Coupling method for improving land surface temperature inversion resolution and precision of desert area

Abstract

The invention discloses a coupling method for improving land surface temperature inversion resolution and precision in a desert area, which comprises the following steps of: (1) obtaining fine classification of land surface types based on a land surface fine classification model obtained by training, and then calculating to obtain first high-resolution land surface emissivity data; and (2) calculating to obtain the adjusted Landsat emissivity, and further calculating to obtain second high-resolution surface emissivity data. And (3) endowing the first high-resolution surface emissivity and the second high-resolution surface emissivity with different weights by using linear weighted fusion to obtain high-resolution surface emissivity data. And (4) calculating the atmospheric water vapor content. And (5) generating a high-resolution thermal infrared image based on an adversarial neural network algorithm. And (6) calculating high-resolution radiation brightness and brightness temperature according to the DN value of the high-resolution thermal infrared image, and performing surface temperature inversion according to all high-resolution parameter data in combination with a single-channel algorithm to finally obtain high-precision and high-resolution surface temperature data.

Description

Technical Field

[0001] The invention belongs to the field of surface temperature inversion, and in particular relates to a coupling method for improving the resolution and accuracy of surface temperature inversion in desert areas. Background Art

[0002] In the global perspective, ecological and environmental issues have become increasingly prominent and have become a core topic of common concern to all countries. The land surface temperature is an extremely important physical quantity. It can reflect the thermal balance between the atmosphere and the land surface and is one of the important indicators for measuring the natural ecological environment on the earth's surface. At the same time, LST (Land surface temperature) also carries a wealth of environmental information, such as the dynamic changes in land cover and the evaporation of water bodies. This information is crucial for in-depth research on the evolution of the ecological environment and the formulation of protection strategies. Its acquisition plays a vital role in global climate change detection, disaster prevention, weather forecasting, urban construction and other aspects.

[0003] The traditional LST acquisition method mainly relies on temperature measuring equipment at ground measurement sites. Although this method can provide some high-precision data and obtain better temperature inversion results, its high cost and data limitations (such as uneven distribution) limit its application.

[0004] At present, with the rapid development of satellite remote sensing technology, using remote sensing technology to obtain LST has become a new way. Compared with traditional measurement methods, remote sensing technology has many advantages. It can cover a wider measurement area, provide more comprehensive LST information, and save high costs. But at the same time, the commonly used single-channel algorithm, multi-channel algorithm or split window algorithm itself has many disadvantages. First, the inaccuracy of parameters, such as surface emissivity and atmospheric water vapor content, are difficult to obtain and are easily affected by factors such as changes in surface cover types and atmospheric conditions. Therefore, the acquired data is uncertain and the accuracy is not high; secondly, the real physical model is very complex, but in order to simplify the calculation and improve efficiency, the inversion algorithm will simplify the model, and this processing may also lead to errors; finally, the existing algorithms are used in very limited scenarios. The single-channel algorithm is more suitable for simple atmospheric conditions and surface types, while the multi-channel algorithm or split window algorithm is more suitable for complex scenes. Therefore, it is necessary to set up an inversion method that can be targeted at specific scenes to effectively improve the inversion accuracy. Summary of the invention

[0005] In view of the problems existing in the prior art, the present invention provides a coupling method for improving the resolution and accuracy of surface temperature inversion in desert areas, which can effectively improve the accuracy and resolution of surface temperature inversion results in desert areas.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: a coupling method for improving the resolution and accuracy of surface temperature inversion in desert areas, comprising the following steps:

[0007] S1. Construct and train several remote sensing images of desert areas and corresponding data with fine surface classification labels to obtain a surface fine classification model U-BiFormer; use the surface fine classification model U-BiFormer to finely classify the surface types of remote sensing images of desert areas, and then calculate and obtain the first high-resolution surface emissivity data based on the fine classification of surface types;

[0008] S2. Classify the pixels in the image according to the normalized vegetation index to obtain vegetation coverage; calculate the bare soil emissivity of the preset band of the ASTER data in combination with the reflectivity of the ASTER data, obtain the thermal infrared band emissivity of the Landsat series satellites through conversion, and then obtain the adjusted Landsat emissivity through vegetation adjustment and interpolation calculation, and finally calculate the second high-resolution surface emissivity data;

[0009] S3, using a linear weighted fusion method to assign different weights to the first high-resolution surface emissivity data and the second high-resolution surface emissivity data, and obtaining the third high-resolution surface emissivity data after weighted calculation;

[0010] S4, using low-resolution NCEP atmospheric water vapor content data, linearly interpolating the data in space and time to obtain high-resolution atmospheric water vapor content data, and then calculating the atmospheric function;

[0011] S5. Based on the adversarial neural network algorithm, the SRGAN model is trained on the preprocessed thermal infrared dataset. Through adversarial training of the generator and the discriminator, the generator is able to generate super-resolution thermal infrared images.

