Surface Temperature Inversion Method, Device, Computer Equipment and Readable Storage Medium

Abstract

The present application relates to a method, device, computer device, readable storage medium and program product for retrieving land surface temperature, which is applied to the field of remote sensing. The method includes: obtaining a multi-spectral image and a thermal infrared image of a to-be-detected area; the spatial resolutions of the multi-spectral image and the thermal infrared image are consistent; for each pixel point at the same position in the multi-spectral image and the thermal infrared image, based on the calibration coefficients of the respective imaging channels of the multi-spectral image and the thermal infrared image and the pixel values of the pixel point on the multi-spectral image and the thermal infrared image, an irradiance information group corresponding to the pixel point is obtained; based on the irradiance information groups corresponding to each pixel point, a land surface temperature image of the to-be-detected area is retrieved. Using this method can improve the retrieval accuracy of land surface temperature products.

Description

Technical Field

[0001] The present application relates to the field of remote sensing technology, and particularly to a method, device, computer equipment, computer-readable storage medium, and computer program product for retrieving land surface temperature. Background Art

[0002] In the field of land surface temperature retrieval, common land surface temperature retrieval algorithms can retrieve land surface temperature information from satellite remote sensing observation data under certain assumptions. These algorithms mainly include the single-window algorithm, split-window algorithm, multi-angle algorithm, day / night algorithm, temperature-emissivity separation algorithm, etc. However, these algorithms require at least one of the land surface emissivity, atmospheric water vapor content (WVC), and atmospheric parameters (atmospheric transmittance, up / downward atmospheric mean radiation) as known prerequisites to retrieve the land surface temperature. At the same time, the above three also need to be retrieved from satellite remote sensing observation data. Therefore, error propagation is bound to occur.

[0003] Therefore, the requirements of traditional algorithms for prior conditions limit the accuracy of land surface temperature retrieval. For different input parameters required by the algorithms, the error propagation during the solution process will further affect the final retrieval quality of the land surface temperature. Summary of the Invention

[0004] Based on this, it is necessary to provide a method, device, computer equipment, computer-readable storage medium, and computer program product for retrieving land surface temperature to solve the technical problem of poor retrieval quality existing in the above-mentioned retrieval algorithms.

[0005] In a first aspect, the present application provides a method for retrieving land surface temperature. The method includes:

[0006] Obtain a multi-spectral image and a thermal infrared image of the area to be measured; the spatial resolutions of the multi-spectral image and the thermal infrared image are the same;

[0007] For each pixel at the same position in the multi-spectral image and the thermal infrared image, obtain a group of radiance information corresponding to the pixel according to the calibration coefficients of the imaging channels corresponding to the multi-spectral image and the thermal infrared image and the pixel values of the pixel in the multi-spectral image and the thermal infrared image;

[0008] Based on the groups of radiance information corresponding to each pixel, retrieve a land surface temperature image of the area to be measured.

[0009] In one embodiment, the retrieving a land surface temperature image of the area to be measured based on the groups of radiance information corresponding to each pixel includes:

[0010] Input the radiance information groups corresponding to each of the pixel points into a trained land surface temperature inversion model for processing to obtain the land surface temperature corresponding to each of the pixel points;

[0011] Based on the land surface temperature corresponding to each of the pixel points, obtain the land surface temperature image of the area to be measured.

[0012] In one embodiment, the input variables of the land surface temperature inversion model are determined by the following method:

[0013] Determine a first relational expression for land surface temperature inversion; the input variables in the first relational expression include the brightness temperature of the thermal infrared channel, the atmospheric transmittance, and the land surface emissivity;

[0014] Convert the land surface emissivity of the thermal infrared channel to be characterized by the radiance from visible light to near-infrared and short-wave infrared to obtain a first function, and convert the atmospheric transmittance and brightness temperature of the thermal infrared channel to be characterized by the radiance to obtain a second function;

[0015] According to the first function and the second function, adjust the first relational expression to obtain a target relational expression for land surface temperature inversion, and use the input variables in the target relational expression as the input variables of the land surface temperature inversion model.

[0016] In one embodiment, the step of adjusting the first relational expression according to the first function and the second function to obtain a target relational expression for land surface temperature inversion includes:

[0017] Adjust the first relational expression according to the first function to obtain a second relational expression for land surface temperature inversion;

[0018] Adjust the second relational expression according to the second function to obtain a third relational expression for land surface temperature inversion;

[0019] Perform a simplification process on the input variables in the third relational expression by using the principal component analysis method to obtain a target relational expression for land surface temperature inversion.

[0020] In one embodiment, the step of obtaining the multi-spectral image and the thermal infrared image of the area to be measured includes:

[0021] Obtain an initial multi-spectral image and an initial thermal infrared image of the area to be measured;

[0022] Identify the overlapping area between the initial multi-spectral image and the initial thermal infrared image;

[0023] Based on the overlapping area, perform matching on the initial multi-spectral image and the initial thermal infrared image to obtain the multi-spectral image and the thermal infrared image of the area to be measured.

[0024] In one embodiment, identifying the overlapping area between the initial multispectral image and the initial thermal infrared image includes:

[0025] Converting the pixel coordinates of the initial multispectral image and the initial thermal infrared image into longitude and latitude coordinates respectively;

[0026] Determining the positioning information of the matching area of the initial multispectral image and the initial thermal infrared image under the longitude and latitude coordinates; the positioning information includes the longitude and latitude coordinates of the head and tail pixels of the matching area;

[0027] Converting the longitude and latitude coordinates of the head and tail pixels back to the pixel coordinates of the initial multispectral image and the initial thermal infrared image to obtain the overlapping area.

[0028] In a second aspect, the present application also provides a device for retrieving surface temperature. The device includes:

[0029] An image acquisition module for acquiring a multispectral image and a thermal infrared image of a region to be measured; the spatial resolutions of the multispectral image and the thermal infrared image are the same;

[0030] A data processing module for obtaining a set of radiance information corresponding to each pixel point at the same position in the multispectral image and the thermal infrared image according to the calibration coefficients of the imaging channels corresponding to the multispectral image and the thermal infrared image respectively and the pixel values of the pixel point on the multispectral image and the thermal infrared image;

[0031] A temperature inversion module for inversely obtaining the surface temperature image of the region to be measured based on the set of radiance information corresponding to each pixel point.

[0032] In a third aspect, the present application also provides a computer device. The computer device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0033] Acquiring a multispectral image and a thermal infrared image of a region to be measured; the spatial resolutions of the multispectral image and the thermal infrared image are the same;

[0034] For each pixel point at the same position in the multispectral image and the thermal infrared image, obtaining a set of radiance information corresponding to the pixel point according to the calibration coefficients of the imaging channels corresponding to the multispectral image and the thermal infrared image respectively and the pixel values of the pixel point on the multispectral image and the thermal infrared image;

[0035] Based on the set of radiance information corresponding to each pixel point, inversely obtaining the surface temperature image of the region to be measured.

