CN116384237A - Thermal infrared atmospheric parameter inversion method and device and electronic equipment - Google Patents
Thermal infrared atmospheric parameter inversion method and device and electronic equipment Download PDFInfo
- Publication number
- CN116384237A CN116384237A CN202310327016.2A CN202310327016A CN116384237A CN 116384237 A CN116384237 A CN 116384237A CN 202310327016 A CN202310327016 A CN 202310327016A CN 116384237 A CN116384237 A CN 116384237A
- Authority
- CN
- China
- Prior art keywords
- neural network
- term memory
- short
- network model
- memory neural
- 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 38
- 230000015654 memory Effects 0.000 claims abstract description 83
- 238000003062 neural network model Methods 0.000 claims abstract description 79
- 239000013598 vector Substances 0.000 claims abstract description 43
- 238000004088 simulation Methods 0.000 claims abstract description 34
- 238000012549 training Methods 0.000 claims abstract description 30
- 238000013528 artificial neural network Methods 0.000 claims abstract description 22
- 230000007246 mechanism Effects 0.000 claims abstract description 14
- 238000000605 extraction Methods 0.000 claims description 24
- 230000006870 function Effects 0.000 claims description 19
- 230000007787 long-term memory Effects 0.000 claims description 15
- 238000004422 calculation algorithm Methods 0.000 claims description 14
- 230000004913 activation Effects 0.000 claims description 10
- 238000001228 spectrum Methods 0.000 claims description 8
- 238000010276 construction Methods 0.000 claims description 5
- 230000002441 reversible effect Effects 0.000 claims description 5
- 230000002238 attenuated effect Effects 0.000 claims description 4
- 238000011022 operating instruction Methods 0.000 claims 1
- 238000007418 data mining Methods 0.000 abstract description 4
- 238000012545 processing Methods 0.000 abstract description 3
- 238000013507 mapping Methods 0.000 description 9
- 230000008569 process Effects 0.000 description 8
- 230000005855 radiation Effects 0.000 description 7
- 238000002834 transmittance Methods 0.000 description 5
- 238000004590 computer program Methods 0.000 description 3
- 230000003287 optical effect Effects 0.000 description 3
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Chemical compound O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 3
- 230000002457 bidirectional effect Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000007774 longterm Effects 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 238000012216 screening Methods 0.000 description 2
- 230000003068 static effect Effects 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000003213 activating effect 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
- 238000000802 evaporation-induced self-assembly Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 230000001373 regressive effect Effects 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 239000002689 soil Substances 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01W—METEOROLOGY
- G01W1/00—Meteorology
- G01W1/02—Instruments for indicating weather conditions by measuring two or more variables, e.g. humidity, pressure, temperature, cloud cover or wind speed
-
- 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
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
- G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
-
- 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
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Environmental & Geological Engineering (AREA)
- Environmental Sciences (AREA)
- Ecology (AREA)
- Biodiversity & Conservation Biology (AREA)
- Geometry (AREA)
- Computer Hardware Design (AREA)
- Atmospheric Sciences (AREA)
- Medical Informatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Transforming Light Signals Into Electric Signals (AREA)
- Radiation Pyrometers (AREA)
- Photometry And Measurement Of Optical Pulse Characteristics (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to the technical field of remote sensing image processing, and relates to a thermal infrared atmospheric parameter inversion method, a device and electronic equipment, wherein the method comprises the following steps: extracting surface emissivity information and atmospheric information; determining atmospheric parameters and establishing a simulation data set; constructing a two-way long-short-term memory neural network model; model training to determine structural parameters of the model; weighting each channel information of the output feature vector of the two-way long-short-term memory neural network model by using an attention mechanism; training a model; iteratively updating until the output of the model converges to obtain an atmospheric parameter inversion model; and obtaining a thermal infrared atmospheric parameter inversion result. The invention effectively solves the problem that the deep neural network has weaker data mining capability, provides a two-way long-short-term memory neural network structure, weights different channel information by utilizing the channel attention module, acquires the correlation characteristics of different channel radiance data, and obtains a thermal infrared atmospheric parameter inversion result.
Description
Technical Field
The invention belongs to the technical field of remote sensing image processing, and particularly relates to a long-term and short-term memory neural network thermal infrared atmospheric parameter inversion method and device based on an attention mechanism, and electronic equipment.
