WO2024113427A1 - 一种地表温度遥感产品降尺度方法、系统、设备及介质 - Google Patents

一种地表温度遥感产品降尺度方法、系统、设备及介质 Download PDF

Info

Publication number
WO2024113427A1
WO2024113427A1 PCT/CN2022/139124 CN2022139124W WO2024113427A1 WO 2024113427 A1 WO2024113427 A1 WO 2024113427A1 CN 2022139124 W CN2022139124 W CN 2022139124W WO 2024113427 A1 WO2024113427 A1 WO 2024113427A1
Authority
WO
WIPO (PCT)
Prior art keywords
surface temperature
spatial resolution
pixel
temperature data
change rate
Prior art date
Application number
PCT/CN2022/139124
Other languages
English (en)
French (fr)
Inventor
郭善昕
陈劲松
韩宇
姜小砾
Original Assignee
深圳先进技术研究院
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 深圳先进技术研究院 filed Critical 深圳先进技术研究院
Publication of WO2024113427A1 publication Critical patent/WO2024113427A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/766—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using regression, e.g. by projecting features on hyperplanes
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/13—Satellite images
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation

Definitions

  • the present invention belongs to the technical field of satellite remote sensing earth observation, and in particular relates to a method, system, equipment and medium for downscaling a surface temperature remote sensing product.
  • Land surface temperature is an important component of the surface energy and water balance, providing information on the spatiotemporal changes of the surface balance state from local to global.
  • land surface temperature is widely used in evapotranspiration estimation, urban heat island monitoring, local climate change, vegetation monitoring, agricultural monitoring, and forest fire monitoring; however, due to the relatively low energy of surface thermal radiation and the limitation of satellite sensors on cost and hardware technology, the remote sensing data obtained by a single satellite sensor currently has a "space-time contradiction", that is, the obtained surface temperature is difficult to meet both high spatial resolution and high temporal resolution, which greatly limits the application demand of surface temperature.
  • the algorithm theory it is divided into downscaling methods based on scale factors and downscaling methods based on space-time fusion; among them, the downscaling method based on space-time fusion can more effectively reduce the model's dependence on auxiliary data, and realize the downscaling of surface temperature at low cost, conveniently and efficiently;
  • weight function-based methods learning-based methods and unmixing-based methods; among them, for LST data with strong temporal fluctuations and not rich spectral features, the parameters of the weight function method are more sensitive, which also affects the effectiveness of the training amount of the learning method, resulting in a lower overall fusion effect of the two methods; for the LST that changes at all times, the unmixing-based method combines the assumption that the change trend ratios of different sensors are equal in areas with the same LST change trend, which can improve the fusion effect of LST in different phases.
  • the model based on the unmixing strategy is the main model.
  • This type of model can effectively reduce the impact of surface environmental changes on prediction accuracy.
  • this type of method has low learning cost and high time efficiency.
  • the existing model still has the following two shortcomings:
  • the linear unmixing model cannot effectively balance the number of homogeneous units and the unmixing accuracy. Specifically, the explanatory, high-spatial-resolution surface problem information requires a large number of surface homogeneous units. However, due to the insufficient solving ability of the current linear unmixing model, when the number of homogeneous units exceeds a certain range, the linear unmixing model cannot be solved.
  • the remote sensing image spatial downscaling model cannot effectively downscale the surface temperature in scenarios where the surface temperature changes rapidly.
  • There are technical defects such as insufficient accuracy of the daily high spatial resolution surface temperature products produced and the inability to provide accurate high temporal and spatial resolution surface temperature products.
  • the purpose of the present invention is to provide a method, system, device and medium for downscaling surface temperature remote sensing products to solve the technical problems that the existing linear unmixing model cannot effectively estimate the surface temperature and the existing weight function fails near the rate of change -1.
  • a first aspect of the present invention provides a method for downscaling a land surface temperature remote sensing product, comprising the following steps:
  • the original low spatial resolution surface temperature data before T and after T are resampled to a spatial resolution consistent with the original high spatial resolution surface temperature data, and the sampled low spatial resolution surface temperature data before T and after T are obtained, as well as the ID value of each pixel after sampling corresponding to the original pixel;
  • the pixel-by-pixel change rate is calculated at three time points before T, T, and after T to obtain the pixel-by-pixel change rate of the low spatial resolution surface temperature data;
  • the pixel-by-pixel change rate of low spatial resolution surface temperature data, the ID value of each pixel corresponding to the original pixel after sampling, and the pre-acquired surface homogeneous unit ID are taken as input, and the pre-trained dynamic multi-layer perception network DyNet is used to predict the category-by-category change rate of high spatial resolution surface temperature data; the absolute value of the difference between the original high spatial resolution surface temperature data before T and after T and the obtained category-by-category change rate of high spatial resolution surface temperature data are taken as input, and the pre-trained weight perception network WNet is used to predict the surface temperature at the target time T.