[0036] In a fourth aspect, the present application also provides a computer-readable storage medium. On the computer-readable storage medium, there is stored a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0037] Obtain a hyperspectral image and a thermal infrared image of the area to be measured; the spatial resolutions of the hyperspectral image and the thermal infrared image are the same;

[0038] For each pixel point at the same position in the hyperspectral image and the thermal infrared image, according to the calibration coefficients of the imaging channels corresponding to the hyperspectral image and the thermal infrared image respectively and the pixel values of the pixel point in the hyperspectral image and the thermal infrared image, obtain the radiance information group corresponding to the pixel point;

[0039] Based on the radiance information groups corresponding to each of the pixel points, inversely calculate to obtain the surface temperature image of the area to be measured.

[0040] In a fifth aspect, the present application also provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0041] Obtain a hyperspectral image and a thermal infrared image of the area to be measured; the spatial resolutions of the hyperspectral image and the thermal infrared image are the same;

[0042] For each pixel point at the same position in the hyperspectral image and the thermal infrared image, according to the calibration coefficients of the imaging channels corresponding to the hyperspectral image and the thermal infrared image respectively and the pixel values of the pixel point in the hyperspectral image and the thermal infrared image, obtain the radiance information group corresponding to the pixel point;

[0043] Based on the radiance information groups corresponding to each of the pixel points, inversely calculate to obtain the surface temperature image of the area to be measured.

[0044] The above-mentioned surface temperature inversion method, device, computer equipment, storage medium and computer program product obtain the multi-spectral image and thermal infrared image of the area to be measured, and process the spatial resolutions of the multi-spectral image and the thermal infrared image to be consistent, so that the pixel points on the two images correspond one by one. Further, for each pixel point at the same position in the multi-spectral image and the thermal infrared image, according to the calibration coefficients of the imaging channels corresponding to the multi-spectral image and the thermal infrared image respectively and the pixel values of the pixel point on the multi-spectral image and the thermal infrared image, an irradiance information group corresponding to the pixel point is obtained. Finally, based on the irradiance information groups corresponding to each pixel point, a surface temperature image of the area to be measured is inversely obtained. This method models the relationship between the irradiance in the visible to near-infrared and thermal infrared bands and the surface temperature, avoiding the dependence of surface temperature inversion on prior conditions such as surface emissivity, atmospheric water vapor content (WVC) and atmospheric parameters, getting rid of the bondage of prior information on the surface temperature inversion algorithm, and can inversely obtain the surface temperature without any prior conditions. At the same time, since the irradiance information is directly used as the input of the inversion, the surface temperature can be inversely obtained from the remote sensing image obtained from a single source of hyperspectral observation satellites only, without being converted into indirect parameters such as apparent reflectance, normalized difference vegetation index and vegetation coverage, avoiding the error transfer in the traditional surface temperature inversion algorithm, and making full use of the advantages of multi-channel satellite remote sensing observation data, thereby improving the accuracy of the inversion result of the surface temperature product. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a schematic flowchart of the surface temperature inversion method in an embodiment;

[0046] Figure 2 It is a schematic diagram of the construction process of the simulation data set in an embodiment;

[0047] Figure 3 It is a schematic diagram of the test result of the neural network structure in an embodiment;

[0048] Figure 4 It is a schematic flowchart of the derivation process of the input variables of the surface temperature inversion model in an embodiment;

[0049] Figure 5 It is a schematic diagram of the analysis result of the principal component analysis method on the generated simulation data set in an embodiment;

[0050] Figure 6 It is a schematic diagram of the analysis result of the principal component analysis method on the generated simulation data set in another embodiment;

[0051] Figure 7 It is a schematic diagram of the structure of the neural network in an embodiment;

[0052] Figure 8Schematic flowchart of the matching process of multi-spectral image and thermal infrared image in an embodiment;

[0053] Figure 9 Schematic diagram of the observation ranges of the multi-spectral image and the thermal infrared image collected by a hyperspectral observation satellite in an embodiment;

[0054] Figure 10 Structural block diagram of a surface temperature inversion device in an embodiment;

[0055] Figure 11 Internal structure diagram of a computer device in an embodiment. Specific embodiments

[0056] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0057] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects and do not necessarily need to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein.

[0058] In the related art, the mainstream land surface temperature inversion algorithms (mainly including the single-window algorithm, split-window algorithm, multi-angle algorithm, day / night algorithm, temperature / emissivity separation algorithm, etc.) can invert the land surface temperature information from satellite remote sensing observation data under certain assumptions. Among them, the single-window algorithm simplifies the radiative transfer equation by exploring the empirical linear relationship between the atmospheric transmittance and the atmospheric water vapor content (WVC), as well as the average atmospheric temperature and the near-surface air temperature, and can invert the land surface temperature on the premise of knowing the near-surface air temperature and water vapor content. The split-window algorithm is based on the land surface thermal radiative transfer equation, and removes the atmospheric influence by using the different atmospheric absorptions of two adjacent channels. On the premise of knowing the land surface emissivity and water vapor of the adjacent channels, an equation is established to invert the land surface temperature. The multi-angle algorithm assumes that the land surface temperature is independent of the viewing zenith angle (VZA), and the atmosphere is horizontally uniform and stable during the observation time. Therefore, on the premise of knowing different viewing zenith angles and land surface emissivity, the land surface temperature can be inverted through precise geometric registration, but it is only applicable to homogeneous land surfaces. The day / night algorithm assumes that the land surface emissivity remains unchanged or changes very little during day and night. Therefore, it does not require land surface emissivity information as a priori, but requires precise atmospheric parameters as input for correction, the same viewing zenith angle during day and night, and multiple channels distributed in the mid-infrared and thermal infrared. The temperature / emissivity separation algorithm uses the normalized emissivity method (NEM) module to iteratively solve the initial land surface emissivity and land surface temperature estimates, and then restores the emissivity spectral shape through the ratio (RAT) module and restores the emissivity spectral amplitude through the maximum-minimum normalized emissivity difference (MMD) module to obtain the final land surface emissivity estimate, which requires precise atmospheric parameters as a priori.

[0059] However, whether it is the split-window algorithm, the temperature / emissivity separation algorithm, or other algorithms, at least one of the surface emissivity, atmospheric water vapor content (WVC), and atmospheric parameters (such as atmospheric transmittance, up / downwelling atmospheric mean radiance) must be known in advance. At the same time, the above three also need to be retrieved from satellite remote sensing observation data. Therefore, error propagation is bound to occur. For the split-window algorithm, an error of 0.01 in surface emissivity will result in an inversion error of 1-2 K in surface temperature, and the accuracy of the current optimal surface emissivity inversion method based on the normalized difference vegetation index is only better than 0.01. For the temperature / emissivity separation algorithm, accurate atmospheric parameters need to be calculated through a radiative transfer model under the premise of an accurate atmospheric profile, which requires very high requirements for satellite sensors. Even under the premise of accurate atmospheric correction, the theoretical accuracies of this algorithm for inverting surface temperature and surface emissivity are ±1.5 K and ±0.015 respectively, and for ground objects with low emissivity spectral contrast (water bodies, ice and snow, vegetation) and atmospheric conditions with high water vapor content, the errors are even greater. In summary, the requirements of prior conditions limit the accuracy of surface temperature inversion. For different input parameters required by the algorithm, the error propagation in the solution process will further affect the final quality of surface temperature inversion.

