CN117951485A - Temperature profile inversion method based on deep learning - Google Patents
Temperature profile inversion method based on deep learning Download PDFInfo
- Publication number
- CN117951485A CN117951485A CN202410332386.XA CN202410332386A CN117951485A CN 117951485 A CN117951485 A CN 117951485A CN 202410332386 A CN202410332386 A CN 202410332386A CN 117951485 A CN117951485 A CN 117951485A
- Authority
- CN
- China
- Prior art keywords
- data
- temperature profile
- inversion
- temperature
- channel
- 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 80
- 238000013135 deep learning Methods 0.000 title claims abstract description 24
- 230000006870 function Effects 0.000 claims abstract description 30
- 238000012549 training Methods 0.000 claims abstract description 22
- 238000000605 extraction Methods 0.000 claims abstract description 9
- 238000010606 normalization Methods 0.000 claims abstract description 9
- 238000012545 processing Methods 0.000 claims abstract description 8
- 230000007246 mechanism Effects 0.000 claims abstract description 7
- 238000004590 computer program Methods 0.000 claims description 10
- 230000004913 activation Effects 0.000 claims description 6
- 239000013598 vector Substances 0.000 claims description 6
- 238000013507 mapping Methods 0.000 claims description 4
- 230000008569 process Effects 0.000 claims description 4
- 238000005259 measurement Methods 0.000 claims description 3
- 238000011176 pooling Methods 0.000 claims description 3
- 238000012360 testing method Methods 0.000 description 6
- 238000004458 analytical method Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 5
- 238000012795 verification Methods 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 3
- 238000013527 convolutional neural network Methods 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000000053 physical method Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
Classifications
-
- 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
- Radiation Pyrometers (AREA)
Abstract
The invention discloses a temperature profile inversion method based on deep learning, which comprises the following steps: 1. collecting GIIRS long-wave channel bright temperature data, exploring observation data and ERA5 temperature profile data, and adjusting the matching of the data in space-time; 2. constructing a deep learning training sample, and carrying out normalization processing on the numerical values in the deep learning training sample; 3. and constructing a temperature profile inversion model, setting a loss function, and training the temperature profile inversion model to obtain a trained temperature profile inversion model. The invention combines the characteristics of high space time and vertical resolution of the infrared hyperspectral bright temperature data of the stationary satellite, utilizes the priori acquired channel information to construct a channel attention mechanism to guide the inversion model feature extraction, more accords with the physical rule through the inversion result of the embedded guide model of the priori physical information, adopts the depth residual network to deeply excavate the nonlinear relation between the infrared hyperspectral bright temperature data and the atmospheric temperature profile, and improves the accuracy of the inversion temperature profile.
Description
Technical Field
The invention belongs to the technical field of atmospheric science, and particularly relates to a temperature profile inversion method based on deep learning.
Background
The temperature profile is an indispensable parameter for describing the atmospheric thermodynamic state, and plays an important role in atmospheric science research applications such as numerical weather forecast, strong convection weather forecast and analysis and the like. Therefore, the acquisition of an accurate atmospheric temperature profile is of great importance.
The temperature profile can be obtained by inversion of infrared hyperspectral data, and the current inversion method mainly comprises a physical method, a statistical regression method and a machine learning method. The above method has met with some success in the field of temperature profile inversion, but has some drawbacks. The statistical regression method processes the bright temperature data by using a statistical method, and the algorithm is stable and simple, but the description capability of the nonlinear relation between the satellite observed bright temperature and the atmospheric parameter is poor. The physical inversion method is complex to calculate, takes a long time and depends on the accuracy of the initial background field. The BP neural network algorithm in the existing machine learning method is widely applied to inversion of the atmospheric temperature profile, and a 3-layer BP neural network inversion temperature profile is generally selected for reducing inversion complexity and inversion time. But the BP neural network model has simple structure and poor network generalization capability. The convolutional neural network is used as one of the representative algorithms of deep learning and is preliminarily applied in the inversion of the atmospheric temperature profile, has better nonlinear mapping capability, is driven by data completely, lacks certain physical information constraint, and meanwhile, a reference model of a convolutional neural network method with residual connection is not generally introduced in the existing inversion research, so that the depth and feature extraction capability of the convolutional neural network are limited to influence the inversion precision.
