises atmospheric temperature and humidity profiles which are carefully selected from a large number of atmospheric samples in different periods and global ranges by using a topology method, belong to sounding observation, and are distributed in all places of the world in 2311 total atmospheric profiles. 716 sunny open lines are selected as the atmospheric profile data in the method.
The method effectively solves the problem that the traditional cross calibration requires similar spectrum bands between the remote sensor to be calibrated and the target remote sensor, trains the convolutional neural network model with a multi-head attention mechanism to enable the convolutional neural network model to have good band selection capacity and screen out a band with stronger representativeness in hyperspectral data, performs spectrum reconstruction by utilizing a long-term and short-term memory network after band selection, fits out required band information, and further meets the requirement of cross calibration.
Example 2
Based on the same principle as the method shown in the embodiment 1 of the present invention, as shown in fig. 2, the embodiment of the present invention further provides a hyperspectral cross scaling device based on spectrum reconstruction, which comprises a model building unit, a selecting unit, a model training unit, a reconstruction unit and a cross scaling unit;
the model building unit is used for building a convolutional neural network model with a multi-head attention mechanism;
The selection unit is used for selecting hyperspectral wave band data of a target wave band by using the convolutional neural network model;
the model training unit is used for building a long-term memory neural network model and training the long-term memory neural network model by adopting a back propagation algorithm;
The reconstruction unit is used for reconstructing hyperspectral band data of the selected target band by using the trained long-short-term memory neural network model to obtain reconstructed hyperspectral band data of the target band;
And the cross scaling unit is used for cross scaling by using the hyperspectral data of the target wave band.
Optionally, a plurality of multi-head attention modules and a full connection layer are adopted as the basic structure of the long-short-period memory neural network model, and a step function is added as a filter during output.
Optionally, selecting hyperspectral band data of the target band by using the convolutional neural network model includes: performing feature extraction on hyperspectral wave band data to obtain a weight matrix of a target wave band; screening the weight matrix of the target wave band, and reserving the hyperspectral wave band with the largest weight by using a step function to obtain hyperspectral wave band data of the target wave band.
Optionally, when the trained long-short-term memory neural network model is used for reconstructing hyperspectral band data of the selected target band, the radiance of each band is used as one-dimensional characteristics, and the filtered band set is used as a sequence to perform long-short-term memory neural network calculation.
Optionally, updating the model parameters of the convolutional neural network by adopting a back propagation algorithm, and calculating the mean square error between the original matrix and the matrix reconstructed by using the convolutional neural network model; the band satisfying the original full band information is determined according to the mean square error.
Optionally, cross-scaling using target band hyperspectral data includes:
After the reconstructed target band hyperspectral data are obtained, one-dimensional radiance information is expanded and calculated to obtain a multidimensional tensor;
performing dimension reduction operation on tensors by using two linear regression networks, and fitting the reconstructed wave band information to obtain radiance;
inputting data into a trained long-short-period memory neural network model, and fitting out the radiance of a missing wave band;
and taking the fitted wave band radiance as a target value, and calculating a scaling coefficient by combining with the DN value of the remote sensor to be scaled to finish cross scaling.
Example 3
Based on the same principle as the method shown in the embodiment of the present invention, there is also provided in the embodiment of the present invention an electronic device, as shown in the accompanying drawings, which may include, but is not limited to: a processor and a memory; a memory for storing a computer program; a processor for performing the method of any of the embodiments of the invention by invoking a computer program.
In an alternative embodiment, an electronic device is provided, the electronic device 30 shown in fig. 3 comprising: a processor 310 and a memory 330. Wherein the processor 310 is coupled to the memory 330, such as via a bus 320.
Optionally, the electronic device 30 may further comprise a transceiver 340, and the transceiver 340 may be used for data interaction between the electronic device and other electronic devices, such as transmission of data and/or reception of data, etc. It should be noted that, in practical applications, the transceiver 340 is not limited to one, and the structure of the electronic device 30 is not limited to the embodiment of the present invention.
The processor 310 may be a CPU central processing unit, a general purpose processor, a DSP data signal processor, an ASIC specific integrated circuit, an FPGA field programmable gate array or other programmable logic device, a hardware component, or any combination thereof. Processor 310 may also be a combination that performs computing functions, e.g., including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.