[0060] Based on the above problems, the present application proposes a method for improving the quality of surface temperature inversion. By modeling the relationship between the radiance in the visible to near-infrared and thermal infrared bands and the surface temperature based on a neural network, the dependence on prior conditions such as surface emissivity, atmospheric water vapor content (WVC), and atmospheric parameters in surface temperature inversion is avoided. At the same time, since the radiance information is directly used as the input of the neural network without being converted into indirect parameters such as apparent reflectance, normalized difference vegetation index, and vegetation cover, the error propagation in traditional surface temperature inversion algorithms is avoided, and the advantages of multi-channel satellite remote sensing observation data are fully utilized, thereby improving the inversion accuracy of surface temperature products.

[0061] Reference Figure 1 , is a schematic flow chart of a surface temperature inversion method provided by an embodiment of the present application. In this embodiment, an example is given where the method is applied to a terminal. It can be understood that the method can also be applied to a server and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0062] Step S110, obtaining a multi-spectral image and a thermal infrared image of the area to be measured; the spatial resolutions of the multi-spectral image and the thermal infrared image are the same.

[0063] Among them, the multi-spectral image represents an image obtained by capturing the reflected or radiated energy of the area to be measured through multiple discrete bands (specifically, visible to near-infrared), reflecting the reflection characteristics of the object.

[0064] Among them, the thermal infrared image represents an image generated by capturing the thermal radiation energy of the area to be measured in the thermal infrared band, reflecting the temperature distribution of the object.

[0065] In a specific implementation, for the area to be measured, an all-spectrum imager can be used to collect images in multiple discrete bands and the thermal infrared band, obtaining an initial multi-spectral image and an initial thermal infrared image. Further, the initial multi-spectral image and the initial thermal infrared image are matched pixel by pixel to adjust the spatial resolutions of the initial multi-spectral image and the initial thermal infrared image to be the same. Specifically, the initial multi-spectral image is downsampled to the same spatial resolution as the thermal infrared image.

[0066] Step S120: For each pixel point at the same position in the multi-spectral image and the thermal infrared image, according to the calibration coefficients of the imaging channels corresponding to the multi-spectral image and the thermal infrared image respectively and the pixel values of the pixel point on the multi-spectral image and the thermal infrared image, an irradiance information group corresponding to the pixel point is obtained.

[0067] It can be understood that for a multi-spectral image and a thermal infrared image with the same spatial resolution, the number of pixel points is the same and they correspond one by one. The multi-spectral image and the thermal infrared image each correspond to multiple imaging channels, and each imaging channel has a corresponding calibration coefficient. A pixel point can be converted into an irradiance value through the calibration coefficient of each imaging channel. These irradiance values can form an irradiance information group.

[0068] Specifically, the irradiance value of any pixel point in any imaging channel can be obtained through the following formula:

[0069] (1)

[0070] Among them, L is the irradiance value corresponding to the pixel; DN is the pixel value corresponding to the pixel; GAINS and OFFSET are radiation correction coefficients (i.e., calibration coefficients), which are obtained by reading the parameter file of the satellite image.

[0071] For example, the imaging channels selected through the channel analysis of this application are: four channels including the blue channel, green channel, red channel, and near-infrared channel corresponding to the multi-spectral image, and four channels corresponding to the thermal infrared image. Then, for each pixel point at the same position in the multi-spectral image and the thermal infrared image, four irradiance values can be obtained on the multi-spectral image, and four irradiance values can be obtained on the thermal infrared image. An irradiance information group including eight irradiance values is formed in the order from visible light to near-infrared and then to thermal infrared. Thus, each pixel point corresponds to an irradiance information group, and this irradiance information group includes the irradiance values corresponding to eight imaging channels.

[0072] Step S130: Based on the radiance information groups corresponding to each pixel, inversely calculate the surface temperature image of the area to be measured.

[0073] In specific implementation, after obtaining the radiance information groups corresponding to each pixel, using the radiance information groups as input variables to inversely calculate the surface temperature. Among them, inversely calculating according to the radiance information group of each pixel can obtain the surface temperature of this pixel. Matching the surface temperatures corresponding to each pixel according to the spatial positions of each pixel can form a surface temperature image, which is used as the surface temperature product of the area to be measured.

[0074] In the above surface temperature inversion method, obtain the multi-spectral image and the thermal infrared image of the area to be measured, and make the spatial resolutions of the multi-spectral image and the thermal infrared image consistent, so that the pixels on the two images correspond one by one. Further, for each pixel at the same position in the multi-spectral image and the thermal infrared image, according to the calibration coefficients of the imaging channels corresponding to the multi-spectral image and the thermal infrared image respectively and the pixel values of the pixel on the multi-spectral image and the thermal infrared image, obtain the radiance information group corresponding to the pixel. Finally, based on the radiance information groups corresponding to each pixel, inversely calculate the surface temperature image of the area to be measured. This method models the relationship between the radiance from visible light to near-infrared and thermal infrared bands and the surface temperature, avoiding the dependence of surface temperature inversion on prior conditions such as surface emissivity, atmospheric water vapor content (WVC), and atmospheric parameters, getting rid of the bondage of prior information on the surface temperature inversion algorithm, and can inversely calculate the surface temperature without any prior conditions. At the same time, since directly using the radiance information as the input for inversion, the surface temperature can be inversely calculated from the remote sensing image obtained only from a single source of hyperspectral observation satellites, without the need to convert it into indirect parameters such as apparent reflectance, normalized difference vegetation index, and vegetation coverage, avoiding the error transmission in the traditional surface temperature inversion algorithm, and making full use of the advantages of multi-channel satellite remote sensing observation data, thereby improving the accuracy of the inversion result of the surface temperature product.

[0075] In an exemplary embodiment, the above step S130 based on the radiance information groups corresponding to each pixel inversely calculates the surface temperature image of the area to be measured, including: inputting the radiance information groups corresponding to each pixel into a trained surface temperature inversion model for processing to obtain the surface temperature corresponding to each pixel; obtaining the surface temperature image of the area to be measured according to the surface temperature corresponding to each pixel.

[0076] In specific implementation, a land surface temperature inversion model can be pre-constructed. Considering that the relationship between the input and output variables for land surface temperature inversion is a non-linear function, and neural networks have excellent fitting performance for non-linear functions, neural networks are thus used as the land surface temperature inversion model. Further, a simulation dataset is constructed. Using the land surface temperature in the simulation dataset as the sample label, and the radiance information of a total of eight channels from visible light to near-infrared and thermal infrared as the input variables, the simulation dataset is divided into a training set, a validation set, and a test set according to a preset ratio (such as 8:1:1) for the training of the neural network, obtaining a trained land surface temperature inversion model. The radiance information groups of each corresponding pixel point in the multi-spectral image and the thermal infrared image are formed into a matrix form (n rows and 8 columns, where n is the number of pixel points in the multi-spectral image / thermal infrared image) and input into the trained neural network for prediction, obtaining the land surface temperature prediction results for each pixel point, and then embedding according to the positional relationship of each pixel in space to obtain the land surface temperature image.