[0012] S6. Calculate the high-resolution radiation brightness and brightness temperature based on the DN value of the super-resolution thermal infrared image. Finally, combine the third high-resolution surface emissivity data and atmospheric function, use a single-channel algorithm to invert the surface temperature, and finally obtain high-precision and high-resolution surface temperature data.

[0013] Furthermore, the aforementioned step S1 includes the following sub-steps:

[0014] S1.1. Obtain high-resolution visible light data and select clear images from remote sensing images of desert areas; crop the images and add fine classification labels of surface types; then annotate the images and divide them into training sets, validation sets, and test sets;

[0015] S1.2. Add a U-type decoder based on the BiFormer model to build a surface fine classification network based on deep learning, then set the initial parameters, and use the CE+Focal mixed loss function when training the deep learning surface fine classification network, as follows:

[0016] Loss=Loss CE +Loss Focal

[0017] Among them, the cross entropy loss function Loss CE And the focal loss function Loss Focal The calculation formulas for a single pixel in multi-category tasks are:

[0018]

[0019]

[0020] Among them, y i Represents the true label of the predicted sample, p i Indicates the probability that the model predicts that the pixel belongs to the i-th category, FL(y i ,p i ) represents the focal loss of the i-th category, α is a balancing factor used to adjust the weights between positive and negative categories, and γ is a tuning parameter used to control the weights of difficult and easy samples;

[0021] S1.3. Set the training algorithm parameters, adjust the weight connection parameters according to the loss function learning results, until the preset accuracy is reached, and obtain the surface fine classification model U-BiFormer. Use the surface fine classification model U-BiFormer to finely classify the surface types of remote sensing images of desert areas, and obtain the corresponding surface types, including towns, farmlands, deserts, water bodies, and mountains;

[0022] S1.4. Take different surface emissivity for different surface types after classification to obtain the first high-resolution surface emissivity data:

[0023]

[0024] Furthermore, the aforementioned step S2 includes the following sub-steps:

[0025] S2.1. Obtain the low-resolution reflectance data of bands 13 and 14 in the ASTER dataset;

[0026] S2.2. Calculate the normalized vegetation index NDVI using near infrared NIR data and infrared R data, as follows:

[0027]

[0028] S2.3. Based on the NDVI threshold of soil and vegetation, the vegetation index is used to classify the pixels in the image into pure water bodies, mixed pixels and pure vegetation, and the vegetation coverage PV is calculated:

[0029]

[0030] S2.4. Determine the surface emissivity. When the NDVI is less than or equal to the soil threshold, the pixel is considered to belong to the water body. When the NDVI is greater than or equal to the vegetation threshold, the pixel is considered to belong to the vegetation. For pixels that are divided into mixed pixels, the PV value that has been calculated is used for further calculation to obtain the bare soil emissivity of the ASTER 13th and 14th bands, as shown in the following formula:

[0031]

[0032] Among them, ε ASTER,bare represents the ASTER bare soil emissivity, ε ASTER Indicates the emissivity of the preset band, ε ASTER,veg represents the emissivity of pure vegetation of ASTER;

[0033] S2.5. Convert the bare soil emissivity of ASTER bands 13 and 14 to the thermal infrared band emissivity of Landsat series satellites as follows:

[0034] ε L10,bare =c0ε A13,bare +c1ε A14,bare +c2

[0035] Among them, ε L10,bare represents the bare soil emissivity in band 10 of the Landsat TIRS sensor, ε A13,bare and ε A14,bare They represent the bare soil emissivity of ASTER bands 13 and 14 obtained using calculations;

[0036] S2.6. Calculate the vegetation-adjusted Landsat emissivity based on the bare soil emissivity, pure vegetation emissivity, and PV as follows:

[0037] ε=PV·ε L10,veg +(1-PV)·ε L10,bare

[0038] Where ε represents the surface emissivity of the 10th band as the input for the subsequent single-channel algorithm, and ε L10,veg represents the pure vegetation emissivity of the 10th band of Landsat, ε L10,bare represents the bare soil emissivity in the 10th band of Landsat;

[0039] S2.7, in the single-channel algorithm, is calculated using the following formula:

[0040]

[0041] S2.8. Interpolate the acquired low-resolution surface emissivity data to the same resolution as the visible light data to obtain the second highest resolution surface emissivity ε2.

[0042] Furthermore, the aforementioned step S3 includes the following sub-steps:

[0043] S3.1. Use the linear weighted fusion method to assign different weights to the first high-resolution surface emissivity data and the second high-resolution surface emissivity data.

[0044] S3.2. According to the set weights, the high-resolution surface emissivity ε finally used in the single-channel algorithm is calculated according to the following calculation formula:

[0045]

[0046] where ε i refers to the surface emissivity obtained in step i, w i Refers to the weight corresponding to the i-th step.