Disclosure of Invention
The invention aims to: the invention aims to provide a temperature profile inversion method based on deep learning. A priori channel information attention mechanism is fused, a satellite observation channel attention mechanism is constructed by utilizing priori temperature inversion sensitive channel information in an inversion model, the inversion model is guided to pay more attention to feature extraction of the partial channel information, a depth residual error network is constructed to perform feature extraction, the higher inversion precision is tested through effectively training a deeper network model, sounding data and ERA5 analysis data are fused when a training model data set is constructed, the loss function of the inversion model is further optimized, and the inversion precision of the model is improved.
The technical scheme is as follows: the invention discloses a temperature profile inversion method based on deep learning, which comprises the following steps:
step 1, collecting GIIRS long-wave channel bright temperature data, exploring observation data and ERA5 temperature profile data, and adjusting the matching of the data in time and space to obtain adjusted data;
Step 2, based on the adjusted data, constructing a deep learning training sample, performing linear interpolation processing on the exploratory observation data, replacing the 1-1000 hpa air pressure layer missing measurement data with ERA5 temperature profile data, and performing normalization processing on the numerical values in the deep learning training sample;
And 3, constructing a temperature profile inversion model based on the deep learning training sample, taking GIIRS long-wave channel bright temperature data as input, taking the temperature profile data as output, setting a loss function, and training the temperature profile inversion model to obtain a trained temperature profile inversion model.
Further, the step 1 specifically includes: collecting GIIRS long-wave channel bright temperature data, sounding observation data and ERA5 temperature profile data, matching the longitude and latitude of a sounding site with the longitude and latitude of a long-wave infrared appearance observation field of FY-4A/GIIRS by adopting a distance threshold method, wherein the formula is defined as follows:
Wherein the method comprises the steps of And/>Respectively represent GIIRS observed field longitude and latitude,/>And/>And respectively representing the longitude and latitude of the sounding site, wherein R represents the earth radius, extracting GIIRS long-wave channel bright temperature data and sounding observation data which are successfully matched, and collecting nearest neighbor ERA5 temperature profile data.
Further, in step 2, the normalization processing is performed on the numerical value in the deep learning training sample, specifically: the adopted methods are maximum and minimum normalization methods, and the formulas are as follows:
the Min represents the minimum value in the sample, including the minimum value of the brightness temperature in the GIIRS long-wave channel brightness temperature data and the minimum value of the temperature in the label data, and the Max corresponds to the maximum value of the two.
Further, the step 3 specifically includes the following steps:
step 3.1 GIIRS long-wave channel bright temperature data are original input data, the original input data are subjected to a channel attention module, the channel attention module comprises two layers of 1×7 convolutions, a vector X= [1,0, 1..1 ] with the same dimension as the original input is set, 1 and 0 respectively represent a priori selected temperature profile inversion sensitive channel and a priori selected insensitive channel, sensitive inversion sensitive channel information can be selected based on a method of accessing a weight function peak value, the parameter output by the channel attention module of the original input and the vector calculate cross entropy loss, and the optimized parameter are subjected to a sigmoid activation function operation and then subjected to a dot multiplication operation with the original input, and the process can be represented by the following formula:
Where F represents the input data and where, Convolution operation representing a convolution kernel of 1 x 7,/>Representing a sigmoid function,/>Representing a dot product operation,/>Representing a ReLU function,/>Representing a learnable weight parameter,/>Representing the data characteristics after the attention mechanism enhancement;
step 3.2, through the feature extraction module, totally comprise 4 stages, each stage comprises 2 residual modules and one layer of pooling operation, the feature map channel output by each stage is respectively set to 64, 128, 256 and 512, the convolution kernel size is set to 1×3, each residual module comprises 2 convolution layers and one layer of residual connection operation, simultaneously, after each layer of convolution operation, nonlinear mapping capacity is increased by using a BN layer and finally using a ReLU as an activation function output, loss functions of an inversion model are optimized when inversion output and tag set loss are calculated, and finally the extracted features are output through two layers of complete connection layers.