[0077] In some embodiments, considering that the order of magnitude of the radiance values of each channel and the land surface temperature values is different, the simulation dataset can first be normalized to eliminate the order of magnitude difference, and at the same time prevent gradient explosion and accelerate the training efficiency of the neural network. For example, after scaling it to the range of [0,1] using the maximum-minimum normalization model, it is then input into the neural network for prediction. Correspondingly, after obtaining the prediction result through neural network processing, the inverse transformation is performed through the normalization model to obtain the predicted land surface temperature.

[0078] The construction process of the above simulation dataset includes:

[0079] 1. Data collection.

[0080] (1.1) Obtain multi-band images of the full spectrum from shortwave to thermal infrared through the full-spectrum imager (VIMI) on the hyperspectral observation satellite, and use its visible light to near-infrared band and thermal infrared band, specifically the blue light channel VIMI-1, green light channel VIMI-2, red light channel VIMI-3, near-infrared channel VIMI-4, thermal infrared channels VIMI-9, VIMI-10, VIMI-11, and VIMI-12, as well as the shortwave infrared channels VIMI-5 and VIMI-6.

[0081] (1.2) Screen out the atmospheric profiles over land under clear sky conditions from the atmospheric profile library. Specifically, from the atmospheric profile library, eliminate the atmospheric profiles where the relative humidity of any layer exceeds 90%, the relative humidity of two consecutive layers is greater than 85%, and the relative humidity of any layer within 2 kilometers from the ground is greater than 80%, obtaining the atmospheric profiles over land under clear sky conditions.

[0082] (1.3) Select several target surface emissivity samples from the ground object spectral library according to the categories of water bodies, soils, minerals, vegetation, and artificial materials.

[0083] 2. Data processing.

[0084] Convolve the target surface emissivity samples selected in (1.3) with the spectral response functions of each channel in (1.1) respectively to obtain the effective emissivity of the target surface emissivity samples in each channel. Input the effective emissivity of each channel and the clear sky atmospheric profile selected in (1.2) into the radiative transfer model, set the aerosol type to rural, the observation zenith angle to 0° (hyperspectral observation satellites all observe the surface vertically at the lowest point of the orbit), and the surface temperature is taken as -5K, +0K, +5K, +10K, +15K, and +20K of the lower boundary temperature corresponding to the atmospheric profile library, and calculate the radiance curves corresponding to each channel; finally, convolve the radiance curves corresponding to each channel with the spectral response functions of each channel to obtain the effective radiance of each channel. Repeat the above calculation process to obtain a total of 943 (number of atmospheric profiles) * 67 (ground object samples) * 6 (surface temperature estimations) = 379086 groups of data. Each group of data consists of surface temperature, emissivities corresponding to the four thermal infrared bands, and a total of ten radiance information from visible to near-infrared, shortwave infrared, and thermal infrared, which is used as the simulation dataset. The overall flowchart is as Figure 2 shown.

[0085] After obtaining the simulation dataset, several groups of data (such as 100000 groups) can be selected first for testing the neural network structure, and the test results are obtained as Figure 3 shown. The test results show that as the depth of the neural network increases, the inversion ability first increases and then decreases. The increase in depth increases the number of neurons and the non-linear representation ability, but an overly deep network will be difficult to learn and lead to a decrease in the inversion ability; as the width of the neural network increases, the inversion ability first increases and then decreases. The increase in width increases the number of neurons, but there is an upper limit to the enhancement of the learning ability of the neural network by the width, and it is not as large as the improvement by increasing the depth. Therefore, a network structure with 7 hidden layers and 100 neurons is selected as the relatively optimal neural network structure.

[0086] Finally, the selected neural network structure includes 8 inputs, 7 hidden layers, a Dropout layer, and 1 Self Attention layer. Among them, each hidden layer has 100 neurons; the Dropout layer sets a probability of 0.5 to lose neurons to increase the generalization ability; the Self Attention layer is used to enhance the learning ability.

[0087] In this embodiment, a neural network is used as the land surface temperature inversion model. After training, the land surface temperature is inverted. By virtue of the excellent fitting performance of the neural network for non-linear relationships, the inversion accuracy can be further improved.

[0088] In one exemplary embodiment, as Figure 4 shown, the input variables of the land surface temperature inversion model (i.e., which variables are used as the basis for land surface temperature inversion) are determined in the following manner:

[0089] Step S410, determine the first relationship for land surface temperature inversion; the input variables in the first relationship include the brightness temperature of the thermal infrared channel, the atmospheric transmittance, and the land surface emissivity.

[0090] Step S420, convert the land surface emissivity of the thermal infrared channel to be characterized by the radiance from visible light to near-infrared and short-wave infrared to obtain the first function, and convert the atmospheric transmittance of the thermal infrared channel to be characterized by radiance to obtain the second function.

[0091] Step S430, adjust the first relationship according to the first function and the second function to obtain the target relationship for land surface temperature inversion, and use the input variables in the target relationship as the input variables of the land surface temperature inversion model.

[0092] In a specific implementation, the first relationship for land surface temperature inversion can be determined by the split-window algorithm. The first relationship takes the brightness temperature of the thermal infrared channel, the atmospheric transmittance, and the land surface emissivity as inputs and the land surface temperature as the output. Further, to avoid the influence of prior information such as atmospheric transmittance and land surface emissivity on subsequent inversion, the land surface emissivity of the thermal infrared channel can be converted to be characterized by the radiance from visible light to near-infrared and short-wave infrared through an improved normalized difference vegetation index algorithm to obtain the first function, and the atmospheric transmittance of the thermal infrared channel can be converted to be characterized by brightness temperature through a linear split-window algorithm, and further, the converted brightness temperature and the brightness temperature of the thermal infrared channel are converted to be characterized by radiance through the Planck function to obtain the second function. Through the first function and the second function, the brightness temperature, the atmospheric transmittance, and the land surface emissivity of the thermal infrared channel in the first relationship can all be converted to be characterized by radiance. Therefore, the obtained target relationship is an inversion relationship with multiple radiance information as inputs and the land surface temperature as the output.

[0093] Further, in one exemplary embodiment, the above step S430 adjusts the first relationship according to the first function and the second function to obtain the target relationship for land surface temperature inversion, including: adjusting the first relationship according to the first function to obtain the second relationship for land surface temperature inversion; adjusting the second relationship according to the second function to obtain the third relationship for land surface temperature inversion; and simplifying the input variables in the third relationship through principal component analysis to obtain the target relationship for land surface temperature inversion.