  • a further improvement of the present invention is that the step of obtaining the pre-trained dynamic multi-layer perception network DyNet comprises:
  • each training sample in the first training sample set includes as input the pixel-by-pixel change rate of low spatial resolution surface temperature data, the ID value of each pixel corresponding to the original pixel after sampling and the surface homogeneous unit ID, and as a label the category-by-category change rate of high spatial resolution surface temperature data; wherein the category change rate of high spatial resolution surface temperature is based on the original high spatial resolution surface temperature data before T, T, and after T, and is calculated pixel by pixel at three time points before T, T, and after T, and is obtained by using the surface homogeneous unit ID to obtain the average value of each category; the surface homogeneous unit ID is obtained by clustering time series surface temperature data or land cover type data;
  • the pre-constructed dynamic multi-layer perception network is trained using the MSE Loss function. After reaching the preset convergence condition, the pre-trained dynamic multi-layer perception network is obtained.
  • the pre-constructed dynamic multi-layer perception network DyNet includes: a temporary retreat layer for shielding some input neurons, a plurality of fully connected layers and Relu activation function layers connected in series after the temporary retreat layer, and a temporary retreat layer for shielding some output neurons.
  • a further improvement of the present invention is that the step of obtaining the pre-trained weight-aware network WNet comprises:
  • each training sample in the second training sample set includes the absolute value of the difference between the high spatial resolution surface temperature data before T and after T as input and the category-by-category change rate of the high spatial resolution surface temperature data in the first training sample set, and the high spatial resolution surface temperature of the target date as a label;
  • the pre-constructed weight-aware network is trained using the MSE Loss function. After reaching the preset convergence condition, the pre-trained weight-aware network is obtained.
  • the pre-constructed weight-aware network includes: a data mapping layer for inputting data, and a plurality of fully connected layers and Relu activation function layers connected in series after the data mapping layer.
  • a further improvement of the present invention is that, in the pre-constructed weight-aware network WNet, the step of performing data mapping by the data mapping layer comprises:
  • ⁇ LST LST-LST before ;
  • LST is the real surface temperature LST at the target time T
  • LST before is the surface temperature LST before T
  • ⁇ LST is the temperature difference
  • the transformation expression of the category-by-category change rate of the high spatial resolution image is: In the formula, is the change rate of each category of high spatial resolution image, is the category-by-category change rate after transformation;
  • the expression of the predicted value of surface temperature output by the weight perception network is: is the land surface temperature LST predicted by the model.
  • a further improvement of the present invention is that the steps of obtaining the first training sample set and obtaining the second training sample set specifically include:
  • the low spatial resolution surface temperature data is resampled to a spatial resolution consistent with the high spatial resolution surface temperature data according to the principle of nearest neighbor sampling, forming the sampled low spatial resolution surface temperature data and the ID value of each pixel corresponding to the original pixel after sampling;
  • the time series high spatial resolution surface temperature data are used as multiple features to cluster and generate surface homogeneous units
  • the pixel-by-pixel change rate of the low spatial resolution surface temperature data and the high spatial resolution surface temperature data after sampling is calculated at three time points: the moment before the target moment, the target moment, and the moment after the target moment, to obtain the pixel-by-pixel change rate of the low spatial resolution surface temperature data and the pixel-by-pixel change rate of the high spatial resolution surface temperature data; among which, the change rate calculation expression is,
  • ⁇ Lke (i,j) and ⁇ Lok (i,j) are the differences of Landsat remote sensing reflectance in two periods [ tk , te ] and [ t0 , tk ] respectively;
  • Le (i,j) is the Landsat remote sensing reflectance at the end date te;
  • Lk (i,j) is the Landsat remote sensing reflectance at the target date tk;
  • L0 (i,j) is the Landsat remote sensing reflectance at the start date t0.; subscript 0 is the start date, k is the target date; e is the end date;
  • (i,j) represents the pixel in the i-th row and j-th column;
  • the generated surface homogeneous units are used to calculate the average pixel-by-pixel change rate of the generated high spatial resolution surface temperature data, and the category-by-category change rate of the high spatial resolution surface temperature data is obtained;
  • the input features of the first training set are the pixel-by-pixel change rate of low spatial resolution surface temperature data, the ID value of each pixel corresponding to the original pixel after sampling, and the surface homogeneous unit, and the output label is the category-by-category change rate of high spatial resolution surface temperature data;
  • Construct a second QR code training set wherein the input features of the second training set are the absolute value of the difference between the high spatial resolution surface temperature data at one moment before the target moment and at one moment after the target moment and the category-by-category change rate of the high spatial resolution surface temperature data, and the output label is the surface temperature at the target moment.