Further, in step 3.2, the loss function of the optimized inversion model when the inversion output and the tag set loss are calculated is specifically: the loss of the temperature profile inversion model output and the exploratory observation data and ERA5 temperature profile data is calculated respectively, and different weights are given to the two losses, so that the inversion model accuracy is improved, and the overall loss function of the model is as follows:
Wherein, 、/>Respectively representing the loss of the output of the temperature profile inversion model and ERA5 temperature profile data and sounding observation data,/>Representing loss of channel attention module,/>And/>Respectively representing different weights.
The invention also discloses a computer device/equipment/system, comprising a memory, a processor and a computer program stored on the memory, wherein the processor executes the computer program to realize the steps of the method.
The invention also discloses a computer readable storage medium having stored thereon a computer program/instruction which when executed by a processor implements the steps of the method of the invention.
The invention also discloses a computer program product comprising computer programs/instructions which, when executed by a processor, implement the steps of the method of the invention.
The beneficial effects are that: compared with the prior art, the invention has the following remarkable advantages:
(1) The invention combines the characteristics of high space time and vertical resolution of the infrared hyperspectral bright temperature data of the stationary satellite, utilizes the priori acquired channel information to construct a channel attention mechanism to guide the inversion model feature extraction, more accords with the physical rule through the inversion result of the embedded guide model of the priori physical information, adopts the depth residual network to deeply excavate the nonlinear relation between the infrared hyperspectral bright temperature data and the atmospheric temperature profile, and improves the accuracy of the inversion temperature profile.
(2) According to the invention, sounding data and ERA5 data are fused when a training data set is constructed, so that a data set which fully meets the requirements of an inversion model and is more representative is effectively constructed, meanwhile, the loss function of the model is designed and optimized, and the generalization performance and the robustness of the inversion model are improved.
(3) Compared with the BPNN inversion method and the BRNN inversion method, the method has improved root mean square error and average error, and has more excellent inversion performance. In addition, the method has strong popularization and can be popularized in different areas and infrared hyperspectral data.
Drawings
FIG. 1 is a graph of a superparameter analysis, wherein (a) is a superparameterVerification set RMSE map at different values, (b) is a hyper-parameter/>Verification set RMSE map under different values;
FIG. 2 is an inverse model frame diagram;
FIG. 3 is a channel attention module diagram;
FIG. 4 is a schematic diagram of a residual block diagram;
FIG. 5 inverts the root mean square error and average deviation profile;
Fig. 6 inverts an example display diagram.
Detailed Description
The technical scheme of the invention is further described below with reference to the accompanying drawings.
The invention discloses a temperature profile inversion method based on deep learning, which comprises the following steps:
Step 1, collecting GIIRS long-wave channel bright-temperature data sounding observation data ERA5 temperature profile data and solving the problem of space-time matching of the data;
The invention collects GIIRS long wave channel bright temperature data of 6 months to 8 months in 2020 and 6 months to 8 months in 2021, observation data of 120 sounding sites in China and ERA5 temperature profile data at the same time, screens GIIRS long wave channel bright temperature data and ERA5 temperature profile data according to sounding observation time, and matches the longitude and latitude of the sounding site with the longitude and latitude of a long wave infrared appearance sounding field of FY-4A/GIIRS by adopting a distance threshold method in the space matching problem, wherein the formula is as follows:
Wherein the method comprises the steps of And/>Respectively represent GIIRS observed field longitude and latitude,/>And/>And respectively representing the longitude and latitude of the sounding site, wherein R represents the earth radius, extracting GIIRS long-wave channel bright temperature data and sounding observation data which are successfully matched, and collecting nearest neighbor ERA5 temperature profile data. The matching is successful for a total of 89 sites.