[0094] In specific implementation, when adjusting the first relation according to the first function and the second function, the first relation can be adjusted according to the first function first to obtain the second relation for land surface temperature inversion, and then the second relation can be adjusted according to the second function to obtain the third relation for land surface temperature inversion; finally, the input variables in the third relation are simplified by the principal component analysis method to obtain the target relation for land surface temperature inversion.

[0095] To more clearly clarify the deduction process of the input variables of the above land surface temperature inversion model, the following provides a more specific description of this solution, including the following processes:

[0096] (1) Determine the first relation for land surface temperature inversion according to the split-window algorithm.

[0097] Based on the split-window algorithm for land surface temperature inversion, the land surface temperature can be described as a single-output function determined by the independent variables , , , , , , that is:

[0098] (2)

[0099] There are 4 channels in the thermal infrared band of the hyperspectral observation satellite, and they are distributed in pairs at 8-9um and 10-12um. Therefore, more accurate theoretical results can be obtained by using the split-window algorithm twice and taking the average, that is, the following first relation for land surface temperature inversion can be obtained:

[0100] (3)

[0101] Among them, represents the brightness temperature of the four thermal infrared channels, represents the atmospheric transmittance of the four thermal infrared channels, represents the land surface emissivity of the four thermal infrared channels.

[0102] (2) Adjust the first relation based on the improved normalized difference vegetation index algorithm to obtain the second relation for land surface temperature inversion.

[0103] The first relation for land surface temperature inversion shows that the land surface emissivity of the thermal infrared channel is required to solve the land surface temperature. The process of solving the land surface emissivity by the improved normalized difference vegetation index algorithm is described as follows:

[0104] This algorithm assumes that pixels are composed of a mixture of two surface types, soil and vegetation. The surface emissivity of soil pixels consists of the apparent reflectance from visible to near-infrared and shortwave infrared. For hyperspectral observation satellites, it is the apparent reflectance of VIMI-1, VIMI-2, VIMI-3, VIMI-4, VIMI-5, and VIMI-6. The surface emissivity of vegetation pixels is defined as a linear function of the normalized difference vegetation index (NDVI), and NDVI is obtained by taking the ratio of the apparent reflectance of VIMI-3 and VIMI-4. The apparent reflectance is defined as the ratio of channel radiance to extraterrestrial irradiance. For each channel of hyperspectral observation satellites, the extraterrestrial irradiance is a constant value. Therefore, the surface emissivity in the thermal infrared band can be described as a single-output function (i.e., the first function) composed of visible to near-infrared and shortwave infrared radiance:

[0105] (4)

[0106] Therefore, the first relationship for retrieving surface temperature can be simplified to the second relationship:

[0107] (5)

[0108] (3) Adjust the second relationship according to the linear split-window algorithm to obtain the third relationship for retrieving surface temperature.

[0109] The atmospheric transmittance in the thermal infrared channel is defined as a quadratic function of the atmospheric water vapor function. At the same time, the linear split-window algorithm defines the atmospheric water vapor content as a linear function of the brightness temperature difference in the thermal infrared channel. Therefore, the atmospheric transmittance in the thermal infrared channel can be described as a function of the brightness temperature:

[0110] (6)

[0111] At the same time, the brightness temperature and radiance in the thermal infrared channel can be converted correspondingly through the effective Planck function of the channel. Therefore, the third relationship for retrieving surface temperature can finally be obtained:

[0112] (7)

[0113] (4) Simplify using the principal component analysis method to obtain the fourth relationship for retrieving surface temperature.

[0114] The third relationship for retrieving surface temperature indicates that the surface temperature It is a multi-input single-output function defined by the radiances of channels 1-6, 9-12 of the hyperspectral observation satellite. Although the above relationship is derived from various physics-based algorithms, for the training of neural networks, 10 inputs may be redundant, reducing the training efficiency and results. Therefore, the principal component analysis method is used to analyze the generated simulation dataset. The results show that the contributions of the first principal component and the second principal component to the overall characteristics are 97.28%. Therefore, only two principal component components can characterize the characteristics of the overall dataset. As Figure 5 and Figure 6 shown, by analyzing the contributions of the radiances of 10 channels of the hyperspectral observation satellite to the first principal component, the second principal component, and the third principal component, it is known that the visible to near-infrared channels VIMI-1, VIMI-2, VIMI-3, and VIMI-4 make the main contributions to the first principal component, the thermal infrared channels VIMI-9, VIMI-10, VIMI-11, and VIMI-12 make the main contributions to the second principal component, and the shortwave infrared channels VIMI-5 and VIMI-6 make the main contributions to the third principal component. Since the first and second principal components can already characterize the overall characteristics, the radiance information of the shortwave infrared channels VIMI-5 and VIMI-6 is considered redundant information that needs to be removed. This is also consistent with the physical process. The visible light band is used for surface observation, the near-infrared band is used for vegetation monitoring, the thermal infrared band is used for surface / atmospheric temperature observation, and the shortwave infrared is used for water body and cloud research, which is less useful for surface research.

[0115] Therefore, the fourth relationship for retrieving the surface temperature is finally obtained:

[0116] (8)

[0117] Finally, the surface temperature is characterized as a multi-input single-output function of the radiances of the visible to near-infrared and thermal infrared channels.

[0118] Among them, 、 、 、 are the radiances of channels VIMI-9, VIMI-10, VIMI-11, and VIMI-12; represents the non-linear function relationship. Equation (8) shows that theoretically, the surface temperature can be retrieved only through the radiances of the visible to near-infrared, shortwave infrared, and thermal infrared channels without the input of any prior information. Therefore, the only thing to be determined is the non-linear function relationship . The Universal Approximation Theorem states that a multi-layer feedforward network with at least one hidden layer can approximate any continuous function with arbitrary precision provided that there are enough neurons. However, in engineering applications, shallow and wide networks cannot learn and generalize correctly. Therefore, deep neural networks (DNNs) have become the best choice for learning non-linear functional relationships .

[0119] A deep neural network is a multi-layer feedforward neural network trained based on the error backpropagation algorithm and the stochastic gradient descent algorithm. The network is deepened through fully connected layers, and the non-linearity of the network is enhanced through activation functions. Due to its superior non-linear fitting ability, DNNs are widely used in cross-disciplinary research fields. They can autonomously learn useful features from large amounts of data and handle complex non-linear relationships, thus being suitable for solving the ill-posed problem of land surface temperature inversion. In a DNN, data starts from the input layer, goes through layer-by-layer calculations in the hidden layers, and finally reaches the output layer. The output of each layer of neurons serves as the input to the next layer of neurons, and non-linear transformations are achieved through activation functions:

[0120] (9)

[0121] where , , and represent the output, activation function, weight, and bias of the -th layer of the neural network respectively; represents the input to the -th layer of the neural network. The backpropagation algorithm calculates the error between the output layer and the true label and backpropagates the error to each layer of neurons to update the weights and bias terms of the neurons to minimize the prediction error:

[0122] [[ID=~]]](10)

[0123] where and represent the error and output of the -th layer of the neural network respectively; and represent the error and weight of the -th layer of the neural network respectively; the error of the output layer is calculated by the loss function. Finally, it can be seen from Equation (8) that only a DNN with 8 inputs and 1 output needs to be constructed to learn the non-linear functional relationship and achieve the goal of inverting the land surface temperature.