[0047] Furthermore, the aforementioned step S4 includes the following sub-steps:

[0048] S4.1. Based on the NCEP atmospheric water vapor content data, the data is temporally interpolated to obtain the water vapor content at the time of data capture. The formula is as follows:

[0049]

[0050] Where w represents the water vapor content at the time of Landsat acquisition, and represents the NCEP water vapor content before and after the Landsat image was taken, t represents the Landsat image taking time, t1 and t2 represent the NCEP atmospheric water vapor content before and after the Landsat image was taken, respectively;

[0051] S4.2. Perform spatial interpolation on the atmospheric water vapor content data, interpolate the data to the target area, and then use the obtained atmospheric water vapor content data to calculate the atmospheric functions f1, f2, and f3. The formula is as follows:

[0052]

[0053] Among them, c ij is a constant related to the sensor.

[0054] Furthermore, the aforementioned step S5 includes the following sub-steps:

[0055] S5.1. Construct a thermal infrared image dataset and perform preprocessing. The dataset includes low-resolution and high-resolution thermal infrared image data in different thermal infrared bands, as well as corresponding labels.

[0056] S5.2. Construct an SRGAN model including a generator and a discriminator, input a low-resolution thermal infrared image into the SRGAN model through adversarial training, and output a high-resolution thermal infrared image;

[0057] S5.3. Construct loss functions, including content loss function, adversarial loss function and perceptual loss function. Content loss ensures that the generated image is as consistent as possible with the high-resolution image; adversarial loss trains the generator through the discriminator to improve the authenticity of the image; perceptual loss is based on the pre-trained VGG network and focuses on the similarity of image detail features:

[0058] Content loss function:

[0059] Adversarial loss function:

[0060] Perceptual loss function: Among them, W i,j is the width of the feature map, H i,j is the height of the feature map, -φ i,j is the feature map obtained after the jth convolution activation before the i-th maxpooling layer in the VGG19 network, r 2 represents the scaling factor, Represents the loss function of the standard vgg, I is the pixel, HR represents the real image, and LR represents the scaled image.

[0061] S5.4. Set the training parameters of the generator and discriminator, and start adversarial training of the SRGAN model. The model optimizes the weight parameters of the generator, generates thermal infrared images with high resolution, and refines the texture and radiation features of the crack area.

[0062] Furthermore, the aforementioned step S6 includes the following sub-steps:

[0063] S6.1. Convert the DN value of the thermal infrared band image into an absolute radiance value using the following calculation formula:

[0064]

[0065] Among them, gain is the gain constant and bias is the bias constant;

[0066] S6.2. Calculate the brightness temperature based on the radiation brightness value. The calculation formula is as follows:

[0067]

[0068] Among them, K1 and K2 are the calibration constants of the thermal infrared band.

[0069] S6.3. Combine the third-highest resolution surface emissivity data and atmospheric functions and use a single-channel algorithm to calculate the surface temperature as follows:

[0070]

[0071] Among them, b is a quantity related to the equivalent wavelength; f1, f2, and f3 are atmospheric functions.

[0072] Compared with the prior art, the beneficial technical effects of the above technical solutions adopted in the present invention are as follows: the present invention first establishes a surface fine classification model based on the U-BiFormer deep learning model, realizes the automatic recognition of the surface type corresponding to any visible light remote sensing data, and solves the problem of low accuracy of the current surface emissivity data. Then, combined with the surface emissivity calculation method in the general single-channel algorithm, linear weighted fusion is performed to further improve the accuracy of the surface emissivity. Then, the SRGAN model is trained based on the adversarial neural network algorithm, and the generator is able to generate high-quality, detail-rich super-resolution thermal infrared images through adversarial training of the generator and the discriminator. Finally, based on the high-resolution parameter data, a single-channel algorithm is used to calculate high-resolution and high-precision surface temperature data, providing a new idea for improving the accuracy and resolution of surface temperature inversion.

[0073] The present invention is mainly used for surface temperature inversion in desert areas, and can also be applied to other surface types by modifying the surface fine classification model and the surface emissivity calculation method. The high-resolution and high-precision surface temperature data obtained provide a theoretical basis for global climate change detection, disaster prevention, weather forecasting, and urban construction, with strong practicality and high application and promotion value. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 Schematic diagram of the invention process.

[0075] Figure 2 Test image of the surface fine classification model results based on U-BiFormer.

[0076] Figure 3 Flowchart of the super-resolution reconstruction model based on the adversarial neural network algorithm.

[0077] Figure 4Test image of the super-resolution reconstruction model results based on the adversarial neural network algorithm. DETAILED DESCRIPTION

[0078] In order to better understand the technical content of the present invention, specific embodiments are given and described as follows in conjunction with the accompanying drawings.

[0079] Various aspects of the invention are described herein with reference to the accompanying drawings, in which many illustrative embodiments are shown. The embodiments of the invention are not limited to those described in the accompanying drawings. It should be understood that the invention is implemented by any of the various concepts and embodiments described above, as well as the concepts and embodiments described in detail below, because the concepts and embodiments disclosed in the invention are not limited to any implementation. In addition, some aspects disclosed in the invention may be used alone or in any appropriate combination with other aspects disclosed in the invention.