Bus 320 may include a path that communicates information between the components. Bus 320 may be a PCI peripheral component interconnect standard bus or an EISA extension industry standard architecture bus, among others. The bus 320 may be classified into a control bus, a data bus, an address bus, and the like. For ease of illustration, only one thick line is shown in FIG. 3, but not only one bus or one type of bus.
Memory 330 may be, but is not limited to, ROM read-only memory or other type of static storage device that can store static information and instructions, RAM random-access memory or other type of dynamic storage device that can store information and instructions, EEPROM electrically erasable programmable read-only memory, CD-ROM read-only or other optical disk storage, optical disk storage (including optical disks, laser disks, compact disks, digital versatile disks, etc.), magnetic disk storage media, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer.
The memory 330 is used for storing application program codes (computer programs) for executing the inventive arrangements and is controlled to be executed by the processor 310. The processor 310 is configured to execute the application code stored in the memory 330 to implement what is shown in the foregoing method embodiments.
The above is only a preferred embodiment of the present invention, and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (7)
1. A hyperspectral cross-scaling method based on spectral reconstruction, comprising:
Establishing a convolutional neural network model with a multi-head attention mechanism;
selecting hyperspectral wave band data of a target wave band by using the convolutional neural network model;
Establishing a long-term memory neural network model, and training the long-term memory neural network model by adopting a back propagation algorithm;
reconstructing the hyperspectral band data of the selected target band by using the trained long-short-term memory neural network model to obtain reconstructed hyperspectral data of the target band;
Cross-scaling using the target band hyperspectral data, comprising: after the reconstructed target band hyperspectral data are obtained, one-dimensional radiance information is expanded and calculated to obtain a multidimensional tensor; performing dimension reduction operation on the tensor by using two linear regression networks, and fitting the reconstructed wave band information to obtain radiance; inputting data into the trained long-short-period memory neural network model, and fitting out the radiance of a missing wave band; and taking the fitted wave band radiance as a target value, and calculating a scaling coefficient by combining with the DN value of the remote sensor to be scaled to finish cross scaling.
2. The hyperspectral cross scaling method based on spectrum reconstruction according to claim 1, wherein a plurality of multi-head attention modules and full connection layers are adopted as the basic structure of the long-short-term memory neural network model, and step functions are added as filters during output.
3. The hyperspectral cross-scaling method based on spectral reconstruction as claimed in claim 1, wherein selecting hyperspectral band data of a target band using the convolutional neural network model includes: performing feature extraction on the hyperspectral wave band data to obtain a weight matrix of a target wave band; and screening the weight matrix of the target wave band, and reserving the hyperspectral wave band with the largest weight by using a step function to obtain the hyperspectral wave band data of the target wave band.
4. The hyperspectral cross calibration method based on spectrum reconstruction according to claim 1, wherein when the hyperspectral band data of the selected target band is reconstructed by using the trained long-short-term memory neural network model, the radiance of each band is used as one-dimensional characteristics, and the filtered band set is used as a sequence to perform long-short-term memory neural network calculation.
5. The hyperspectral cross scaling method based on spectrum reconstruction according to claim 1, wherein the parameters of the convolutional neural network model are updated by adopting a back propagation algorithm, and the mean square error between an original matrix and a matrix reconstructed by using the convolutional neural network model is calculated; and determining a wave band meeting the reconstruction of the original full-wave band information according to the mean square error.
6. The hyperspectral cross calibration device based on spectrum reconstruction is characterized by comprising a model building unit, a selecting unit, a model training unit, a reconstruction unit and a cross calibration unit;
The model building unit is used for building a convolutional neural network model with a multi-head attention mechanism;
the selection unit is used for selecting hyperspectral wave band data of a target wave band by utilizing the convolutional neural network model;
the model training unit is used for building a long-period memory neural network model and training the long-period memory neural network model by adopting a back propagation algorithm;
The reconstruction unit is used for reconstructing the hyperspectral band data of the selected target band by using the trained long-short-term memory neural network model to obtain reconstructed hyperspectral data of the target band;
The cross scaling unit, configured to perform cross scaling using the target band hyperspectral data, includes: after the reconstructed target band hyperspectral data are obtained, one-dimensional radiance information is expanded and calculated to obtain a multidimensional tensor; performing dimension reduction operation on the tensor by using two linear regression networks, and fitting the reconstructed wave band information to obtain radiance; inputting data into the trained long-short-period memory neural network model, and fitting out the radiance of a missing wave band; and taking the fitted wave band radiance as a target value, and calculating a scaling coefficient by combining with the DN value of the remote sensor to be scaled to finish cross scaling.