[0124] Reference Figure 7, which is a schematic diagram of the structure of a neural network shown in an embodiment. A deep neural network (DNN) is a multi-layer feedforward neural network trained based on the error backpropagation algorithm and the stochastic gradient descent algorithm. The network is deepened through fully connected layers, and the non-linear expression of the network is enhanced through an activation function (ReLU). Due to its superior non-linear fitting ability, DNN is widely used in cross-research fields, can autonomously learn useful features from a large amount of data and process complex non-linear relationships, and thus is suitable for solving the ill-posed problem of land surface temperature inversion. In DNN, data starts from the input layer, goes through layer-by-layer calculations in the hidden layer, and finally reaches the output layer. The output of each layer of neurons serves as the input of the next layer of neurons, and non-linear transformation is achieved through the activation function. The backpropagation algorithm calculates the error between the output layer and the true label, and backpropagates the error to each layer of neurons to update the weights and bias terms of the neurons to minimize the prediction error.

[0125] In this embodiment, first, the original inputs for retrieving the land surface temperature are refined from the split-window algorithm: the brightness temperatures, atmospheric transmittances, and land surface emissivities of VIMI-9, VIMI-10, VIMI-11, and VIMI-12. Then, through the improved normalized difference vegetation index algorithm, the land surface emissivities of the original inputs VIMI-9, VIMI-10, VIMI-11, and VIMI-12 in the first relational expression are further refined into the radiance information of VIMI-1, VIMI-2, VIMI-3, VIMI-4, VIMI-5, and VIMI-6. Further, through the linear split-window inversion algorithm for atmospheric water vapor content, the atmospheric transmittances of VIMI-9, VIMI-10, VIMI-11, and VIMI-12 are further refined into the brightness temperature information of VIMI-9, VIMI-10, VIMI-11, and VIMI-12. According to the Planck function, the brightness temperature and radiance of each channel correspond one by one. Therefore, the brightness temperatures of VIMI-9, VIMI-10, VIMI-11, and VIMI-12 can be condensed into the radiances of VIMI-9, VIMI-10, VIMI-11, and VIMI-12 (because the radiance information of the images collected by the hyperspectral observation satellite is the most original image information). Thus, the land surface temperature can be solved from the radiance information of VIMI-1, VIMI-2, VIMI-3, VIMI-4, VIMI-5, VIMI-6, VIMI-9, VIMI-10, VIMI-11, and VIMI-12. Finally, by performing principal component analysis on the simulated data set to simplify the redundancy of the neural network input, the analysis shows that the radiance information of VIMI-5 and VIMI-6, that is, the short-wave infrared channels, belongs to the redundancy and needs to be deleted. Therefore, the final input to the neural network is the radiance information of eight channels, namely VIMI-1, VIMI-2, VIMI-3, VIMI-4, VIMI-9, VIMI-10, VIMI-11, and VIMI-12. This method derives a neural network inversion algorithm without prior information from physical processes and simplifies the redundancy using a machine learning algorithm (principal component analysis), and finally obtains an optimal neural network model, which can improve the inversion accuracy and efficiency.

[0126] In an exemplary embodiment, as Figure 8 shown, the above step S110 of obtaining the multi-spectral image and the thermal infrared image of the area to be measured includes:

[0127] Step S810 of obtaining the initial multi-spectral image and the initial thermal infrared image of the area to be measured;

[0128] Step S820 of identifying the overlapping area between the initial multi-spectral image and the initial thermal infrared image;

[0129] Step S830: Based on the overlapping region, match the initial multispectral image and the initial thermal infrared image to obtain the multispectral image and the thermal infrared image of the region to be measured.

[0130] Specifically, the initial multispectral image and the initial thermal infrared image can be acquired by the full-spectrum imager on the hyperspectral observation satellite. At the same time, considering that there are deviations in the corresponding positions of the acquired multispectral image and the thermal infrared image in the region to be measured, therefore, when matching the two images, first identify the overlapping region between the initial multispectral image and the initial thermal infrared image, and match the initial multispectral image and the initial thermal infrared image in the overlapping region to ensure that the regions corresponding to the matched multispectral image and the thermal infrared image are consistent.

[0131] At the same time, since the spatial resolution of the multispectral image is usually high, while the spatial resolution of the thermal infrared image is usually low, therefore, matching the initial multispectral image and the initial thermal infrared image can specifically be to perform a resolution reduction operation on the initial multispectral image to adjust the resolution of the initial multispectral image to be the same as that of the initial thermal infrared image. Since the actual resolution of the multispectral image is twice that of the thermal infrared image, it is necessary to take the average value of 2×2 starting from the center point.

[0132] Specifically, the image region corresponding to the overlapping region in the initial multispectral image can be extracted as the first image, and the image region corresponding to the overlapping region in the initial thermal infrared image can be extracted as the second image; downsample the first image according to the resolution of the second image to obtain the first processed image; use the first processed image as the multispectral image of the region to be measured, and use the second image as the thermal infrared image of the region to be measured.

[0133] In this embodiment, when matching the initial multispectral image and the initial thermal infrared image of the region to be measured, considering that there are deviations in the corresponding positions of the acquired multispectral image and the thermal infrared image in the region to be measured, therefore, first identify the overlapping region between the initial multispectral image and the initial thermal infrared image, and match the initial multispectral image and the initial thermal infrared image in the overlapping region to ensure that the regions corresponding to the matched multispectral image and the thermal infrared image are consistent.

[0134] In an exemplary embodiment, the above step S820 of identifying the overlapping region between the initial multispectral image and the initial thermal infrared image includes: respectively converting the pixel coordinates of the initial multispectral image and the initial thermal infrared image into longitude and latitude coordinates; determining the positioning information of the matching region of the initial multispectral image and the initial thermal infrared image under the longitude and latitude coordinates; the positioning information includes the longitude and latitude coordinates of the first and last pixels of the matching region; convert the longitude and latitude coordinates of the first and last pixels back to the pixel coordinates of the initial multispectral image and the initial thermal infrared image to obtain the overlapping region.

[0135] In specific implementation, first, the initial multispectral image and the initial thermal infrared image are respectively converted from pixel x-y coordinates to longitude and latitude coordinates. According to the characteristics of hyperspectral observation satellite images, that is, the horizontal direction represents longitude and the vertical direction represents latitude. Therefore, when determining the overlapping area between the initial multispectral image and the initial thermal infrared image, the starting position can be determined by the first pixel of the image with a larger latitude.