Step 2, constructing a deep learning training sample, respectively carrying out normalization processing on input data and output data, and dividing a training set, a verification set and a test set according to time;
In order to meet the requirement of an inversion model on consistent output dimension, different from the samples with inconsistent dimension which are directly removed in the prior study, the invention replaces the sounding observation data which is still under the measurement of the 1-1000 hpa air pressure layer with ERA5 temperature profile data, provides more samples for inversion model training, assimilates physical information of a large number of satellite observation and sounding observation data by using ERA5 data, and can provide more physical constraints for the model. In order to accelerate the convergence rate of the model, the numerical values of all samples in the data set are required to be normalized, and the method adopted in the invention is a maximum and minimum normalization method, and the formula is as follows:
The Min represents the minimum value in the sample, including the minimum value of the brightness temperature in the GIIRS long-wave channel brightness temperature data and the minimum value of the temperature in the label data, and the Max corresponds to the maximum value of the two. Data of 6-8 months in 2020 are used as model training samples, data of 6 months in 2021 are used as verification, and 7 months to 8 months in 2021 are used as test samples.
Step 3, constructing a temperature profile inversion model based on a deep learning method;
The model integral framework of the inversion method provided by the invention is shown in fig. 2, the total of 689 channel bright temperatures of GIIRS long-wave channel observation bright temperatures are taken as original input data, the channel attention module designed by the invention (shown in fig. 3) comprises two layers of 1×7 convolutions, meanwhile, vectors X= [1,0, 1..1.) which are the same as the original input dimensions are set, sensitive inversion sensitive channel information can be selected on the basis of a method for accessing weight function peak values to have definite physical meanings, and 1 and 0 respectively represent a priori selected temperature profile inversion sensitive channel and a priori selected insensitive channel. The parameters output by the channel attention module of the original input and the vector calculate cross entropy loss, and the optimized parameters are subjected to sigmoid activation function operation to perform dot multiplication operation with the original input, and the process can be expressed by the following formula:
Where F represents the input data and where, Convolution operation representing a convolution kernel of 1 x 7,/>The sigmoid function is represented as a function,Representing a dot product operation,/>Representing a ReLU function,/>Representing a learnable weight parameter,/>Representing the data characteristics after the attention mechanism enhancement;
The characteristics after data enhancement are subjected to a characteristic extraction module and comprise 4 stages in total, wherein each stage comprises 2 residual error modules and one layer of pooling; the feature map channels output in each stage are respectively set to 64, 128, 256 and 512, the feature width of each pooled layer is halved after the feature map channel in each stage is doubled, as shown in fig. 4, each residual module comprises 2 convolution layers and one layer of residual connection operation, so as to avoid the gradient vanishing problem caused by the over-deep network layer number, and the nonlinear mapping capability is increased by using the final ReLU as an activation function output after each layer of convolution operation through the BN layer. And finally outputting a temperature profile through two layers of complete connecting layers after the feature extraction is finished. According to the invention, the loss function of the inversion model is optimized when the loss of the inversion output and the tag set is calculated, the loss of the inversion output, the loss of the sounding data and the loss of the ERA5 data are calculated respectively in order to embody the physical constraint of the ERA5 data on the model, and different weights are given to the two losses, so that the accuracy of the inversion model is improved. The overall model loss function is as follows:
Wherein, 、/>Respectively representing the loss of the output of the temperature profile inversion model and ERA5 temperature profile data and sounding observation data,/>Representing loss of channel attention module,/>And/>Respectively representing different weights. . The network optimizer selects Adam, the iteration times are set to 1000 times, the learning rate is set to 0.001, after the training convergence of the inversion model, the test sample to be inverted is input into the training convergence inversion model, and the sounding observation data is used for checking the inversion precision.