7. An electronic device, comprising:
A processor and a memory;
the memory is used for storing computer operation instructions;
The processor for performing the hyperspectral cross scaling method based on spectral reconstruction as claimed in any one of claims 1 to 5 by invoking the computer operating instructions.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310286996.6A CN116306819B (en) | 2023-03-22 | 2023-03-22 | Hyperspectral cross calibration method and device based on spectrum reconstruction and electronic equipment |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310286996.6A CN116306819B (en) | 2023-03-22 | 2023-03-22 | Hyperspectral cross calibration method and device based on spectrum reconstruction and electronic equipment |
Publications (2)
Publication Number | Publication Date |
---|---|
CN116306819A CN116306819A (en) | 2023-06-23 |
CN116306819B true CN116306819B (en) | 2024-05-03 |
Family
ID=86794019
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310286996.6A Active CN116306819B (en) | 2023-03-22 | 2023-03-22 | Hyperspectral cross calibration method and device based on spectrum reconstruction and electronic equipment |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116306819B (en) |
Citations (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102279393A (en) * | 2011-07-15 | 2011-12-14 | 北京航空航天大学 | Cross radiometric calibration method of hyper-spectral sensor based on multi-spectral sensor |
CN106940219A (en) * | 2017-02-15 | 2017-07-11 | 中国科学院遥感与数字地球研究所 | A kind of spectral response acquisition methods of broadband satellite remote sensor in orbit |
CN109815892A (en) * | 2019-01-22 | 2019-05-28 | 武汉理工大学 | The signal recognition method of distributed fiber grating sensing network based on CNN |
CN111007021A (en) * | 2019-12-31 | 2020-04-14 | 北京理工大学重庆创新中心 | Hyperspectral water quality parameter inversion system and method based on one-dimensional convolution neural network |
CN111191514A (en) * | 2019-12-04 | 2020-05-22 | 中国地质大学(武汉) | Hyperspectral image band selection method based on deep learning |
CN112861810A (en) * | 2021-03-23 | 2021-05-28 | 中国科学院、水利部成都山地灾害与环境研究所 | Artificial forest planting time automatic detection method based on time sequence remote sensing observation data |
CN114155429A (en) * | 2021-10-09 | 2022-03-08 | 信阳学院 | Reservoir earth surface temperature prediction method based on space-time bidirectional attention mechanism |
CN114279567A (en) * | 2022-03-01 | 2022-04-05 | 潍坊绘圆地理信息有限公司 | On-orbit absolute radiation calibration method for micro-nano hyperspectral satellite constellation |
CN114418776A (en) * | 2021-12-30 | 2022-04-29 | 深圳价值在线信息科技股份有限公司 | Data processing method, device, terminal equipment and medium |
CN115265783A (en) * | 2022-08-01 | 2022-11-01 | 国家卫星海洋应用中心 | Multi-platform satellite-borne point instantaneous cross calibration method and device based on hyperspectral data |
CN115641503A (en) * | 2022-09-30 | 2023-01-24 | 浙江大学 | Hyperspectral band selection method and device based on Transformer network |
-
2023
- 2023-03-22 CN CN202310286996.6A patent/CN116306819B/en active Active
Patent Citations (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102279393A (en) * | 2011-07-15 | 2011-12-14 | 北京航空航天大学 | Cross radiometric calibration method of hyper-spectral sensor based on multi-spectral sensor |
CN106940219A (en) * | 2017-02-15 | 2017-07-11 | 中国科学院遥感与数字地球研究所 | A kind of spectral response acquisition methods of broadband satellite remote sensor in orbit |
CN109815892A (en) * | 2019-01-22 | 2019-05-28 | 武汉理工大学 | The signal recognition method of distributed fiber grating sensing network based on CNN |
CN111191514A (en) * | 2019-12-04 | 2020-05-22 | 中国地质大学(武汉) | Hyperspectral image band selection method based on deep learning |
CN111007021A (en) * | 2019-12-31 | 2020-04-14 | 北京理工大学重庆创新中心 | Hyperspectral water quality parameter inversion system and method based on one-dimensional convolution neural network |
CN112861810A (en) * | 2021-03-23 | 2021-05-28 | 中国科学院、水利部成都山地灾害与环境研究所 | Artificial forest planting time automatic detection method based on time sequence remote sensing observation data |