[0080] refer to Figure 1 The present invention discloses a coupling method for improving the resolution and accuracy of surface temperature inversion in desert areas, which is characterized by comprising the following steps:

[0081] S1. Construct and train several remote sensing images of desert areas and corresponding data with fine surface classification labels to obtain a surface fine classification model U-BiFormer; use the surface fine classification model U-BiFormer to finely classify the surface types of remote sensing images of desert areas, and then calculate and obtain the first high-resolution surface emissivity data based on the fine classification of surface types;

[0082] S2. Classify the pixels in the image according to the normalized vegetation index to obtain vegetation coverage; calculate the bare soil emissivity of the preset band of the ASTER data in combination with the reflectivity of the ASTER data, obtain the thermal infrared band emissivity of the Landsat series satellites through conversion, and then obtain the adjusted Landsat emissivity through vegetation adjustment and interpolation calculation, and finally calculate the second high-resolution surface emissivity data;

[0083] S3, using a linear weighted fusion method to assign different weights to the first high-resolution surface emissivity data and the second high-resolution surface emissivity data, and obtaining the third high-resolution surface emissivity data after weighted calculation;

[0084] S4, using low-resolution NCEP atmospheric water vapor content data, linearly interpolating the data in space and time to obtain high-resolution atmospheric water vapor content data, and then calculating the atmospheric function;

[0085] S5. Based on the adversarial neural network algorithm, the SRGAN model is trained on the preprocessed thermal infrared dataset. Through adversarial training of the generator and the discriminator, the generator is able to generate super-resolution thermal infrared images.

[0086] S6. Calculate the high-resolution radiation brightness and brightness temperature based on the DN value of the super-resolution thermal infrared image. Finally, combine the third high-resolution surface emissivity data and atmospheric function, use a single-channel algorithm to invert the surface temperature, and finally obtain high-precision and high-resolution surface temperature data.

[0087] Further, as a preferred embodiment of the present invention, in step S1, based on the U-BiFormer model, the output of all stage decoders is used to predict the segmentation map, the software environment required for the U-BiFormer model to run is configured, and the data set prepared in step S1 is input into the model for training. After training a large amount of data, a surface classification model based on U-Biformer is obtained, and the model is used to classify the surface of a visible light remote sensing image of a desert area. The results are as follows: Figure 2 As shown in the figure, the green area is farmland, the yellow area is desert, the blue area is water, and the black part is town. This example does not involve mountains. Specifically, step S1 includes the following sub-steps:

[0088] S1.1. Obtain high-resolution visible light data from the Gaofen-1 satellite, and then select clear images from more than 1,700 remote sensing images of the African region; crop the images into 512*512 images to facilitate model training, and then select 1,000 images of each surface category, and use the Labelme image annotation tool to annotate the images; divide the training set, validation set, and test set into 8:1:1.

[0089] S1.2. Add a U-decoder to the BiFormer model to build a surface fine classification network based on deep learning. This network is a novel U-shaped Transformer model U-BiFormer. Using the U-decoder on the basis of BiFormer, the output of all stage decoders is used to predict the segmentation map, which can improve the model's ability to capture details and contextual information in the image, so that the model can better segment similar categories. Improve the hybrid attention module unique to the U-decoder, increase the proportion of the current stage features in the hybrid features, let the decoder pay more attention to the refinement of the current stage features, and improve the model's segmentation effect on similar categories;

[0090] Then, configure the software environment required to run the U-BiFormer model, input the data set prepared in step S1.1 into the model for training, and the initial training parameters are as shown in Table 1:

[0091] Table 1

[0092]

[0093] The CE+Focal mixed loss function is used when training the deep learning surface fine classification network, as follows:

[0094] Loss=Loss CE +Loss Focal

[0095] Among them, the cross entropy loss function Loss CE And the focal loss function Loss Focal The calculation formulas for a single pixel in multi-category tasks are:

[0096]

[0097] Among them, y i Represents the true label of the predicted sample, p i Indicates the probability that the model predicts that the pixel belongs to the i-th category, FL(y i ,p i ) represents the focal loss of the i-th category, α is a balancing factor used to adjust the weights between positive and negative categories, and γ is a tuning parameter used to control the weights of difficult and easy samples.