Experimental results
In order to verify the performance of the inversion method of the invention, experimental contrast inversion accuracy was performed with the most widely used BP neural network and another deep learning inversion method BRNN. The present invention uses Root Mean Square Error (RMSE) and average error (MB) to evaluate inversion performance as follows:
Wherein, The temperature value representing the inversion may also be referred to as a model predictive value,/>And the real temperature value is represented, and N is the number of test samples.
Super parameter analysis
Since the super parameters λ and β are introduced into the loss function of the present invention, in order to verify the influence of the super parameters on the performance of the model, the present invention makes experiments in a preset parameter set [0.1,1,10,50,100], and selects the super parameters using RMSE calculated in the verification set by the model as an index, it can be seen from (a) in fig. 1 that RMSE is minimum at λ=50. Similarly, after determining the super parameter λ, β is selected in the same way, and when β=10, as can be seen in (b) of fig. 1, RMSE calculated by the validation set is minimum, so that it is reasonable to set λ to 50 and β to 10 in the subsequent test.
Table 1 gives the root mean square error and average error comparisons for the three methods. It can be seen that the method of the invention achieves better inversion performance in both root mean square error and average deviation.
Table 1 comparison of root mean square error and average error for the three methods
Root Mean Square Error (RMSE) | Average error (MB) | |
The method of the invention | 2.06K | 0.072K |
BP inversion method | 2.42K | 0.255K |
BRNN inversion method | 2.30K | -0.076K |
Fig. 5 shows the root mean square error and the average error of the 3 methods at different air pressure layers, the horizontal axis label represents the root mean square error and the average error, the unit is K, and the vertical axis represents air pressure. The method 1 shows the result of the method of the invention, and from the figure, we can see that the RMSE of the proposed method 1 at different air pressure layers is better than the other two methods, especially at 125-650 hpa. The RMSE of method 1 is within 2K in 125-650 HPA, the RMSE is optimally 1.55K at 500HPA, the RMSE is not more than 3.1K at 900HPA, the RMSE of BPNN is relatively close to that of BRNN, and the RMSE of BRNN is obviously better than that of BPNN at 200-500 HPA. MB of the method 1 is positive deviation when 250-900 hpa, 100-250 hpa shows negative deviation, and the method 1 is better than MB of the BPNN when 100-500 hpa, and the MB of the three methods is closer to 0K as a whole, so that the inversion method provided by the invention is better, and the method provided by the invention is more stable in performance when inverting the atmospheric temperature profile.
In order to further verify the effectiveness of inversion temperature profile of the method, inversion samples are selected from the test samples for comparison analysis. Fig. 6 shows that when the sounding data of 51431 sites are compared with inversion temperature profiles in the steps of 00UTC and 12UTC of 7 months of 2021, the abscissa shows the temperature, and the ordinate shows the air pressure, and the method 1 shows the method result of the invention, as can be seen from the figure, the inversion result obtained by the method of the invention fits well with the sounding data in different air pressure layers at 00UTC, and more accords with the physical rule, especially at 150-250 hpa, the inversion result obtained by the method of the invention captures the change trend of the temperature well, and the inversion results of the BPNN inversion method and BRNN method can not reflect the change accurately. Meanwhile, the inversion accuracy of the BP inversion method is not high enough compared with other two methods in the inversion result at 600-900 hpa. The inversion result of the method disclosed by the invention can be obviously and well fit with sounding data at 12UTC, and particularly at 800-900 hpa, the retrieval result of the method is superior to BP inversion method and BRNN method. In general, the accuracy of the method of the present invention in inverting the atmospheric temperature profile at 00UTC and 12UTC is higher than the other two methods.