  • a second aspect of the present invention provides a system for downscaling a land surface temperature remote sensing product, comprising:
  • a data acquisition module is used to acquire original low spatial resolution surface temperature data and original high spatial resolution surface temperature data before and after the target time T; wherein, before T is any time with data before the target time T, and after T is any time with data after the target time T;
  • a data processing module is used to resample the original low spatial resolution surface temperature data before T and after T to a spatial resolution consistent with the original high spatial resolution surface temperature data, and obtain the sampled low spatial resolution surface temperature data before T and after T and the ID value of each pixel after sampling corresponding to the original pixel;
  • the change rate acquisition module calculates the pixel-by-pixel change rate of the low spatial resolution surface temperature data after sampling before T and after T according to the three time points before T, T, and after T to obtain the pixel-by-pixel change rate of the low spatial resolution surface temperature data;
  • the prediction module is used to take the pixel-by-pixel change rate of low spatial resolution surface temperature data, the ID value of each pixel corresponding to the original pixel after sampling, and the pre-acquired surface homogeneous unit ID as input, and use the pre-trained dynamic multi-layer perception network DyNet to predict the category-by-category change rate of high spatial resolution surface temperature data; take the absolute value of the difference between the original high spatial resolution surface temperature data before T and after T and the obtained category-by-category change rate of high spatial resolution surface temperature data as input, and use the pre-trained weight perception network WNet to predict the surface temperature at the target time T.
  • a third aspect of the present invention provides an electronic device, comprising:
  • At least one processor and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the surface temperature remote sensing product downscaling method as described in any one of the above-mentioned items of the present invention.
  • a fourth aspect of the present invention provides a computer-readable storage medium storing a computer program, characterized in that when the computer program is executed by a processor, it implements any of the above-mentioned methods for downscaling surface temperature remote sensing products of the present invention.
  • the present invention has the following beneficial effects:
  • the present invention provides a new method for downscaling surface temperature remote sensing products by utilizing the strong nonlinear fitting ability of the multi-layer perception network and the high prediction stability in the change rate interval of -1; the technical solution of the present invention constructs a nonlinear unmixing model through a dynamic multi-layer perception network, which can realize dynamic unmixing from low-resolution satellite pixels to high-resolution satellite pixels; the technical solution of the present invention uses a weight perception network to effectively avoid the failure problem of the existing weight function near the change rate of -1.
  • the method for downscaling surface temperature remote sensing products provided by the present invention can effectively downscale the surface temperature and improve the accuracy of the surface temperature product.
  • FIG1 is a schematic flow chart of a method for downscaling a land surface temperature remote sensing product provided by an embodiment of the present invention
  • FIG2 is a schematic diagram of a process for obtaining a training sample data set in an embodiment of the present invention
  • FIG3 is a schematic diagram of a process for training a dynamic multilayer perceptron (DyNet) in an embodiment of the present invention
  • FIG4 is a schematic diagram of a process of training a weight-aware network (WNet) in an embodiment of the present invention
  • FIG. 5 is a schematic diagram of a process for predicting target date data in an embodiment of the present invention.
  • FIG. 6 is a schematic diagram of a flow chart of data post-processing in an embodiment of the present invention.
  • a method for downscaling a land surface temperature remote sensing product includes the following steps:
  • Step S1 obtaining original low spatial resolution surface temperature data and original high spatial resolution surface temperature data before and after the target time T; wherein, before T is the moment before the target time T, and after T is the moment after the target time T;
  • Step S2 resampling the original low spatial resolution surface temperature data before T and after T to a spatial resolution consistent with the original high spatial resolution surface temperature data, obtaining the sampled low spatial resolution surface temperature data before T and after T and the ID value of each pixel after sampling corresponding to the original pixel;
  • Step S3 based on the sampled low spatial resolution surface temperature data before T and after T, the pixel-by-pixel change rate is calculated at three time points before T, T, and after T to obtain the pixel-by-pixel change rate of the low spatial resolution surface temperature data;
  • Step S4 taking the pixel-by-pixel change rate of the low spatial resolution surface temperature data, the ID value of each pixel corresponding to the original pixel after sampling, and the pre-acquired surface homogeneous units as input, and using a pre-trained dynamic multi-layer perception network to predict the category-by-category change rate of the high spatial resolution surface temperature data; taking the absolute value of the difference between the original high spatial resolution surface temperature data before T and after T and the predicted category-by-category change rate of the high spatial resolution surface temperature data as input, and using a pre-trained weight perception network to predict the surface temperature at the target time T.
  • the method provided by the embodiment of the present invention can effectively downscale the surface temperature based on the pre-trained dynamic multi-layer perception network and the weight perception network, and downscale the original daily low spatial resolution data to daily high spatial resolution data, thereby improving the accuracy of the surface temperature product and serving the high-precision and high-frequency satellite remote sensing monitoring of the surface temperature.
  • Step 1 data preprocessing, including: performing necessary spatial registration, homogeneous unit division, change rate calculation and pixel index calculation on the original low spatial resolution surface temperature data and high spatial resolution surface temperature data to provide sufficient training data for subsequent model training;
  • Step 2 model training, includes: generating training samples using the data obtained in step 1, and training the multi-layer perception network (DyNet) and the weight perception network (WNet).