[0098] S1.4. Set the training algorithm parameters as shown in Table 2, adjust the weight connection parameters according to the loss function learning results, until the preset accuracy is reached, and obtain the surface fine classification model U-BiFormer. Use the surface fine classification model U-BiFormer to finely classify the surface types of remote sensing images of desert areas, and obtain the corresponding surface types, including towns, farmlands, deserts, water bodies, and mountains;

[0099] Table 2

[0100]

[0101]

[0102] S1.5. Take different surface reflectances for different classified surface types to obtain the first high-resolution surface reflectance data:

[0103]

[0104] The environment settings of the U-BiFormer algorithm of the surface fine classification model in this embodiment are as follows:

[0105] Type in the terminal

[0106] python"u-biformer / tools / train.py"

[0107] "u-biformer / configs / u-biformer / u-biformer____-512x512.py"

[0108] You can use the training script to train the model

[0109] Python needs to install third-party packages

[0110] 1. numpy

[0111] 2. scipy

[0112] 3. Pillow

[0113] 4. Cython

[0114] 5.matplotlib

[0115] 6.mmcv>=2.0.0rc1,<2.1.0

[0116] 7.mmengine>=0.4.0,<1.0.0

[0117] 8. prettytable

[0118] 9. torch

[0119] 10. torchvision

[0120] 11.scikit-image

[0121] 12.keras>=2.0.8

[0122] 13.opencv-python

[0123] 14.h5py

[0124] 15.imgaug

[0125] 16.codecov

[0126] 17.flake8

[0127] 18.interrogate

[0128] 19.pytest

[0129] 20.xdoctest>=0.10.0

[0130] 21.Yapf

[0131] Further, as a preferred embodiment of the present invention, step S2 includes the following sub-steps:

[0132] S2.1. Obtain the low-resolution reflectivity data of bands 13 and 14 in the ASTER dataset;

[0133] S2.2. Calculate the normalized vegetation index NDVI using near infrared NIR data and infrared R data, as follows:

[0134]

[0135] S2.3. Based on the NDVI threshold of soil and vegetation, the vegetation index is used to classify the pixels in the image into pure water bodies, mixed pixels and pure vegetation, and the vegetation coverage PV is calculated:

[0136]

[0137] S2.4. Determine the surface emissivity. When the NDVI is less than or equal to the soil threshold, the pixel is considered to belong to the water body and the emissivity is set to 0.995. When the NDVI is greater than or equal to the vegetation threshold, the pixel is considered to belong to vegetation and the emissivity is set to 0.986. For pixels that are divided into mixed pixels, the PV value that has been calculated is used for further calculation to obtain the bare soil emissivity of the ASTER bands 13 and 14, as shown in the following formula:

[0138]

[0139] Among them, ε ASTER,bare represents the ASTER bare soil emissivity, ε ASTER Indicates the emissivity of the preset band, ε ASTER,veg represents the emissivity of pure vegetation in ASTER, ε ASTER,veg =0.986;

[0140] S2.5. Convert the bare soil emissivity of ASTER bands 13 and 14 to the thermal infrared band emissivity of Landsat series satellites as follows:

[0141] ε L10,bare =c0ε A13,bare +c1ε A14,bare +c2

[0142] Among them, ε L10,bare represents the bare soil emissivity in band 10 of the Landsat TIRS sensor, ε A13,bare and ε A14,bareThey represent the bare soil emissivity of ASTER bands 13 and 14 obtained using the calculation, and c0=0.6820, c1=0.2578, and c2=0.0584 are the regression coefficients.

[0143] S2.6. Calculate the vegetation-adjusted Landsat emissivity based on the bare soil emissivity, pure vegetation emissivity, and PV as follows:

[0144] ε=PV·ε L10,veg +(1-PV)·ε L10,bare

[0145] Where ε represents the surface emissivity of the 10th band as the input for the subsequent single-channel algorithm, and ε L10,veg represents the pure vegetation emissivity of the 10th band of Landsat, ε L10,veg =0.986, ε L10,bare Represents the bare soil emissivity in Landsat Band 10.

[0146] S2.7, in the single-channel algorithm, is calculated using the following formula:

[0147] ε2

[0148] S2.8. Interpolate the acquired low-resolution surface emissivity data to the same resolution as the visible light data to obtain the second highest resolution surface reflectivity ε2.

[0149] Further, as a preferred embodiment of the present invention, step S3 includes the following sub-steps:

[0150] S3.1. Use a linear weighted fusion method to assign different weights to the first high-resolution surface reflectance data and the second high-resolution surface reflectance data.

[0151] S3.2. According to the set weight, the high-resolution surface emissivity ε finally used in the single-channel algorithm is calculated according to the following calculation formula:

[0152]

[0153] where ε i refers to the surface emissivity obtained in step i, w i Refers to the weight corresponding to the i-th step, w1=0.3, w2=0.7.

[0154] Further, as a preferred embodiment of the present invention, step S4 includes the following sub-steps:

[0155] S4.1. Download atmospheric water vapor content data from the NCEP official website, perform time interpolation on the data, and obtain the water vapor content at the time of data shooting. The formula is as follows:

[0156]

[0157] Where w represents the water vapor content at the time of Landsat acquisition, and They represent the NCEP water vapor content before and after the Landsat image was taken, t represents the Landsat image taking time, and t1 and t2 represent the NCEP atmospheric water vapor content before and after the Landsat image was taken, respectively.

[0158] S4.2. Perform spatial interpolation on the atmospheric water vapor content data, interpolate the data to the target area, and then use the obtained atmospheric water vapor content data to calculate the atmospheric functions f1, f2, and f3. The formula is as follows:

[0159]

[0160] Among them, c ij is a constant related to the sensor.