Claims (8)
1. The temperature profile inversion method based on deep learning is characterized by comprising the following steps of:
step 1, collecting GIIRS long-wave channel bright temperature data, exploring observation data and ERA5 temperature profile data, and adjusting the matching of the data in time and space to obtain adjusted data;
Step 2, based on the adjusted data, constructing a deep learning training sample, performing linear interpolation processing on the exploratory observation data, replacing the 1-1000 hpa air pressure layer missing measurement data with ERA5 temperature profile data, and performing normalization processing on the numerical values in the deep learning training sample;
And 3, constructing a temperature profile inversion model based on the deep learning training sample, taking GIIRS long-wave channel bright temperature data as input, taking the temperature profile data as output, setting a loss function, and training the temperature profile inversion model to obtain a trained temperature profile inversion model.
2. The temperature profile inversion method based on deep learning of claim 1, wherein step 1 specifically comprises: collecting GIIRS long-wave channel bright temperature data, sounding observation data and ERA5 temperature profile data, matching the longitude and latitude of a sounding site with the longitude and latitude of a long-wave infrared appearance observation field of FY-4A/GIIRS by adopting a distance threshold method, wherein the formula is defined as follows:
Wherein the method comprises the steps of And/>Respectively represent GIIRS observed field longitude and latitude,/>And/>And respectively representing the longitude and latitude of the sounding site, wherein R represents the earth radius, extracting GIIRS long-wave channel bright temperature data and sounding observation data which are successfully matched, and collecting nearest neighbor ERA5 temperature profile data.
3. The temperature profile inversion method based on deep learning according to claim 1, wherein in step 2, the normalization processing is performed on the values in the deep learning training samples, specifically: the adopted methods are maximum and minimum normalization methods, and the formulas are as follows:
the Min represents the minimum value in the sample, including the minimum value of the brightness temperature in the GIIRS long-wave channel brightness temperature data and the minimum value of the temperature in the label data, and the Max corresponds to the maximum value of the two.
4. The temperature profile inversion method according to claim 1, wherein the step3 specifically comprises the following steps:
Step 3.1, GIIRS long-wave channel bright temperature data are original input data, the original input data are passed through a channel attention module, the channel attention module comprises two layers of 1×7 convolutions, a vector X= [1,0, 1..1 ] which is the same as the original input dimension is set, 1 and 0 respectively represent a priori selected temperature profile inversion sensitive channel and a priori selected insensitive channel, sensitive inversion sensitive channel information can be selected based on a method of passing through a weight function peak value, cross entropy loss is calculated between parameters output by the channel attention module and the vector by the original input, and point multiplication operation is carried out on optimized parameters and the original input after the operation of a sigmoid activation function, and the process can be represented by the following formula:
Where F represents the input data and where, Convolution operation representing a convolution kernel of 1 x 7,/>Representing a sigmoid function,/>Representing a dot product operation,/>Representing a ReLU function,/>Representing a learnable weight parameter,/>Representing the data characteristics after the attention mechanism enhancement;
step 3.2, through the feature extraction module, totally comprise 4 stages, each stage comprises 2 residual modules and one layer of pooling operation, the feature map channel output by each stage is respectively set to 64, 128, 256 and 512, the convolution kernel size is set to 1×3, each residual module comprises 2 convolution layers and one layer of residual connection operation, simultaneously, after each layer of convolution operation, nonlinear mapping capacity is increased by using a BN layer and finally using a ReLU as an activation function output, loss functions of an inversion model are optimized when inversion output and tag set loss are calculated, and finally the extracted features are output through two layers of complete connection layers.