  • DyNet multi-layer perception network
  • WNet weight perception network
  • the embodiment of the present invention is exemplary, and on the basis of retaining the optimal model in step 2, surface temperature prediction is carried out for the target date.
  • the embodiment of the present invention also includes data post-processing after obtaining the surface temperature prediction value, including: fixing the target date, selecting multiple groups of observation data pairs in the time series, repeatedly predicting the surface temperature on the target date, integrating multiple prediction results according to certain rules, and synthesizing the final data product.
  • the steps of obtaining training sample data for training a dynamic multi-layer perception network and a weight perception network specifically include:
  • Step 1) resample the historical low spatial resolution image (for example, MODIS 1km scale surface temperature product) to a spatial resolution consistent with the high spatial resolution image (for example, Landsat 30m scale surface temperature product) according to the principle of nearest neighbor sampling, to form a sampled low spatial resolution image (which can be recorded as M30); and the ID value of each pixel after sampling corresponding to the original MODIS pixel (which can be recorded as MID);
  • a spatial resolution image for example, MODIS 1km scale surface temperature product
  • a spatial resolution consistent with the high spatial resolution image for example, Landsat 30m scale surface temperature product
  • Step 2) using historical high spatial resolution images (for example, Landsat 30m scale surface temperature products) as multiple features, using the unsupervised clustering model ISOData algorithm, clustering into multiple surface homogeneous units, recorded as ClassID;
  • historical high spatial resolution images for example, Landsat 30m scale surface temperature products
  • Step 3 in chronological order, the pixel-by-pixel change rate of the sampled low spatial resolution image and the high spatial resolution image is calculated at every three time points (before T, during T, after T), and finally the pixel-by-pixel change rate of the low spatial resolution image (which can be recorded as M_ratio) and the pixel-by-pixel change rate of the high spatial resolution image (which can be recorded as L_ratio) are obtained;
  • ⁇ Lke (i,j) and ⁇ Lok (i,j) are the differences of Landsat remote sensing reflectance in two periods [ tk , te ] and [ t0 , tk ] respectively;
  • Le (i,j) is the Landsat remote sensing reflectance at the end date te;
  • Lk (i,j) is the Landsat remote sensing reflectance at the target date tk;
  • L0 (i,j) is the Landsat remote sensing reflectance at the start date t0.; subscript 0 is the start date, k is the target date; e is the end date;
  • (i,j) represents the pixel in the i-th row and j-th column;
  • Step 4 using the surface homogeneous unit (ClassID) generated in step 2), the pixel-by-pixel change rate of the high spatial resolution image generated in step 3) is averaged to obtain the high spatial resolution class-by-class change rate (denoted as L_ratio_C);
  • Step 5 construct a dynamic multi-layer perception network (DyNet) training data set;
  • the input of the data set is a three-channel feature image, the three channels are the pixel-by-pixel change rate of the low spatial resolution image (M_ratio), the ID value of each pixel corresponding to the original MODIS pixel after sampling (MID) and the surface homogeneous unit (ClassID);
  • the output of the data set is a single-channel feature image, which is the class-by-class change rate of the high spatial resolution image (L_ratio_C);
  • Step 6 construct a weighted network WNet training data set;
  • the input of the data set is a two-channel feature image, the two channels are the absolute value of the difference between the high-resolution image before time T and the time after time T, and the category-by-category change rate of the high-spatial resolution image;
  • the output of the data is a single-channel feature image, which is the true surface temperature at the target time T (LST at T).
  • the steps of training a dynamic multi-layer perception network specifically include:
  • a dropout layer for input data is provided, and the dropout layer (DropOut) is used to use the MID in the input data to only activate the neural nodes existing in the current input data and shield the neural nodes not in the current input data; after the dropout layer, five fully connected layers (FC) and Relu activation function layers are passed continuously, and specifically, each fully connected layer may have 128 neurons; after the fifth fully connected layer (FC) and Relu activation function layer, another dropout layer is provided, and whether the neurons in the dropout layer are activated or not is determined according to the ClassID of the surface homogeneous unit in the input data;
  • MSE Loss root mean square loss
  • P_ratio_C predicted category-by-category change rate
  • L_ratio_C true value
  • the steps of training the weight-aware network specifically include:
  • a data mapping layer is provided; five fully connected layers (FC) and Relu activation function layers are continuously provided after the data mapping layer, and specifically, illustratively, the number of neurons in each fully connected layer may be 40;
  • the input parameters pass through the data mapping layer, where the data mapping is mapped according to the following formula:
  • the surface temperature (true value) on the target date is used to calculate the root mean square loss (MSE Loss), and back propagation is performed to obtain the trained weight perception network after reaching the preset convergence condition.
  • MSE Loss root mean square loss
  • the dynamic multi-layer perception network in the embodiment of the present invention can be replaced by other deep networks or deep convolutional network structures including multiple fully connected layer structures and temporary retreat layer structures, and the weight perception network in the present invention can be replaced by other deep networks or deep convolutional network structures including multiple fully connected layer structures.