Background
Atmospheric parameter inversion has received a great deal of attention in recent years and has found wide application in a number of discipline fields. Inversion of atmospheric parameters has also attracted the interest of many scholars, for example, modeling the correlation of the atmospheric transmittance of different wavelengths, and inverting and calculating the atmospheric transmittance of radiation with adjacent wavelengths; and obtaining the transmittance by modeling the statistical relationship between the water vapor and the atmospheric transmittance in different wave bands. In practical application, the observed bright temperature is obtained by practical observation of a microwave radiometer, and the space resolution is low, so that the atmospheric parameter inversion accuracy is low, and the deep neural network has the problem of weak data mining capability.
Disclosure of Invention
In order to solve the technical problems, the invention provides a thermal infrared atmospheric parameter inversion method, a thermal infrared atmospheric parameter inversion device and electronic equipment.
In a first aspect, the invention provides a thermal infrared atmospheric parameter inversion method comprising:
extracting surface emissivity information and atmospheric information of a ground object spectrum library and an atmospheric profile library;
according to the earth surface emissivity information and the atmosphere information, determining atmosphere parameters, and establishing a simulation data set;
constructing a two-way long-short-term memory neural network model;
training the two-way long-short-term memory neural network model by utilizing hyperspectral data in the simulation data set, and determining structural parameters of the two-way long-short-term memory neural network model;
weighting each channel information of the output feature vector of the two-way long-short-term memory neural network model by using an attention mechanism to obtain a target two-way long-short-term memory neural network model;
performing data training on the target two-way long-short-term memory neural network model through the atmospheric parameters in the simulation data set;
the trained target two-way long-short-term memory neural network model parameters are continuously iteratively updated by adopting a back propagation algorithm until the output of the target two-way long-term memory neural network model converges, and an atmospheric parameter inversion model is obtained;
and acquiring thermal infrared hyperspectral radiance data and inputting the data as the atmospheric parameter inversion model to obtain a thermal infrared atmospheric parameter inversion result.
The invention provides a thermal infrared atmospheric parameter inversion device, which comprises an extraction unit, a simulation data set establishment unit, a first model construction unit, a first model training unit, a second model construction unit, a second model training unit, an iteration updating unit, an input unit and an output unit;
the extraction unit is used for extracting surface emissivity information and atmosphere information of the ground object spectrum library and the atmosphere profile library;
the simulation data set establishing unit is used for determining atmospheric parameters according to the earth surface emissivity information and the atmospheric information and establishing a simulation data set;
the first model construction unit is used for constructing a two-way long-short-term memory neural network model;
the first model training unit is used for training the two-way long-short-term memory neural network model by utilizing hyperspectral data in the simulation data set, and determining structural parameters of the two-way long-short-term memory neural network model;
the second model building unit is used for weighting the channel information of the output feature vector of the two-way long-short-term memory neural network model by using an attention mechanism to obtain a target two-way long-short-term memory neural network model;
the second model training unit is used for carrying out data training on the target two-way long-short-term memory neural network model through the atmospheric parameters in the simulation data set;
the iteration updating unit is used for adopting a back propagation algorithm to continuously and iteratively update the trained parameters of the target two-way long-short-term memory neural network model until the output of the target two-way long-term memory neural network model converges, so as to obtain an atmospheric parameter inversion model;
the input unit is used for acquiring thermal infrared hyperspectral radiance data and taking the data as the input of the atmospheric parameter inversion model;
and the output unit is used for outputting a thermal infrared atmospheric parameter inversion result.
In a third aspect, the present invention discloses an electronic device, comprising:
a processor and a memory;
the memory is used for storing computer operation instructions;
and the processor is used for executing the thermal infrared atmospheric parameter inversion method by calling the computer operation instruction.
The beneficial effects of the invention are as follows: the method not only effectively solves the problem that the deep neural network has weaker data mining capacity, but also strengthens the importance degree of the network on the correlation information between hyperspectral adjacent channel data in the inversion process. The invention provides a two-way long-short-term memory neural network structure which is used for extracting correlation characteristics of different channels; then, weighting different channel information by using a channel attention module so as to improve the proportion of the channel with higher weight in the characteristics; finally, mapping the characteristics to a sample marking space by adopting a full-connection structure, and converting the characteristics into the estimation results of the atmospheric uplink radiation, the downlink radiation and the transmissivity of each channel; the invention utilizes the radiance information of the remote sensing image, utilizes a circulating neural network module with a two-way long-short-term memory neural network to add a attention mechanism, obtains the correlation between the radiance and the atmospheric parameter, and obtains the uplink radiation, the downlink radiation and the transmittance of the atmosphere by inversion.
On the basis of the technical scheme, the invention can be improved as follows.