5. The temperature profile inversion method according to claim 4, wherein in step 3.2, the loss function of the optimized inversion model when calculating the inversion output and the tag set loss is specifically: the loss of the temperature profile inversion model output and the exploratory observation data and ERA5 temperature profile data is calculated respectively, and different weights are given to the two losses, so that the inversion model accuracy is improved, and the overall loss function of the model is as follows:
Wherein, 、/>Respectively representing the loss of the output of the temperature profile inversion model and ERA5 temperature profile data and sounding observation data,/>Representing loss of channel attention module,/>And/>Respectively representing different weights.
6. A computer apparatus/device/system comprising a memory, a processor and a computer program stored on the memory, wherein the processor executes the computer program to perform the steps of the method of claim 1.
7. A computer readable storage medium having stored thereon a computer program/instruction which when executed by a processor performs the steps of the method of claim 1.
8. A computer program product comprising computer programs/instructions which, when executed by a processor, implement the steps of the method of claim 1.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202410332386.XA CN117951485B (en) | 2024-03-22 | 2024-03-22 | Temperature profile inversion method based on deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202410332386.XA CN117951485B (en) | 2024-03-22 | 2024-03-22 | Temperature profile inversion method based on deep learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN117951485A true CN117951485A (en) | 2024-04-30 |
CN117951485B CN117951485B (en) | 2024-06-14 |
Family
ID=90792364
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202410332386.XA Active CN117951485B (en) | 2024-03-22 | 2024-03-22 | Temperature profile inversion method based on deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN117951485B (en) |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114880933A (en) * | 2022-05-13 | 2022-08-09 | 南京星图空间信息科技有限公司 | Atmospheric temperature and humidity profile inversion method and system for non-exploration-site foundation microwave radiometer based on reanalysis data |
CN115755095A (en) * | 2021-12-14 | 2023-03-07 | 中国气象科学研究院 | Satellite-ground synchronous inversion method for mesoscale atmospheric dynamic parameters |
CN115859789A (en) * | 2022-11-23 | 2023-03-28 | 北京信息科技大学 | Method for improving inversion accuracy of polar atmosphere temperature profile |
CN116485894A (en) * | 2023-04-24 | 2023-07-25 | 北京易航远智科技有限公司 | Video scene mapping and positioning method and device, electronic equipment and storage medium |
CN116580017A (en) * | 2023-05-26 | 2023-08-11 | 重庆邮电大学 | Improved Mask-R-CNN lung nodule auxiliary detection method integrating double-path channel attention and cavity space attention |
WO2023207579A1 (en) * | 2022-04-29 | 2023-11-02 | 中国科学技术大学 | Method for detecting traffic pollution source via horizontal distribution of trace gas on basis of hyperspectral remote sensing |
CN117576441A (en) * | 2023-10-26 | 2024-02-20 | 西北工业大学 | Lightweight algorithm for steel surface defect detection by fusing classification and detection |
-
2024
- 2024-03-22 CN CN202410332386.XA patent/CN117951485B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115755095A (en) * | 2021-12-14 | 2023-03-07 | 中国气象科学研究院 | Satellite-ground synchronous inversion method for mesoscale atmospheric dynamic parameters |
WO2023207579A1 (en) * | 2022-04-29 | 2023-11-02 | 中国科学技术大学 | Method for detecting traffic pollution source via horizontal distribution of trace gas on basis of hyperspectral remote sensing |
CN114880933A (en) * | 2022-05-13 | 2022-08-09 | 南京星图空间信息科技有限公司 | Atmospheric temperature and humidity profile inversion method and system for non-exploration-site foundation microwave radiometer based on reanalysis data |
CN115859789A (en) * | 2022-11-23 | 2023-03-28 | 北京信息科技大学 | Method for improving inversion accuracy of polar atmosphere temperature profile |