  • the specific application process of the method provided in the embodiment of the present invention includes: based on the full training of the dynamic multi-layer perception network DyNet and the weight perception network WNet, the high-resolution and low-resolution digital image data of the previous moment (before T) and the next moment (after T) are used to predict the data of the target date (at the middle moment of T); the specific process includes the following steps:
  • the low spatial resolution image in the current image pair (T before - T during - T after) is resampled to the spatial resolution consistent with the high spatial resolution image according to the principle of nearest neighbor sampling, forming a sampled low spatial resolution image and an ID value corresponding to the original MODIS pixel for each pixel after sampling;
  • DyNet dynamic multi-layer perception network
  • the dataset is a three-channel feature image.
  • the three channels are the pixel-by-pixel change rate (M_ratio) of the low spatial resolution image, the ID value (MID) of each pixel corresponding to the original MODIS pixel after sampling, and the surface homogeneous unit (ClassID); use the trained dynamic multi-layer perception network (DyNet) to predict the change rate of homogeneous units and obtain the prediction results;
  • a weight-aware network (WNet) dataset of the current image pair is constructed.
  • the input of the dataset is a two-channel feature image.
  • the two channels are the absolute value of the difference between the high-resolution image (L30) at time T before and time T after, and the predicted value of the homogeneous unit change rate.
  • the trained weight-aware network (WNet) is used to predict the surface temperature on the target date (time T).
  • the preferred embodiment of the present invention further includes data post-processing, which is to use multiple sets of date pairs to predict the same day, and merge multiple prediction values to achieve the purpose of reducing the prediction error.
  • the specific process includes:
  • the difference in surface temperature between the previous and next dates is used to set the threshold, mask out pixels with smaller differences, and predict pixels with lower credibility.
  • the median value in the time series is taken as the final high-resolution surface temperature prediction result for the target date.
  • the technical solution provided by the embodiment of the present invention constructs a nonlinear unmixing model by designing a new dynamic multi-layer perception network, thereby realizing dynamic unmixing of low-resolution satellite pixels (the present invention takes MODIS satellite as an example) to high-resolution satellite pixels (the present invention takes Landsat satellite as an example);
  • the established dynamic multi-layer perception network can effectively solve the problem that the existing linear unmixing model cannot effectively estimate the surface temperature, and essentially avoids the problem that cannot be solved due to the contradiction between the number of homogeneous units on the surface and the number of available equations.
  • the embodiment of the present invention effectively avoids the failure problem of the existing weight function near the rate of change -1 by designing a new multi-layer weight perception network.
  • the network model established by the present invention is constructed based on a mathematical mechanism, and under the premise that the solution conditions remain unchanged, cross-regional and cross-time model generalization can be achieved.
  • An embodiment of the present invention provides a system for downscaling a land surface temperature remote sensing product, comprising:
  • a data acquisition module is used to acquire original low spatial resolution surface temperature data and original high spatial resolution surface temperature data before and after the target time T; wherein, before T is the moment before the target time T, and after T is the moment after the target time T;
  • a data processing module is used to resample the original low spatial resolution surface temperature data before T and after T to a spatial resolution consistent with the original high spatial resolution surface temperature data, and obtain the sampled low spatial resolution surface temperature data before T and after T and the ID value of each pixel after sampling corresponding to the original pixel;
  • the change rate acquisition module is used to calculate the pixel-by-pixel change rate of the low spatial resolution surface temperature data after sampling before T and after T according to the three time points before T, T, and after T, so as to obtain the pixel-by-pixel change rate of the low spatial resolution surface temperature data;
  • the prediction module is used to take the pixel-by-pixel change rate of low spatial resolution surface temperature data, the ID value of each pixel corresponding to the original pixel after sampling, and the pre-acquired surface homogeneous units as input, and use the pre-trained dynamic multi-layer perception network to predict the category-by-category change rate of high spatial resolution surface temperature data; take the absolute value of the difference between the original high spatial resolution surface temperature data before T and after T and the predicted category-by-category change rate of high spatial resolution surface temperature data as input, and use the pre-trained weight perception network to predict the surface temperature at the target time T.
  • the weight function-based method assumes that the temporal variation of surface reflectance is consistent in spatial scale, and combines the information of all input images through the weight function to predict the pixel value of the target image with high spatial resolution.
  • Gao et al. first proposed the Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM), which assumes that the pixels in the coarse-resolution image are pure pixels, introduces the concepts of moving windows and similar pixels, and calculates weights from spectral differences, temporal differences, and spatial distances; for areas where land types change in complex landforms, the STARFM method cannot accurately predict the real land object type in the unknown phase, resulting in large errors in the fusion results.
  • STARFM Spatial and Temporal Adaptive Reflectance Fusion Model
  • STAARCH Spatial Temporal Adaptive Algorithm for Mapping Reflectance Change
  • ESTARFM enhanced spatial and temporal adaptive reference fusion model
  • Zhao et al. proposed a robust adaptive spatial and temporal fusion model (RASTFM), which consists of a weighted average module based on non-local linear regression and an image super-resolution module based on non-local linear regression, which are used to predict non-shape changes (including phenological changes and land cover changes without shape changes) and shape changes (land cover changes with shape changes) of the surface, respectively.