Further, the two-way long-short-term memory neural network model comprises two LSTM neural networks with mutually independent parameters, the hyperspectral data in the simulation data set are respectively input into the LSTM neural networks in a positive sequence and a reverse sequence to perform feature extraction, two feature extraction vectors are obtained, and the feature extraction vectors are spliced to form an output vector.
Further, weighting each channel information of the output feature vector of the two-way long-short term memory neural network model by using an attention mechanism, including:
through convolution operation, the input data with the given characteristic channel number being the first channel number is transformed to obtain a characteristic vector with the characteristic channel number being the second channel number;
compressing the feature vector in a space dimension by using a global method, and compressing the two-dimensional feature vector of each channel into a real number to serve as a feature value of each channel;
forming a Bottleneck structure through a plurality of full-connection layers to generate weight values for the characteristic values of all channels, and normalizing the weight values;
and weighting the normalized weight to the characteristic value of each channel, and carrying out weighted summation on the weight value and the characteristic vector.
Further, the full connection layer adopts three layers of full connection modules for mapping; the fully connected modules of adjacent layers are activated by using an activation function.
Further, the trained parameters of the target two-way long-short-term memory neural network model are continuously and iteratively updated by adopting a back propagation algorithm until the output of the target two-way long-term memory neural network model converges, and an atmospheric parameter inversion model is obtained, which comprises the following steps:
acquiring hyperspectral atmospheric parameter inversion data from the simulation data set as training data;
reading a plurality of atmosphere parameters corresponding to the radiance under different atmosphere parameters from the tag data, and matching the atmosphere parameters with the radiance information under the same atmosphere parameter state;
updating parameters of the target two-way long-short-term memory neural network model by adopting a back propagation algorithm, and inputting the radiance under different states to enable the root mean square error value of the target two-way long-short-term memory neural network model to be gradually attenuated to a set value;
and using the mean square error as a loss function, and selecting a network module with the minimum final loss function as the atmospheric parameter inversion model.
Further, by setting a channel attention module and setting the number of downsampling channels at equal intervals in the channel attention module, the number of downsampling channels of the channel attention module is determined according to the convergence speed of the target two-way long-short-term memory neural network model.
Drawings
FIG. 1 is a flow chart of a thermal infrared atmospheric parameter inversion method provided in embodiment 1 of the present invention;
FIG. 2 is a schematic diagram of a thermal infrared atmospheric parameter inversion apparatus according to embodiment 1 of the present invention;
fig. 3 is a schematic diagram of an electronic device according to 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 embodiment provides a thermal infrared atmospheric parameter inversion method, which includes:
extracting surface emissivity information and atmospheric information of a ground object spectrum library and an atmospheric profile library;
according to the earth surface emissivity information and the atmosphere information, determining atmosphere parameters, and establishing a simulation data set;
constructing a two-way long-short-term memory neural network model;
training the two-way long-short-term memory neural network model by utilizing hyperspectral data in the simulation data set, and determining structural parameters of the two-way long-short-term memory neural network model;
weighting each channel information of the output feature vector of the two-way long-short-term memory neural network model by using an attention mechanism to obtain a target two-way long-short-term memory neural network model;
performing data training on the target two-way long-short-term memory neural network model through simulating atmospheric parameters in the data set;
the back propagation algorithm is adopted to continuously and iteratively update the trained parameters of the target two-way long-short-term memory neural network model until the output of the target two-way long-term memory neural network model converges, and an atmospheric parameter inversion model is obtained;
and acquiring thermal infrared hyperspectral radiance data and inputting the data as an atmospheric parameter inversion model to obtain a thermal infrared atmospheric parameter inversion result.
In the practical application process, the earth surface emissivity information and the atmosphere information of an earth surface spectrum library and an atmosphere profile library (thermodynamic initial guess retrieval, TIGR) are extracted, and the atmosphere parameters are calculated by using MODTRA to establish a simulation data set. Hyperspectral data are obtained through the simulation data set, the radiance under different emissivity and the atmospheric parameters corresponding to the radiance are matched, and the atmospheric parameters are matched with the radiance information under the same state. 2311 different regional atmosphere profiles are contained in an original database, 945 atmosphere profiles are selected through cloud removal screening and water vapor and bottom temperature screening, different surface emissivity information is extracted through a ground feature emissivity library, the emissivity of 66 different ground features including ground feature types such as bare soil, vegetation and water is simulated through MODTRA, 74844000 simulated radiance data are formed through setting different ground features and different atmosphere profiles, adjacent 10 wave numbers are processed into a wave band through an arithmetic average value (The average value), and 24 wave bands are obtained.