[0136] The observation ranges of the multispectral image and the thermal infrared image collected by the hyperspectral observation satellite are as Figure 9 shown. The solid frame in the figure represents the observation range corresponding to the multispectral image, and the dashed frame represents the observation range corresponding to the thermal infrared image. It can be seen from the figure that the actual coverage of the thermal infrared image will be smaller than that of the multispectral image, that is, the upper boundary of the dashed frame may exceed the upper boundary of the solid frame, or it may not exceed, but the other three boundaries must conform to the illustrated rules. Therefore, the overlapping range needs to be determined by the first pixel of the image with a higher latitude.

[0137] When the latitude of the first pixel of the multispectral image is lower (Figure a), the longitude and latitude of the first pixel of the matching area are the longitude of the first pixel of the thermal infrared image plus the latitude of the first pixel of the multispectral image; when the latitude of the first pixel of the multispectral image is higher (Figure b), the longitude and latitude of the first pixel of the matching area are the longitude of the first pixel of the thermal infrared image plus the latitude of the first pixel of the thermal infrared image. The longitude and latitude of the last pixel of the matching area (i.e., the ending position) must be the latitude of the last pixel of the thermal infrared image plus the longitude of the last pixel of the multispectral image.

[0138] By converting the longitude and latitude of the first and last pixels after matching into the xy coordinates of the multispectral image and the thermal infrared image respectively, the overlapping area can be extracted, and further the overlapping area of the multispectral image is downsampled to the resolution of the thermal infrared overlapping area to obtain two images with the same resolution.

[0139] In this embodiment, after converting the pixel coordinates of the initial multispectral image and the initial thermal infrared image into longitude and latitude coordinates respectively, by determining the longitude and latitude coordinates of the first and last pixels of the matching area between the initial multispectral image and the initial thermal infrared image in longitude and latitude coordinates, and then converting the longitude and latitude coordinates of the first and last pixels back into the pixel coordinates of the initial multispectral image and the initial thermal infrared image, the overlapping area is obtained, realizing the determination of the overlapping area at the pixel level, which can ensure the accuracy of the extracted overlapping area.

[0140] To verify the effect of the surface temperature inversion algorithm based on neural network proposed in this application, the method of this application is verified on the ground on different ecological networks, and compared with the split window algorithm and the temperature / emissivity separation algorithm, and the comparison results shown in Table 1 are obtained.

[0141] Table 1

[0142]

[0143] The verification of Table 1 is to select satellite images of different ecological network base stations, then perform spatial matching between the visible to near-infrared band and the thermal infrared band, calculate the matching pixels in the multi-spectral image and the thermal infrared image from the longitude and latitude of the base station, select the average value of the 3*3 pixel range of the matching center to calculate the radiance information and send it into the neural network for prediction, and compare the prediction result with the surface temperature measured by the base station to obtain the result.

[0144] By synthesizing the data of each site, it can be seen that the surface temperature inversion algorithm based on neural network proposed in this application achieves the optimal inversion effect. The root mean square error (RMSE) is significantly lower than that of the split-window algorithm and the temperature / emissivity separation algorithm, showing a better fitting effect. The deviation rate (BIAS) is closest to 0, indicating the accuracy of the inversion. Among them, the maximum root mean square error (RMSE) appears at the wetland site, which can be considered due to the higher water vapor content and stronger surface heterogeneity at the wetland site, making it more difficult to obtain atmospheric water vapor content, surface emissivity, and atmospheric parameters, resulting in larger errors. The surface temperature inversion algorithm based on neural network proposed in this application has the best effect, which can be determined because the algorithm of this application avoids prior conditions and reduces the error transmission caused by it.

[0145] The surface temperature inversion algorithm based on neural network proposed in this application highly abstracts the physics-based algorithm to obtain the fundamental representation affecting the inversion result, getting rid of the dependence on intermediate variables such as surface emissivity, atmospheric water vapor content, and atmospheric parameters. Without any prior conditions, the surface temperature can be inverted. This advantage mainly stems from the fact that the inversion algorithms of intermediate variables also start from the radiance received by the satellite for estimation. Tracing back to the source, the starting points of development are the same, so they can be abstracted and combined. The surface temperature inversion algorithm based on neural network of this application eliminates the error transmission of intermediate variables by avoiding prior information. At the same time, since the input is only composed of radiance information, the number of sensitive sources is reduced, and higher inversion accuracy and less sensitivity to the error of input radiance information can be achieved.

[0146] It should be understood that although the steps in the flowcharts involved in the above-described embodiments are sequentially shown according to the indications of the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless specifically stated herein, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-described embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be executed alternately or in turn with at least a part of other steps or steps or stages in other steps.

[0147] Based on the same inventive concept, an embodiment of the present application further provides a surface temperature inversion device for implementing the above-mentioned surface temperature inversion method. The solution provided by this device for solving problems is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the surface temperature inversion device provided below can refer to the limitations on the surface temperature inversion method in the foregoing, and will not be elaborated herein.

[0148] In one embodiment, as Figure 10 shown, a surface temperature inversion device is provided, including:

[0149] An image acquisition module 1001, configured to acquire a multi-spectral image and a thermal infrared image of a to-be-detected area; the spatial resolutions of the multi-spectral image and the thermal infrared image are consistent;

[0150] A data processing module 1002, configured to obtain a group of radiance information corresponding to a pixel point according to the calibration coefficients of the imaging channels corresponding to the multi-spectral image and the thermal infrared image and the pixel values of the pixel point on the multi-spectral image and the thermal infrared image for each pixel point at the same position in the multi-spectral image and the thermal infrared image;

[0151] A temperature inversion module 1003, configured to inversely obtain a surface temperature image of the to-be-detected area based on the group of radiance information corresponding to each pixel point.

[0152] In one of the embodiments, the temperature inversion module 1003 is further configured to input the group of radiance information corresponding to each pixel point into a trained surface temperature inversion model for processing to obtain the surface temperature corresponding to each pixel point; and obtain a surface temperature image of the to-be-detected area according to the surface temperature corresponding to each pixel point.

[0153] In one embodiment, the device further includes a derivation module configured to determine a first relational expression for surface temperature inversion; the input variables in the first relational expression include the brightness temperature of the thermal infrared channel, the atmospheric transmittance, and the surface emissivity; the surface emissivity of the thermal infrared channel is converted to be characterized by the radiance from visible light to near-infrared and shortwave infrared, obtaining a first function, and the atmospheric transmittance and brightness temperature of the thermal infrared channel are converted to be characterized by the radiance, obtaining a second function; according to the first function and the second function, the first relational expression is adjusted to obtain a target relational expression for surface temperature inversion, and the input variables in the target relational expression are used as the input variables of the surface temperature inversion model.

[0154] In one embodiment, the derivation module is further configured to adjust the first relational expression according to the first function to obtain a second relational expression for surface temperature inversion; adjust the second relational expression according to the second function to obtain a third relational expression for surface temperature inversion; perform a simplification process on the input variables in the third relational expression through principal component analysis to obtain a target relational expression for surface temperature inversion.

[0155] In one embodiment, the image acquisition module 1001 is further configured to acquire an initial multispectral image and an initial thermal infrared image of the area to be measured; identify the overlapping area between the initial multispectral image and the initial thermal infrared image; based on the overlapping area, match the initial multispectral image and the initial thermal infrared image to obtain a multispectral image and a thermal infrared image of the area to be measured.