  • RASTFM adaptive spatial and temporal fusion model
  • the STARFM series of models have simple parameters and high computational efficiency, but they are still limited in predicting changes in complex land cover and land cover types at different time phases.
  • the learning-based method uses a machine learning algorithm to model high- and low-resolution image pairs in a nonlinear way to predict high-spatial and temporal resolution images.
  • Huang et al. proposed a sparse representation-based spatiotemporal reflectance fusion model (SPSTFM), which uses dictionary pair learning to establish the corresponding reflectivity change relationship between low spatial resolution images and high spatial resolution images, and predicts the target high-resolution image through time weighting.
  • SPSTFM sparse representation-based spatiotemporal reflectance fusion model
  • STFDCNN deep convolutional neural networks
  • combining convolutional neural networks and nonlinear mapping models to design a dual convolutional neural network, which can automatically extract image features and improve prediction accuracy, effectively solving the defects of sparse representation methods.
  • Learning-based methods can capture more surface spatial details and are suitable for heterogeneous areas, but such methods are overly dependent on training samples and model parameters and have high time complexity.
  • Unmixing-based methods make up for the shortcomings of the above methods. Based on the linear spectral mixing theory, such methods extract the category and abundance of objects from high spatial resolution images to decompose low spatial resolution pixels to obtain the spectral values of the categories.
  • the implementation process generally includes endmember selection, abundance calculation, unmixing, etc.
  • STDFA spatial temporal data fusion approach
  • the STDFA algorithm does not take into account the spatial heterogeneity of image endmember reflectance, Zhang et al. based on the STDFA algorithm, combined with the multi-scale segmentation algorithm and the ISODATA algorithm to generate classification maps, and used the moving window method to unmix low spatial resolution images.
  • the concept of time weight is introduced to predict unknown phase images; in order to improve the prediction accuracy of the model in the area where the land object type changes, Huang et al.
  • U-STFM unmixing-based spatio-temporal reflectance fusion model
  • HCRs homogeneous change regions
  • This type of algorithm has small computational complexity and strong operability, and improves the fusion accuracy of remote sensing images for land object types that change over time, and is often used in reflectance data.
  • the parameters of the weight function method are more sensitive, which also affects the effectiveness of the training amount of the learning method, resulting in a low overall fusion effect of the two methods; while the method based on unmixing improves the fusion effect of LST at different phases by combining the assumption that the change trend ratios of different sensors are equal in the area with the same LST change trend for LST that changes at all times.
  • the technical solution provided by the embodiment of the present invention constructs a nonlinear unmixing model through a dynamic multi-layer perception network, which can realize dynamic unmixing of low-resolution satellite pixels to high-resolution satellite pixels; the technical solution provided by the embodiment of the present invention can effectively avoid the failure problem of the existing weight function near the change rate -1 through the weight perception network.
  • a computer device in another embodiment, includes a processor and a memory, the memory is used to store a computer program, the computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer storage medium.
  • the processor may be a central processing unit (CPU), and may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc., which are the computing core and control core of the terminal, which are suitable for implementing one or more instructions, and are specifically suitable for loading and executing one or more instructions in the computer storage medium to implement the corresponding method flow or corresponding function; the processor described in the embodiment of the present invention can be used for the operation of the downscaling method of the surface temperature remote sensing product.
  • the present invention further provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device for storing programs and data.
  • a storage medium specifically a computer-readable storage medium (Memory), which is a memory device in a computer device for storing programs and data.
  • the computer-readable storage medium here can include both built-in storage media in the computer device and, of course, extended storage media supported by the computer device.
  • the computer-readable storage medium provides a storage space, which stores the operating system of the terminal.
  • one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space, and these instructions can be one or more computer programs (including program codes).
  • the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory (non-volatile memory), such as at least one disk memory.
  • the processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the method for downscaling the surface temperature remote sensing product in the above embodiment.
  • the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.
  • a computer-usable storage media including but not limited to disk storage, CD-ROM, optical storage, etc.
  • each flow process and/or box in the flow chart and/or block diagram and the combination of the flow process and/or box in the flow chart and/or block diagram can be realized by computer program instructions.
  • These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processing machine or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device for realizing the function specified in one flow chart or multiple flows and/or one box or multiple boxes of the block diagram.
  • These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device that implements the functions specified in one or more processes in the flowchart and/or one or more boxes in the block diagram.