Optionally, the two-way long-short-term memory neural network model comprises two LSTM neural networks with mutually independent parameters, hyperspectral data in the simulation data set are respectively input into the LSTM neural networks in a positive sequence and a reverse sequence to perform feature extraction, two feature extraction vectors are obtained, and the feature extraction vectors are spliced to form an output vector.
In the practical application process, a two-way long-short-term memory neural network model is constructed, the two-way long-term memory neural network is adopted to perform feature extraction on the simulation data set, the long-term neural network operation is used for mapping the input data set X with the size of H multiplied by W multiplied by S into the output data set Y with the size of H multiplied by W multiplied by T, and the formula is as follows:
Y h',w',t for outputting data;
is input data; c (C) m,n,s,t The weight of a certain current position; b m,n,s,t Is the bias of a certain current position; h is the batch of data; w is the number of channels; s is the data length; m is the current batch; n is the current channel number; s is the current data length; t is the time.
The specific calculation process is as follows:
f(t)＝σ(Wfh t-1 +Ufx t +bf)；
i(t)＝σ(W i h t-1 +U i x t +b i )；
a(t)＝tanh(W a h t-1 +U a x t +b a )；
o(t)＝σ(W o h t-1 +U o x t +b o )；
wherein x is t Input at time t, h t-1 The state value of the hidden layer at the time t-1 is represented; w (W) f 、W i 、W o And W is a Respectively representing forgetting gate, input gate, output gate and h in the characteristic extraction process t-1 Weight coefficient of (2); u (U) f 、U i 、U o And U a Respectively representing forgetting gate, input gate, output gate and x in the characteristic extraction process t Weight coefficient of (2); b f 、b i 、b o And b a Respectively representing the bias values in the forgetting gate, the input gate, the output gate and the feature extraction process, wherein tan h (·) represents a tangent hyperbolic function, sigma (·) represents an activation function Sigmoid, and the formula is as follows:
the result of the forget gate and input gate calculation acts on c (t-1), forming the variable state c (t) at time t, expressed as:
c(t)＝c(t-1)⊙f(t)+i(t)⊙a(t)；
wherein, as follows, hadamard product, f (t) is the information left by the forgetting gate at the moment t; i (t) is information of input gate update at the moment t; a (t) is a value for transition to the variable state at time t. Finally, the state h (t) of the hidden layer at the time t is obtained from the output gate o (t) and the variable state c (t) at the current time, and is expressed as:
h(t)＝o(t)⊙tanh(c(t))；
the bidirectional long-short term neural network module consists of 2 LSTM with mutually independent parameters, the data set is respectively input into the 2 LSTM neural networks in positive sequence and reverse sequence for feature extraction, and a new feature vector formed after 2 output vectors (namely feature extraction vectors) are spliced is obtained and is used as the final feature expression of the data. Under this concept, the bidirectional LSTM module allows feature data obtained at time t to have correlation information between the past and future at the same time.
Optionally, weighting each channel information of the output feature vector of the two-way long-short term memory neural network model by using an attention mechanism includes:
through convolution operation, the input data with the given characteristic channel number being the first channel number is transformed to obtain a characteristic vector with the characteristic channel number being the second channel number;
compressing the feature vector in the space dimension by using a global method, and compressing the two-dimensional feature vector of each channel into a real number to serve as the feature value of each channel;
a plurality of full-connection layers form a Bottleneck structure to generate weight values for the characteristic values of all channels, and the weight values are normalized;
and weighting the normalized weights to the characteristic values of the channels, and carrying out weighted summation on the weight values and the characteristic vectors.
In the practical application process, a characteristic channel number T is obtained by convolution operation for the input conversion of a given characteristic channel number T 1 Is characterized by (2); compressing the features in the space dimension by using a global pooling method, compressing the two-dimensional features of each channel into a real number, and obtaining an image area corresponding to the feature value; wherein the C-th element Z of the feature Z is calculated C The values are formulated as:
wherein H is the height of the image; w is the width of the image; H×W is a channelIs a two-dimensional feature of (2); i is a counter that accumulates in height; j is a counter that accumulates over the width; u (u) c The pixel value at the C channel with the height of i and the width of j;
the importance of each channel was evaluated by two fully-connected layer composition Bottleneck structures, by Bottleneck Z C Generating a weight value for each characteristic channel of the model, and normalizing the weight; finally, the normalized weights are weighted to the features of each channel, and the weights and feature vectors are weighted and summed.