[0161] The specific values ​​are shown in Table 3 below.

[0162] Table 3

[0163]

[0164] Further, refer to Figure 3 As a preferred embodiment of the present invention, step S5 includes the following sub-steps:

[0165] S5.1. Construct a thermal infrared image dataset and perform preprocessing: The dataset includes more than 20,000 30×30 low-resolution and 15×15 high-resolution thermal infrared image data, mainly containing data from different thermal infrared bands to ensure data diversity and generalization ability; then perform image preprocessing; for the collected thermal infrared images, perform noise removal, image normalization and other operations, and perform appropriate magnification, rotation and cropping to ensure the consistency of image quality after processing; generate label files for training according to the corresponding low-resolution and high-resolution images, and divide the data into a training set and a test set with a ratio of 8:2, and then use the k-fold cross-validation method to further divide the training set to ensure the stability of the model performance;

[0166] S5.2. Construct an SRGAN model including a generator and a discriminator, input a low-resolution thermal infrared image into the SRGAN model through adversarial training, and output a high-resolution thermal infrared image: Configure the software environment required for the operation of the SRGAN model, and input the data set prepared in step S5.1 into the SRGAN model for training; the model consists of a generator (Generator) and a discriminator (Discriminator). Through adversarial training, the generator gradually learns to reconstruct low-resolution images into high-resolution images. The content loss (Content Loss) is calculated in the generator to measure the content similarity between the generated image and the target high-resolution image. The generated high-resolution image is input into the discriminator model and compared with the real high-resolution image to help the generated model learn how to better generate high-resolution images that are close to the real one. The adversarial loss (Adversarial Loss) is calculated in the discriminator to evaluate the difference between the generated image and the real image, thereby guiding the generation model to improve. The whole process optimizes the output of the generation model through the feedback of content loss and adversarial loss, so that the low-resolution image is gradually upgraded to a high-resolution image. A thermal infrared image of the study area is used for result testing. The test results are as follows: Figure 4 As shown, (a) shows the thermal infrared image before reconstruction, and (b) shows the thermal infrared image after reconstruction.

[0167] The environment settings of the algorithm in the embodiment are as follows:

[0168] python "python train.py – mode = train" # Train the model based on the data set and get the model parameters

[0169] "python train.py – mode = eval" # Perform reasoning or testing to generate high-resolution images Python needs to install third-party packages

[0170] 1. numpy

[0171] 2.cv2

[0172] 3.SSL

[0173] 4.tensorlayerx 0.5.8

[0174] 5. torch 1.10.1+cu111

[0175] 6.IPython[all]python 3.7.3

[0176] The initial training parameters are shown in Table 4:

[0177] Table 4 Initial training parameters

[0178]

[0179]

[0180] S5.4. Construct loss functions, including content loss function, adversarial loss function and perceptual loss function. Content loss ensures that the generated image is as consistent as possible with the high-resolution image; adversarial loss trains the generator through the discriminator to improve the authenticity of the image; perceptual loss is based on the pre-trained VGG network and focuses on the similarity of image detail features to achieve realistic detail performance:

[0181] Content loss function:

[0182] Adversarial loss function:

[0183] Perceptual loss function:

[0184] Among them, W i,j is the width of the feature map, H i,j is the height of the feature map, -φ i,j is the feature map obtained after the jth convolution activation before the i-th maxpooling layer in the VGG19 network, r 2 represents the scaling factor, G θG Represents the loss function of the standard vgg, I is the pixel, HR represents the real image, and LR represents the scaled image.

[0185] S5.5. Set the training parameters of the generator and discriminator, and start adversarial training of the SRGAN model. The model gradually optimizes the weight parameters of the generator to generate thermal infrared images with high resolution and refine the texture and radiation features of the crack area, making the visual effect of the reconstructed image realistic and rich in details.

[0186] Further, as a preferred embodiment of the present invention, step S6 includes the following sub-steps:

[0187] S6.1. Convert the DN value of the thermal infrared band image into an absolute radiance value using the following calculation formula:

[0188]

[0189] Among them, gain = 0.0003342 is the gain constant, bias = 0.1 is the bias constant;

[0190] S6.2. Calculate the brightness temperature based on the radiation brightness value. The calculation formula is as follows:

[0191]

[0192] Among them, K1 and K2 are the calibration constants of the thermal infrared band, K1 = 774.8853 W / m·μm·sr, K2 = 1321.0789 K;

[0193] S6.3. Combining the third-highest resolution surface reflectance data and atmospheric functions, using a single-channel algorithm to calculate the surface temperature, we can obtain a high-precision, high-resolution surface temperature result, as shown in the following formula:

[0194]

[0195] Among them, b is a quantity related to the equivalent wavelength, b=1324; f1, f2, f3 are atmospheric functions.