[0156] In one embodiment, the image acquisition module 1001 is further configured to respectively convert the pixel coordinates of the initial multispectral image and the initial thermal infrared image into longitude and latitude coordinates; determine the positioning information of the matching area of the initial multispectral image and the initial thermal infrared image in the longitude and latitude coordinates; the positioning information includes the longitude and latitude coordinates of the head and tail pixels of the matching area; convert the longitude and latitude coordinates of the head and tail pixels back to the pixel coordinates of the initial multispectral image and the initial thermal infrared image to obtain the overlapping area.

[0157] Each module in the above surface temperature inversion device can be implemented in whole or in part by software, hardware, and their combination. The above modules can be embedded in the processor of the computer device in hardware form or be independent of it, or can be stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to the above respective modules.

[0158] In one embodiment, a computer device is provided. The computer device can be a server, and its internal structure diagram can be as Figure 11As shown in the figure. The computer device includes a processor, a memory, and a network interface connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data during the surface temperature inversion process. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, it implements a surface temperature inversion method.

[0159] Those skilled in the art can understand that Figure 11 the structure shown in the figure is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0160] In one embodiment, a computer device is further provided, including a memory and a processor. A computer program is stored in the memory. When the processor executes the computer program, the steps in the above method embodiments are implemented.

[0161] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by the processor, the steps in the above method embodiments are implemented.

[0162] In one embodiment, a computer program product is provided, including a computer program. When the computer program is executed by the processor, the steps in the above method embodiments are implemented.

[0163] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or fully authorized by all parties. The collection, use, and processing of relevant data need to comply with relevant laws, regulations, and standards of relevant countries and regions.

[0164] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memories. Non-volatile memories can include read-only memory (ROM), magnetic tapes, floppy disks, flash memories, optical memories, high-density embedded non-volatile memories, resistive random access memories (ReRAM), magnetoresistive random access memories (MRAM), ferroelectric random access memories (FRAM), phase change memories (PCM), graphene memories, etc. Volatile memories can include random access memory (RAM) or external cache memories, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, data processing logics based on quantum computing, etc., without limitation.

[0165] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0166] The above-described embodiments merely represent several implementation manners of the present application. The description thereof is relatively specific and detailed, but it should not be construed as a limitation on the patent scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the appended claims.

Claims

1. A method for retrieving land surface temperature, characterized in that, The method includes: Obtaining a multi-spectral image and a thermal infrared image of the area to be measured; the spatial resolutions of the multi-spectral image and the thermal infrared image are the same; For each pixel point at the same position in the multi-spectral image and the thermal infrared image, according to the calibration coefficients of the imaging channels corresponding to the multi-spectral image and the thermal infrared image respectively, and the pixel values of the pixel point on the multi-spectral image and the thermal infrared image, obtaining a group of radiance information corresponding to the pixel point; Inputting the groups of radiance information corresponding to each pixel point into a trained land surface temperature inversion model for processing to obtain the land surface temperature corresponding to each pixel point; the trained land surface temperature inversion model is obtained by constructing a simulated data set, using the groups of radiance information in the simulated data set as input variables and the ground temperature as sample labels to train a neural network; According to the land surface temperature corresponding to each pixel point, obtaining a land surface temperature image of the area to be measured.

2. The method according to claim 1, wherein The method further includes: Performing normalization processing on the simulated data set; Training a neural network with the normalized simulated data set to obtain the trained land surface temperature inversion model.

3. The method according to claim 1, wherein The input variables of the land surface temperature inversion model are determined by the following method: Determining a first relational expression for land surface temperature inversion; the input variables in the first relational expression include the brightness temperature of the thermal infrared channel, the atmospheric transmittance, and the land surface emissivity; Converting the land surface emissivity of the thermal infrared channel into radiance representation from visible light to near-infrared and short-wave infrared to obtain a first function, and converting the atmospheric transmittance and brightness temperature of the thermal infrared channel into radiance representation to obtain a second function; According to the first function and the second function, adjusting the first relational expression to obtain a target relational expression for land surface temperature inversion, and using the input variables in the target relational expression as the input variables of the land surface temperature inversion model.

4. The method according to claim 3, characterized in that, The adjusting the first relational expression according to the first function and the second function to obtain a target relational expression for land surface temperature inversion includes: Adjusting the first relational expression according to the first function to obtain a second relational expression for land surface temperature inversion; Adjusting the second relational expression according to the second function to obtain a third relational expression for land surface temperature inversion; Performing simplification processing on the input variables in the third relational expression by principal component analysis to obtain a target relational expression for land surface temperature inversion.

5. The method according to claim 1, wherein The obtaining the multi-spectral image and the thermal infrared image of the area to be measured includes: Obtaining an initial multi-spectral image and an initial thermal infrared image of the area to be measured; Identifying the overlapping area between the initial multi-spectral image and the initial thermal infrared image; Based on the overlapping area, matching the initial multi-spectral image and the initial thermal infrared image to obtain the multi-spectral image and the thermal infrared image of the area to be measured.

6. The method according to claim 5, wherein The identifying the overlapping area between the initial multi-spectral image and the initial thermal infrared image includes: Converting the pixel coordinates of the initial multi-spectral image and the initial thermal infrared image into longitude and latitude coordinates respectively; Determine the positioning information of the matching region of the initial multispectral image and the initial thermal infrared image under the latitude and longitude coordinates; the positioning information includes the latitude and longitude coordinates of the first and last pixels of the matching region. Convert the latitude and longitude coordinates of the first and last pixels back to the pixel coordinates of the initial multispectral image and the initial thermal infrared image to obtain the overlapping region.

7. A surface temperature inversion device, characterized in that, The device includes: An image acquisition module for acquiring a multispectral image and a thermal infrared image of a region to be measured; the spatial resolutions of the multispectral image and the thermal infrared image are consistent. A data processing module for, for each pixel point at the same position in the multispectral image and the thermal infrared image, obtaining a set of radiance information corresponding to the pixel point according to the calibration coefficients of the imaging channels corresponding to the multispectral image and the thermal infrared image respectively, and the pixel values of the pixel point on the multispectral image and the thermal infrared image. A temperature inversion module for inputting the set of radiance information corresponding to each pixel point into a trained land surface temperature inversion model for processing to obtain the land surface temperature corresponding to each pixel point; obtaining a land surface temperature image of the region to be measured according to the land surface temperatures corresponding to each pixel point; the trained land surface temperature inversion model is obtained by constructing a simulation dataset, using the set of radiance information in the simulation dataset as input variables and the ground temperature as sample labels to train a neural network.

8. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, the steps of the land surface temperature inversion method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, the steps of the land surface temperature inversion method according to any one of claims 1 to 6 are implemented.

10. A computer program product comprising a computer program, characterized in that, When the computer program is executed by the processor, the steps of the land surface temperature inversion method according to any one of claims 1 to 6 are implemented.

Images