Optionally, the full connection layer adopts three layers of full connection modules for mapping; the fully connected modules of adjacent layers are activated by using an activation function.
In the practical application process, the characteristics of the weighted channels are output to the full-connection layer, the extracted characteristics of the channels are mapped to the sample marking space, and through testing, when the network uses a three-layer full-connection layer structure, the regressive atmosphere parameters are stable, and the accuracy is higher.
Optionally, the method for obtaining the atmospheric parameter inversion model by adopting a back propagation algorithm to continuously and iteratively update the trained target two-way long-short-term memory neural network model parameters until the output of the target two-way long-term memory neural network model converges comprises the following steps:
acquiring hyperspectral atmospheric parameter inversion data from the simulation data set as training data;
reading a plurality of atmosphere parameters corresponding to the radiance under different atmosphere parameters from the tag data, and matching the atmosphere parameters with the radiance information under the same atmosphere parameter state;
updating parameters of the target two-way long-short-term memory neural network model by adopting a back propagation algorithm, and inputting the radiance under different states to enable the root mean square error value of the target two-way long-term memory neural network model to be gradually attenuated to a set value;
and using the mean square error as a loss function, and selecting a network module with the minimum final loss function as an atmospheric parameter inversion model.
In the practical application process, the invention uses the combined structure of the channel attention mechanism and the full-connection layer to form the channel attention module, and performs dimension lifting on the spectrum characteristics extracted by the long-term and short-term network. And finally determining the number of downsampling channels of the attention module by setting the number of downsampling channels of the attention modules at equal intervals, mapping the extracted features to a full connection layer (Linear), normalizing the features by a normalization function, and activating the features by a Sigmoid activation function. Feature mapping is performed by using three layers of fully connected modules (Linear 1, linear2 and Linear 3), feature activation is performed between Linear1 and Linear2 by using a Sigmoid activation function through testing, and feature activation is performed between Linear2 and Linear3 by using a Softmax activation function.
And mapping the characteristics extracted by the channel attention module to a sample space through a multi-layer full-connection module, wherein the output of the module is the required atmospheric parameter.
Optionally, the channel attention module is set, and the number of downsampling channels of the channel attention module is determined according to the convergence speed of the target two-way long-short-term memory neural network model in a mode that the channel attention module is set with the number of downsampling channels at equal intervals.
The method not only effectively solves the problem that the deep neural network has weaker data mining capacity, but also strengthens the importance degree of the network on the correlation information between hyperspectral adjacent channel data in the inversion process. The invention provides a two-way long-short-term memory neural network structure which is used for extracting correlation characteristics of different channels; then, weighting different channel information by using a channel attention module so as to improve the proportion of the channel with higher weight in the characteristics; and finally, mapping the characteristics to a sample marking space by adopting a full-connection structure, and converting the characteristics into the estimation results of the atmospheric uplink radiation, the downlink radiation and the transmissivity of each channel.
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 thermal infrared atmospheric parameter inversion device, which includes an extraction unit, a simulation data set building unit, a first model training unit, a second model building unit, a second model training unit, an iteration updating unit, an input unit and an output unit;
the extraction unit is used for extracting the earth surface emissivity information and the atmosphere information of the earth surface spectrum library and the atmosphere profile library;
the simulation data set establishing unit is used for determining atmospheric parameters according to the earth surface emissivity information and the atmospheric information and establishing a simulation data set;
the first model building unit is used for building a two-way long-short-term memory neural network model;
the first model training unit is used for training the two-way long-short-term memory neural network model by utilizing hyperspectral data in the simulation data set, and determining structural parameters of the two-way long-term memory neural network model;
the second model construction unit is used for weighting the information of each channel of the output characteristic vector of the two-way long-short-term memory neural network model by using an attention mechanism to obtain a target two-way long-short-term memory neural network model;
the second model training unit is used for carrying out data training on the target two-way long-short-term memory neural network model through simulating the atmospheric parameters in the data set;
the iteration updating unit is used for adopting a back propagation algorithm to continuously and iteratively update the trained target two-way long-short-term memory neural network model parameters until the output of the target two-way long-term memory neural network model converges, so as to obtain an atmospheric parameter inversion model;
the input unit is used for acquiring the thermal infrared hyperspectral radiance data and inputting an atmospheric parameter inversion model;
and the output unit is used for outputting a thermal infrared atmospheric parameter inversion result.
Optionally, the two-way long-short-term memory neural network model comprises two LSTM neural networks with mutually independent parameters, hyperspectral data in the simulation data set are respectively input into the LSTM neural networks in a positive sequence and a reverse sequence to perform feature extraction, two feature extraction vectors are obtained, and the feature extraction vectors are spliced to form an output vector.