[0196] Based on the single-channel algorithm, the calculation of surface emissivity in the algorithm process was innovated. The surface emissivity was obtained using two methods: (1) Based on the U-BiFormer surface fine classification model, the visible light remote sensing data was finely classified, and then different surface emissivity values ​​were set for different surface types; (2) The normalized vegetation index was calculated using infrared and near-infrared data, and then the surface emissivity data was inferred by combining the reflectivity of the ASTER13 and 14 bands. Finally, the linear weighted fusion method was used to assign weights of 0.3 and 0.7 to the results obtained by the two methods, respectively, and the final surface emissivity data was obtained by weighted calculation, which improved the accuracy of the parameter and thus improved the accuracy of the surface temperature inversion results.

[0197] Although the present invention has been described above with preferred embodiments, it is not intended to limit the present invention. A person skilled in the art of the present invention may make various modifications and improvements without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention shall be determined by the definition of the claims.

Claims

1. A coupling method for improving the resolution and accuracy of surface temperature inversion in desert areas, characterized in that: The following steps are involved: S1. Construct and train several remote sensing images of desert areas and corresponding data with fine surface classification labels to obtain a surface fine classification model U-BiFormer; use the surface fine classification model U-BiFormer to finely classify the surface types of remote sensing images of desert areas, and then calculate and obtain the first high-resolution surface emissivity data based on the fine classification of surface types; S2. Classify the pixels in the image according to the normalized vegetation index to obtain vegetation coverage; calculate the bare soil emissivity of the preset band of the ASTER data in combination with the reflectivity of the ASTER data, obtain the thermal infrared band emissivity of the Landsat series satellites through conversion, and then obtain the adjusted Landsat emissivity through vegetation adjustment and interpolation calculation, and finally calculate the second high-resolution surface emissivity data; S3, using a linear weighted fusion method to assign different weights to the first high-resolution surface emissivity data and the second high-resolution surface emissivity data, and obtaining the third high-resolution surface emissivity data after weighted calculation; S4, using low-resolution NCEP atmospheric water vapor content data, linearly interpolating the data in space and time to obtain high-resolution atmospheric water vapor content data, and then calculating the atmospheric function; S5. Based on the adversarial neural network algorithm, the SRGAN model is trained on the preprocessed thermal infrared dataset. Through adversarial training of the generator and the discriminator, the generator is able to generate super-resolution thermal infrared images. S6. Calculate the high-resolution radiation brightness and brightness temperature based on the DN value of the super-resolution thermal infrared image. Finally, combine the third high-resolution surface emissivity data and atmospheric function, use a single-channel algorithm to invert the surface temperature, and finally obtain high-precision and high-resolution surface temperature data.

2. A coupling method for improving the resolution and accuracy of surface temperature inversion in desert areas according to claim 1, characterized in that: Step S1 includes the following sub-steps: S1.

1. Obtain high-resolution visible light data and filter out clear images from remote sensing images of desert areas; crop the images and add fine classification labels for surface types; then annotate the images and divide them into training sets, validation sets, and test sets; S1.

2. Add a U-type decoder based on the BiFormer model to build a surface fine classification network based on deep learning, then set the initial parameters, and use the CE+Focal mixed loss function when training the deep learning surface fine classification network, as follows: Loss=Loss CE +Loss Focal Among them, the cross entropy loss function Loss CE And the focal loss function Loss Focal The calculation formulas for a single pixel in multi-category tasks are: Among them, y i Represents the true label of the predicted sample, p i Indicates the probability that the model predicts that the pixel belongs to the i-th category, FL(y i ,p i ) represents the focal loss of the i-th category, α is a balancing factor used to adjust the weights between positive and negative categories, and γ is a tuning parameter used to control the weights of difficult and easy samples; S1.

3. Set the training algorithm parameters, adjust the weight connection parameters according to the loss function learning results, until the preset accuracy is reached, and obtain the surface fine classification model U-BiFormer. Use the surface fine classification model U-BiFormer to finely classify the surface types of remote sensing images of desert areas, and obtain the corresponding surface types, including towns, farmlands, deserts, water bodies, and mountains; S1.

4. Take different surface emissivity for different surface types after classification to obtain the first high-resolution surface emissivity data:

3. A coupling method for improving the resolution and accuracy of surface temperature inversion in desert areas according to claim 1, characterized in that: Step S2 includes the following sub-steps: S2.

1. Obtain the low-resolution reflectivity data of bands 13 and 14 in the ASTER dataset; S2.

2. Calculate the normalized vegetation index NDVI using near infrared NIR data and infrared R data, as follows: S2.

3. Based on the NDVI threshold of soil and vegetation, the vegetation index is used to classify the pixels in the image into pure water bodies, mixed pixels and pure vegetation, and the vegetation coverage PV is calculated: S2.

4. Determine the surface emissivity. When the NDVI is less than or equal to the soil threshold, the pixel is considered to belong to the water body. When the NDVI is greater than or equal to the vegetation threshold, the pixel is considered to belong to the vegetation. For pixels that are divided into mixed pixels, the PV value that has been calculated is used for further calculation to obtain the bare soil emissivity of the ASTER 13th and 14th bands, as shown in the following formula: Among them, ε ASTER,bare represents the ASTER bare soil emissivity, ε ASTER Indicates the emissivity of the preset band, ε ASTER,veg represents the emissivity of pure vegetation of ASTER; S2.