CN114155429A (en) * | 2021-10-09 | 2022-03-08 | 信阳学院 | Reservoir earth surface temperature prediction method based on space-time bidirectional attention mechanism |
CN114418776A (en) * | 2021-12-30 | 2022-04-29 | 深圳价值在线信息科技股份有限公司 | Data processing method, device, terminal equipment and medium |
CN114279567A (en) * | 2022-03-01 | 2022-04-05 | 潍坊绘圆地理信息有限公司 | On-orbit absolute radiation calibration method for micro-nano hyperspectral satellite constellation |
CN115265783A (en) * | 2022-08-01 | 2022-11-01 | 国家卫星海洋应用中心 | Multi-platform satellite-borne point instantaneous cross calibration method and device based on hyperspectral data |
CN115641503A (en) * | 2022-09-30 | 2023-01-24 | 浙江大学 | Hyperspectral band selection method and device based on Transformer network |
Non-Patent Citations (5)
Title |
---|
"radiometric calibration of fengyun-3d meris ii satellite a case study in lake qinghai china";lin yan;《IGARSS 2020 IEEE international geoscience and remote sensing symposium》;20210217;全文 * |
"radiometric cross calibration of the ZY1-02D Hyperspectral imager using the GF-5 ahsi imager";牛超;《IEEE transactions on geoscience and remote sensing》;20211130;全文 * |
"基于FY-3E星载红外观测的交叉定标匹配不确定性分析";杨天杭;《红外与激光工程》;20221201;全文 * |
"基于卫星定标参数时间序列的趋势预测研究";马金锋;《中国优秀硕士学位论文全文数据库 基础科学辑》;20220115;全文 * |
"逆合成孔径雷达成像横向定标方法研究";黄鑫;《中国优秀硕士学位论文全文数据库 信息科技辑》;20210115;全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN116306819A (en) | 2023-06-23 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Huser et al. | Advances in statistical modeling of spatial extremes | |
Sharma et al. | An information theoretic alternative to model a natural system using observational information alone | |
Webber et al. | Practical rare event sampling for extreme mesoscale weather | |
Lagerquist et al. | Using deep learning to emulate and accelerate a radiative transfer model | |
Gholipour et al. | Extracting the main patterns of natural time series for long-term neurofuzzy prediction | |
CN111931983B (en) | Precipitation prediction method and system | |
US7882105B2 (en) | Method of ranking a set of electronic documents of the type possibly containing hypertext links to other electronic documents | |
CN116449462B (en) | Method, system, storage medium and equipment for predicting effective wave height space-time sequence of sea wave | |
Quinting et al. | EuLerian Identification of ascending AirStreams (ELIAS 2.0) in numerical weather prediction and climate models–Part 1: Development of deep learning model | |
CN115660887A (en) | Photovoltaic output prediction method and system based on limited weather forecast information | |
CN113111975A (en) | SAR image target classification method based on multi-kernel scale convolutional neural network | |
CN116306819B (en) | Hyperspectral cross calibration method and device based on spectrum reconstruction and electronic equipment | |
Garai et al. | An MRA Based MLR Model for Forecasting Indian Annual Rainfall Using Large Scale Climate Indices | |
Karami et al. | Image denoising using generalised Cauchy filter | |
US20080147351A1 (en) | Source Decay Parameter System and Method for Automatic Grid Generation | |
Wang et al. | Penalized local polynomial regression for spatial data | |
CN111950198B (en) | Ground clutter simulation method based on neural network | |
Wang et al. | Statistical and machine learning methods applied to the prediction of different tropical rainfall types | |
CN115238967A (en) | Photovoltaic power prediction method and device combining cloud picture and adjacent power station cluster | |
CN112446550B (en) | Short-term building load probability density prediction method | |
Lin et al. | Smart building uncertainty analysis via adaptive Lasso | |
CN117556716B (en) | Design method, device, equipment and storage medium of electromagnetic glass | |
CN114398228A (en) | Method and device for predicting equipment resource use condition and electronic equipment | |
Hess et al. | Deep learning for improving numerical weather prediction of rainfall extremes | |
CN116337240B (en) | Thermal infrared hyperspectral data band selection method based on graph 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 |