Optionally, weighting each channel information of the output feature vector of the two-way long-short term memory neural network model by using an attention mechanism includes:
through convolution operation, the input data with the given characteristic channel number being the first channel number is transformed to obtain a characteristic vector with the characteristic channel number being the second channel number;
compressing the feature vector in the space dimension by using a global method, and compressing the two-dimensional feature vector of each channel into a real number to serve as the feature value of each channel;
a plurality of full-connection layers form a Bottleneck structure to generate weight values for the characteristic values of all channels, and the weight values are normalized;
and weighting the normalized weights to the characteristic values of the channels, and carrying out weighted summation on the weight values and the characteristic vectors.
Optionally, the full connection layer adopts three layers of full connection modules for mapping; the fully connected modules of adjacent layers are activated by using an activation function.
Optionally, the method for obtaining the atmospheric parameter inversion model by adopting a back propagation algorithm to continuously and iteratively update the trained target two-way long-short-term memory neural network model parameters until the output of the target two-way long-term memory neural network model converges comprises the following steps:
acquiring hyperspectral atmospheric parameter inversion data from the simulation data set as training data;
reading a plurality of atmosphere parameters corresponding to the radiance under different atmosphere parameters from the tag data, and matching the atmosphere parameters with the radiance information under the same atmosphere parameter state;
updating parameters of the target two-way long-short-term memory neural network model by adopting a back propagation algorithm, and inputting the radiance under different states to enable the root mean square error value of the target two-way long-term memory neural network model to be gradually attenuated to a set value;
and using the mean square error as a loss function, and selecting a network module with the minimum final loss function as an atmospheric parameter inversion model.
Optionally, the channel attention module is set, and the number of downsampling channels of the channel attention module is determined according to the convergence speed of the target two-way long-short-term memory neural network model in a mode that the channel attention module is set with the number of downsampling channels at equal intervals.
Example 3
Based on the same principle as the method shown in the embodiment of the present invention, there is also provided an electronic device in the embodiment of the present invention, as shown in fig. 3, which may include, but is not limited to: a processor and a memory; a memory for storing a computer program; a processor for executing the method according to 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 350. Processor 310 is coupled to memory 350, such as via 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.
Memory 350 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 350 is used to store 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 350 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 (8)
1. The thermal infrared atmospheric parameter inversion method is characterized by comprising the following steps of:
extracting surface emissivity information and atmospheric information of a ground object spectrum library and an atmospheric profile library;
according to the earth surface emissivity information and the atmosphere information, determining atmosphere parameters, and establishing a simulation data set;
constructing a two-way long-short-term memory neural network model;
training the two-way long-short-term memory neural network model by utilizing hyperspectral data in the simulation data set, and determining structural parameters of the two-way long-short-term memory neural network model;
weighting each channel information of the output feature vector of the two-way long-short-term memory neural network model by using an attention mechanism to obtain a target two-way long-short-term memory neural network model;
performing data training on the target two-way long-short-term memory neural network model through the atmospheric parameters in the simulation data set;
the trained target two-way long-short-term memory neural network model parameters are continuously iteratively updated by adopting a back propagation algorithm until the output of the target two-way long-term memory neural network model converges, and an atmospheric parameter inversion model is obtained;
and acquiring thermal infrared hyperspectral radiance data and inputting the data as the atmospheric parameter inversion model to obtain a thermal infrared atmospheric parameter inversion result.
2. The thermal infrared atmospheric parameter inversion method according to claim 1, wherein the two-way long-short-term memory neural network model comprises two LSTM neural networks with mutually independent parameters, the hyperspectral data in the simulation data set are respectively input into the LSTM neural networks in a positive sequence and a reverse sequence to perform feature extraction, two feature extraction vectors are obtained, and the feature extraction vectors are spliced to form an output vector.
3. The thermal infrared atmospheric parameter inversion method according to claim 1, wherein weighting each channel information of the output feature vector of the two-way long-short-term memory neural network model by using an attention mechanism comprises:
through convolution operation, the input data with the given characteristic channel number being the first channel number is transformed to obtain a characteristic vector with the characteristic channel number being the second channel number;
compressing the feature vector in a space dimension by using a global method, and compressing the two-dimensional feature vector of each channel into a real number to serve as a feature value of each channel;
forming a Bottleneck structure through a plurality of full-connection layers to generate weight values for the characteristic values of all channels, and normalizing the weight values;
and weighting the normalized weight to the characteristic value of each channel, and carrying out weighted summation on the weight value and the characteristic vector.