5. Convert the bare soil emissivity of ASTER bands 13 and 14 to the thermal infrared band emissivity of Landsat series satellites as follows: e L10,bare =c0ε A13,bare +c1e A14,bare +c2 Among them, ε L10,bare represents the bare soil emissivity in band 10 of the Landsat TIRS sensor, ε A13,bare and ε A14,bare They represent the bare soil emissivity of ASTER bands 13 and 14 obtained using calculations; S2.

6. Calculate the vegetation-adjusted Landsat emissivity based on the bare soil emissivity, pure vegetation emissivity, and PV as follows: ε=PV·ε L10,veg +(1-PV)·ε L10,bare Where ε represents the surface emissivity of the 10th band as the input for the subsequent single-channel algorithm, and ε L10,veg represents the pure vegetation emissivity of the 10th band of Landsat, ε L10,bare represents the bare soil emissivity in the 10th band of Landsat; S2.7, in the single-channel algorithm, is calculated using the following formula: S2.

8. Interpolate the acquired low-resolution surface emissivity data to the same resolution as the visible light data to obtain the second highest resolution surface emissivity ε2.

4. A coupling method for improving the resolution and accuracy of surface temperature inversion in desert areas according to claim 1, characterized in that: Step S3 includes the following sub-steps: S3.

1. Use the linear weighted fusion method to assign different weights to the first high-resolution surface emissivity data and the second high-resolution surface emissivity data. S3.

2. According to the set weights, the high-resolution surface emissivity ε finally used in the single-channel algorithm is calculated according to the following calculation formula: where ε i refers to the surface emissivity obtained in step i, w i Refers to the weight corresponding to the i-th step.

5. A coupling method for improving the resolution and accuracy of surface temperature inversion in desert areas according to claim 1, characterized in that: Step S4 includes the following sub-steps: S4.

1. Based on the NCEP atmospheric water vapor content data, the data is temporally interpolated to obtain the water vapor content at the time of data capture. The formula is as follows: Where w represents the water vapor content at the time of Landsat acquisition, and represents the NCEP water vapor content before and after the Landsat image was taken, t represents the Landsat image taking time, t1 and t2 represent the NCEP atmospheric water vapor content before and after the Landsat image was taken, respectively; S4.

2. Perform spatial interpolation on the atmospheric water vapor content data, interpolate the data to the target area, and then use the obtained atmospheric water vapor content data to calculate the atmospheric functions f1, f2, and f3. The formula is as follows: Among them, c ij is a constant related to the sensor.

6. A coupling method for improving the resolution and accuracy of surface temperature inversion in desert areas according to claim 1, characterized in that: Step S5 includes the following sub-steps: S5.

1. Construct a thermal infrared image dataset and perform preprocessing. The dataset includes low-resolution and high-resolution thermal infrared image data in different thermal infrared bands, as well as corresponding labels. S5.

2. Construct an SRGAN model including a generator and a discriminator, input a low-resolution thermal infrared image into the SRGAN model through adversarial training, and output a high-resolution thermal infrared image; S5.

3. Construct loss functions, including content loss function, adversarial loss function and perceptual loss function. Content loss ensures that the generated image is as consistent as possible with the high-resolution image; adversarial loss trains the generator through the discriminator to improve the authenticity of the image; perceptual loss is based on the pre-trained VGG network and focuses on the similarity of image detail features: Content loss function: Adversarial loss function: Perceptual loss function: Among them, W i,j is the width of the feature map, H i,j is the height of the feature map, -φ i,j is the feature map obtained after the jth convolution activation before the i-th maxpooling layer in the VGG19 network, r 2 represents the scaling factor, Represents the loss function of the standard vgg, I is the pixel, HR represents the real image, and LR represents the scaled image. S5.

4. Set the training parameters of the generator and discriminator, and start adversarial training of the SRGAN model. The model optimizes the weight parameters of the generator, generates thermal infrared images with high resolution, and refines the texture and radiation features of the crack area.

7. A coupling method for improving the resolution and accuracy of surface temperature inversion in desert areas according to claim 1, characterized in that: Step S6 includes the following sub-steps: S6.

1. Convert the DN value of the thermal infrared band image into an absolute radiance value using the following calculation formula: Among them, gain is the gain constant and bias is the bias constant; S6.

2. Calculate the brightness temperature based on the radiation brightness value. The calculation formula is as follows: Among them, K1 and K2 are the calibration constants of the thermal infrared band. S6.

3. Combine the third-highest resolution surface emissivity data and atmospheric functions and use a single-channel algorithm to calculate the surface temperature as follows: Among them, b is a quantity related to the equivalent wavelength; f1, f2, and f3 are atmospheric functions.

Images