CN116485894A (en) * | 2023-04-24 | 2023-07-25 | 北京易航远智科技有限公司 | Video scene mapping and positioning method and device, electronic equipment and storage medium |
CN116580017A (en) * | 2023-05-26 | 2023-08-11 | 重庆邮电大学 | Improved Mask-R-CNN lung nodule auxiliary detection method integrating double-path channel attention and cavity space attention |
CN117576441A (en) * | 2023-10-26 | 2024-02-20 | 西北工业大学 | Lightweight algorithm for steel surface defect detection by fusing classification and detection |
Non-Patent Citations (5)
Title |
---|
HONGKUN WANG等: "Retrieval of Atmospheric Temperature Profile from Historical Data and Ground-Based Observations by Using a Machine Learning Algorithm", 《REMOTE SENSING》, 24 May 2023 (2023-05-24), pages 1 - 18 * |
RENLONG HANG等: "Hyperspectral Image Classification with Attention Aided CNNs", 《IEEE TRANSACTIONS ON GEESCIENCE AND REMOTE SENSING 》, vol. 59, no. 03, 16 July 2020 (2020-07-16), pages 1 - 12 * |
徐桂荣等: "青藏高原东侧九龙夏季非降水云的观测特征", 《暴雨灾害》, 21 February 2024 (2024-02-21), pages 1 - 11 * |
杭仁龙等: "模态间匹配学习的高光谱和激光雷达联合分类", 《遥感学报》, vol. 28, no. 01, 25 January 2024 (2024-01-25), pages 154 - 167 * |
黄鹏宇等: "FY-4A/GIIRS资料云上温度廓线反演研究", 《 激光与光电子学进展》, vol. 58, no. 17, 10 September 2021 (2021-09-10), pages 125 - 133 * |
Also Published As
Publication number | Publication date |
---|---|
CN117951485B (en) | 2024-06-14 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110084221B (en) | Serialized human face key point detection method with relay supervision based on deep learning | |
CN111814611B (en) | Multi-scale face age estimation method and system embedded with high-order information | |
CN112016596B (en) | Farmland soil fertility evaluation method based on convolutional neural network | |
CN115146700B (en) | Runoff prediction method based on transform sequence-to-sequence model | |
CN111580151A (en) | SSNet model-based earthquake event time-of-arrival identification method | |
CN111915007A (en) | Magnetic resonance spectrum noise reduction method based on neural network | |
CN111598460A (en) | Method, device and equipment for monitoring heavy metal content in soil and storage medium | |
CN115937693A (en) | Road identification method and system based on remote sensing image | |
CN117253122B (en) | Corn seed approximate variety screening method, device, equipment and storage medium | |
CN114663777A (en) | Hyperspectral image change detection method based on spatio-temporal joint graph attention mechanism | |
CN117951485B (en) | Temperature profile inversion method based on deep learning | |
CN116740567A (en) | Soil moisture content detection method and system for Paulownia seedling cultivation | |
CN117221816A (en) | Multi-building floor positioning method based on Wavelet-CNN | |
CN113609809B (en) | Method, system, equipment, medium and terminal for diagnosing faults of radio frequency low-noise discharge circuit | |
CN115909332A (en) | Agricultural remote sensing image semantic segmentation method based on channel feature enhancement | |
CN113537573A (en) | Wind power operation trend prediction method based on dual space-time feature extraction | |
CN111914212B (en) | Neighborhood reconstruction method for data dimension reduction | |
CN114842300B (en) | Crop pest detection method suitable for rainy day environment | |
CN113743363B (en) | Shielded target identification method based on small sample of unmanned aerial vehicle system | |
CN113361417B (en) | Human behavior recognition method based on variable time sequence | |
CN114241491B (en) | Handwritten letter recognition method based on lightweight deep learning network | |
CN115758855B (en) | Surface reflectivity inversion method based on LSTM and attention mechanism | |
CN114663779B (en) | Multi-temporal hyperspectral image change detection method based on time-space-spectrum attention mechanism | |
CN117236201B (en) | Diffusion and ViT-based downscaling method | |
CN115267883B (en) | Earthquake response prediction model training and predicting method, system, equipment and medium |
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 |