4. The thermal infrared atmospheric parameter inversion method according to claim 3 wherein said fully-connected layer is mapped using a three-layer fully-connected module; the fully connected modules of adjacent layers are activated by using an activation function.
5. The thermal infrared atmospheric parameter inversion method according to claim 1, wherein the step of obtaining the atmospheric parameter inversion model by using a back propagation algorithm to iteratively update the trained parameters of the target two-way long-short-term memory neural network model until the output of the target two-way long-term memory neural network model converges comprises the steps of:
acquiring hyperspectral atmospheric parameter inversion data from the simulation data set as training data;
reading a plurality of atmosphere parameters corresponding to the radiance under different atmosphere parameters from the tag data, and matching the atmosphere parameters with the radiance information under the same atmosphere parameter state;
updating parameters of the target two-way long-short-term memory neural network model by adopting a back propagation algorithm, and inputting the radiance under different states to enable the root mean square error value of the target two-way long-short-term memory neural network model to be gradually attenuated to a set value;
and using the mean square error as a loss function, and selecting a network module with the minimum final loss function as the atmospheric parameter inversion model.
6. The thermal infrared atmospheric parameter inversion method according to claim 1, wherein the number of downsampling channels of the channel attention module is determined according to the convergence speed of the target two-way long-short-term memory neural network model by setting the channel attention module and setting the number of downsampling channels at equal intervals in the channel attention module.
7. The thermal infrared atmospheric parameter inversion device is characterized by comprising an extraction unit, a simulation data set establishment unit, a first model training unit, a second model establishment unit, a second model training unit, an iteration updating unit, an input unit and an output unit;
the extraction unit is used for extracting surface emissivity information and atmosphere information of the ground object spectrum library and the atmosphere profile library;
the simulation data set establishing unit is used for determining atmospheric parameters according to the earth surface emissivity information and the atmospheric information and establishing a simulation data set;
the first model construction unit is used for constructing a two-way long-short-term memory neural network model;
the first model training unit is used for training the two-way long-short-term memory neural network model by utilizing hyperspectral data in the simulation data set, and determining structural parameters of the two-way long-short-term memory neural network model;
the second model building unit is used for weighting the channel information of the output feature vector of the two-way long-short-term memory neural network model by using an attention mechanism to obtain a target two-way long-short-term memory neural network model;
the second model training unit is used for carrying out data training on the target two-way long-short-term memory neural network model through the atmospheric parameters in the simulation data set;
the iteration updating unit is used for adopting a back propagation algorithm to continuously and iteratively update the trained parameters of the target two-way long-short-term memory neural network model until the output of the target two-way long-term memory neural network model converges, so as to obtain an atmospheric parameter inversion model;
the input unit is used for acquiring thermal infrared hyperspectral radiance data and taking the data as the input of the atmospheric parameter inversion model;
and the output unit is used for outputting a thermal infrared atmospheric parameter inversion result.
8. An electronic device, comprising:
a processor and a memory;
the memory is used for storing computer operation instructions;
the processor for performing the thermal infrared atmospheric parameter inversion method of any one of claims 1 to 6 by invoking the computer operating instructions.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310327016.2A CN116384237B (en) | 2023-03-29 | 2023-03-29 | Thermal infrared atmospheric parameter inversion method and device and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310327016.2A CN116384237B (en) | 2023-03-29 | 2023-03-29 | Thermal infrared atmospheric parameter inversion method and device and electronic equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116384237A true CN116384237A (en) | 2023-07-04 |
| CN116384237B CN116384237B (en) | 2024-08-23 |
Family
ID=86965077
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310327016.2A Active CN116384237B (en) | 2023-03-29 | 2023-03-29 | Thermal infrared atmospheric parameter inversion method and device and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116384237B (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200202211A1 (en) * | 2018-12-25 | 2020-06-25 | Abbyy Production Llc | Neural network training utilizing loss functions reflecting neighbor token dependencies |