  • These computer program instructions may also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more processes in the flowchart and/or one or more boxes in the block diagram.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Computing Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Software Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Multimedia (AREA)
  • Biophysics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Medical Informatics (AREA)
  • Databases & Information Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Molecular Biology (AREA)
  • Remote Sensing (AREA)
  • Astronomy & Astrophysics (AREA)
  • Radiation Pyrometers (AREA)
  • Image Analysis (AREA)

Abstract

本发明公开了一种地表温度遥感产品降尺度方法、系统、设备及介质，所述方法包括以下步骤：获取目标时刻T的T 前、T 后的原始低空间分辨率地表温度数据和原始高空间分辨率地表温度数据；获得T 前、T 后的采样后低空间分辨率地表温度数据以及采样后每个像素对应于原始像素的ID值；获得低空间分辨率地表温度数据逐像素变化率；利用预训练好的动态多层感知网络DyNet预测获得高空间分辨率地表温度数据逐类别变化率；利用预训练好的权重感知网络WNet预测获得目标时刻T的地表温度。本发明提供的技术方案能够解决现有线性解混模型无法有效估计地表温度以及现有权重函数在变化率-1附近失效的技术问题。

Description

一种地表温度遥感产品降尺度方法、系统、设备及介质 技术领域
本发明属于卫星遥感对地观测技术领域，特别涉及一种地表温度遥感产品降尺度方法、系统、设备及介质。
背景技术
地表温度(Land Surface Temperature，LST)是地表能量和水平衡中的一个重要成分，提供了从局部到全球表面平衡状态的时空变化信息。现如今，地表温度被广泛应用于蒸散发估算、城市热岛监测、局部气候变化、植被监测、农业监测以及森林火灾监测等；然而，由于地表热辐射的能量相对较低以及卫星传感器受限于成本和硬件技术，目前单一的卫星传感器获取的遥感数据存在“时空的矛盾”，即获取的地表温度难以同时满足高空间分辨率和高时间分辨率，极大程度上限制了地表温度的应用需求。
针对如何获取高时空LST这一问题，众多学者已提出了不同LST降尺度方法，包括：
1)按照算法理论分为基于尺度因子的降尺度方法和基于时空融合的降尺度方法；其中，基于时空融合的降尺度方法更能有效降低模型对辅助数据的依懒性，且低成本、便捷高效地实现地表温度的降尺度；
2)按照实现原理可以分为基于权重函数的方法、基于学习的方法和基于解混的方法；其中，对于时间变化波动性强、光谱特征不丰富的LST数据，使权重函数类方法的参数更加敏感，也影响学习类方法训练量的有效性，导致这俩种方法总体融合效果较低；基于解混的方法对于时刻发生变化的LST，在LST变化趋势相同的区域上结合不同传感器变化趋势比相等的假设，可提高不同时相LST的融合效果。
当前基于解混策略的模型是主要模型，该类模型能够有效减少地表环境变化对预测精度的影响，同时相比于基于学习的模型，该类方法学习成本低，时间效率比较高；然而，现有该类 模型尚存在以下两点缺点：
1)线性解混模型无法有效平衡刻画同质单元数量与解混精度；具体解释性的，高空间分辨率的地表问题信息一方面需要大量的地表同质单元，但由于当前线性解混模型求解能力的不足，导致当同质单元数量超过一定范围，线性解混模型出现无法解算的情况；
2)当前的权重函数由于其数学本身缺点的影响，导致在地表温度前后变化率为-1的区间中，出现无法预测的情况，导致地表温度的计算出现较大的误差。
综上所述，现有的LST降尺度方法中，遥感影像空间降尺度模型无法在地表温度快速变化的场景下，对地表温度进行有效的降尺度估计，存在生产的逐日高空间分辨率地表温度产品精度不足以及无法提供准确的高时空分辨率地表温度产品的技术缺陷。
发明内容
本发明的目的在于提供一种地表温度遥感产品降尺度方法、系统、设备及介质，以解决现有线性解混模型无法有效估计地表温度以及现有权重函数在变化率-1附近失效的技术问题。
为达到上述目的，本发明采用以下技术方案：
本发明第一方面提供的一种地表温度遥感产品降尺度方法，包括以下步骤：
获取目标时刻T的T 前、T 后的原始低空间分辨率地表温度数据和原始高空间分辨率地表温度数据；其中，T 前为目标时刻T之前的任意有数据时刻，T 后为目标时刻T之后的任意有数据时刻；
将T 前、T 后的原始低空间分辨率地表温度数据重采样到与原始高空间分辨率地表温度数据一致的空间分辨率，获得T 前、T 后的采样后低空间分辨率地表温度数据以及采样后每个像素对应于原始像素的ID值；