| CN111735772A (en) * | 2020-05-13 | 2020-10-02 | 中国科学院空天信息创新研究院 | Improved high-spectrum data earth surface reflectivity inversion method of cascade neural network |
| CN112733394A (en) * | 2020-12-21 | 2021-04-30 | 国家卫星气象中心(国家空间天气监测预警中心) | Atmospheric parameter inversion method and device |
| CN112733725A (en) * | 2021-01-12 | 2021-04-30 | 西安电子科技大学 | Hyperspectral image change detection method based on multistage cyclic convolution self-coding network |
| CN113379146A (en) * | 2021-06-24 | 2021-09-10 | 合肥工业大学智能制造技术研究院 | Pollutant concentration inversion method based on multi-feature selection algorithm |
| CN115422703A (en) * | 2022-07-19 | 2022-12-02 | 南京航空航天大学 | Surface thermal infrared emissivity inversion method based on MODIS data and Transformer network |
-
2023
- 2023-03-29 CN CN202310327016.2A patent/CN116384237B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200202211A1 (en) * | 2018-12-25 | 2020-06-25 | Abbyy Production Llc | Neural network training utilizing loss functions reflecting neighbor token dependencies |
| CN111735772A (en) * | 2020-05-13 | 2020-10-02 | 中国科学院空天信息创新研究院 | Improved high-spectrum data earth surface reflectivity inversion method of cascade neural network |
| CN112733394A (en) * | 2020-12-21 | 2021-04-30 | 国家卫星气象中心(国家空间天气监测预警中心) | Atmospheric parameter inversion method and device |
| CN112733725A (en) * | 2021-01-12 | 2021-04-30 | 西安电子科技大学 | Hyperspectral image change detection method based on multistage cyclic convolution self-coding network |
| CN113379146A (en) * | 2021-06-24 | 2021-09-10 | 合肥工业大学智能制造技术研究院 | Pollutant concentration inversion method based on multi-feature selection algorithm |
| CN115422703A (en) * | 2022-07-19 | 2022-12-02 | 南京航空航天大学 | Surface thermal infrared emissivity inversion method based on MODIS data and Transformer network |
Non-Patent Citations (2)
| Title |
|---|
| 毛克彪;唐华俊;李丽英;许丽娜;: "一个从MODIS数据同时反演地表温度和发射率的神经网络算法", 遥感信息, no. 04, pages 9 - 16 * |
| 虞浩跃;沈韬;朱艳;刘英莉;余正涛;: "基于双向长短期记忆网络的太赫兹光谱识别", 光谱学与光谱分析, no. 12, pages 91 - 96 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116384237B (en) | 2024-08-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109508655B (en) | SAR target recognition method based on incomplete training set of twin network | |
| Saralioglu et al. | Semantic segmentation of land cover from high resolution multispectral satellite images by spectral-spatial convolutional neural network | |
| US11720727B2 (en) | Method and system for increasing the resolution of physical gridded data | |
| US20230153615A1 (en) | Neural network distillation method and apparatus | |
| CN111814607B (en) | Deep learning model suitable for small sample hyperspectral image classification | |
| CN112183718A (en) | Deep learning training method and device for computing equipment | |
| CN107563497A (en) | Computing device and method | |
| CN111797970B (en) | Method and device for training neural network | |
| CN110472280B (en) | Power amplifier behavior modeling method based on generation of antagonistic neural network | |
| Maier et al. | Exploding the myths: An introduction to artificial neural networks for prediction and forecasting | |
| Gad et al. | A robust deep learning model for missing value imputation in big NCDC dataset | |
| CN114912673A (en) | Water level prediction method based on whale optimization algorithm and long-term and short-term memory network | |
| Mai et al. | Optimization of interval type-2 fuzzy system using the PSO technique for predictive problems | |
| Anochi et al. | Two geoscience applications by optimal neural network architecture | |
| Mukherjee et al. | Identification of the types of disease for tomato plants using a modified gray wolf optimization optimized MobileNetV2 convolutional neural network architecture driven computer vision framework | |
| Rivas-Perea et al. | Statistical and neural pattern recognition methods for dust aerosol detection | |
| JP2019096237A (en) | System, method, program and storage medium storing the program for generating emulation model and generating and displaying information on uncertainty of target variable of the generated emulation model | |
| CN114444657A (en) | Image processing method, system, equipment and readable storage medium | |
| CN116384237B (en) | Thermal infrared atmospheric parameter inversion method and device and electronic equipment | |
| CN117830701A (en) | Attention mechanism-based multiscale feature fusion star map identification method and device | |
| Ghosh et al. | Deep learning enabled surrogate model of complex food processes for rapid prediction | |
| Kuter et al. | Modern applied mathematics for alternative modeling of the atmospheric effects on satellite images | |
| CN116958709A (en) | Hyperspectral classification method of lightweight tensor convolution long-short-term memory network | |
| CN116822716A (en) | Typhoon prediction method, system, equipment and medium based on space-time attention | |
| Tao et al. | Error variance estimation for individual geophysical parameter retrievals |
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 |