基于T 前、T 后的采样后的低空间分辨率地表温度数据，按照T 前、T、T 后三个时间点计算逐像素的变化率，获得低空间分辨率地表温度数据逐像素变化率；
将低空间分辨率地表温度数据逐像素变化率、采样后每个像素对应于原始像素的ID值以及预获取的地表同质单元ID作为输入，利用预训练好的动态多层感知网络DyNet预测获得高空间分辨率地表温度数据逐类别变化率；将T 前、T 后的原始高空间分辨率地表温度数据的差值的绝对值和获得的高空间分辨率地表温度数据逐类别变化率作为输入，利用预训练好的权重感知网络WNet预测获得目标时刻T的地表温度。
本发明的进一步改进在于，所述预训练好的动态多层感知网络DyNet的获取步骤包括：
获取第一训练样本集；其中，所述第一训练样本集中的每个训练样本均包括作为输入的低空间分辨率地表温度数据逐像素变化率、采样后每个像素对应于原始像素的ID值和地表同质单元ID，以及作为标签的高空间分辨率地表温度数据逐类别变化率；其中，所述高空间分辨率地表温度类别变化率是基于T 前、T、T 后的原始高空间分辨率地表温度数据，按照T 前、T、T 后三个时间点逐像素计算，并利用地表同质单元ID求取逐个类别的均值得到；地表同质单元ID由时间序列地表温度数据聚类或土地覆被类型数据获得；
基于所述第一训练样本集对预构建的动态多层感知网络，采用MSE Loss损失函数进行训练，达到预设收敛条件后，获得所述预训练好的动态多层感知网络。
本发明的进一步改进在于，所述预构建的动态多层感知网络DyNet包括：用于屏蔽部分输入神经元的暂退层，设置于暂退层后串接的多个全连接层和Relu激活函数层，以及用于屏蔽部分输出神经元的暂退层。
本发明的进一步改进在于，所述预训练好的权重感知网络WNet的获取步骤包括：
获取第二训练样本集；其中，所述第二训练样本集中的每个训练样本均包括作为输入的高空间分辨率地表温度数据T 前和T 后的差值的绝对值和所述第一训练样本集中所述高空间分辨率地表温度数据逐类别变化率，以及作为标签的目标日期的高空间分辨率地表温度；
基于所述第二训练样本集对预构建的权重感知网络，采用MSE Loss损失函数进行训练， 达到预设收敛条件后，获得所述预训练好的权重感知网络。
本发明的进一步改进在于，所述预构建的权重感知网络包括：用于输入数据的数据映射层，以及设置于数据映射层后串接的多个全连接层和Relu激活函数层。
本发明的进一步改进在于，所述预构建的权重感知网络WNet中，所述数据映射层进行数据映射的步骤包括：
对标签真值数据进行变换，变换表达式为，ΔLST＝LST-LST 前；式中，LST为目标时刻T真实的地表温度LST，LST 前为T 前时刻的地表温度LST，ΔLST为温度差；
当T 后时刻的地表温度值小于T 前时刻的地表温度值时，高空间分辨率影像逐类别变化率的变换表达式为，
Figure PCTCN2022139124-appb-000001
式中，
Figure PCTCN2022139124-appb-000002
为高空间分辨率影像逐类别变化率，
Figure PCTCN2022139124-appb-000003
为变换后的逐类别变化率；
当T 后时刻的地表温度值大于T 前时刻的地表温度值时，高空间分辨率影像逐类别变化率的变换表达式为，
Figure PCTCN2022139124-appb-000004
权重感知网络输出的地表温度预测值的表达式为，
Figure PCTCN2022139124-appb-000005
为模型预测的地表温度LST。
本发明的进一步改进在于，获取第一训练样本集和获取第二训练样本集的步骤具体包括：
获取样本原始的低空间分辨率地表温度数据和高空间分辨率地表温度数据；
将低空间分辨率地表温度数据按照最邻近采样的原则，重采样到与高空间分辨率地表温度数据一致的空间分辨率，形成采样后低空间分辨率地表温度数据以及采样后每个像素对应于原始像素的ID值；
将时间序列高空间分辨率地表温度数据作为多个特征，聚类生成地表同质单元；
按照时间顺序，采样后低空间分辨率地表温度数据和高空间分辨率地表温度数据按照目标时刻前一时刻、目标时刻、目标时刻后一时刻三个时间点计算逐像素的变化率，获得低空间分 辨率地表温度数据逐像素变化率和高空间分辨率地表温度数据逐像素变化率；其中，变化率计算表达式为，
Figure PCTCN2022139124-appb-000006
式中，ΔL ke(i,j)和ΔL ok(i,j)分别是两个周期[t k,t e]和[t 0,t k]中Landsat遥感反射率的差值；L e(i,j)是结束日期te的Landsat遥感反射率；L k(i,j)是目标日期tk的Landsat遥感反射率；L 0(i,j)是开始日期t0.的Landsat遥感反射率；下标0是开始日期，k是目标日期；e是结束日期；(i，j)代表像素第i行j列；
利用生成的地表同质单元对生成的高空间分辨率地表温度数据逐像素变化率进行均值计算，得到高空间分辨率地表温度数据逐类别变化率；
构建第一训练集；其中，所述第一训练集的输入特征为低空间分辨率地表温度数据逐像素变化率、采样后每个像素对应于原始像素的ID值和地表同质单元，输出标签为高空间分辨率地表温度数据逐类别变化率；
构建第二维码训练集；其中，所述第二训练集的输入特征为目标时刻前一时刻与目标时刻后一时刻高空间分辨率地表温度数据差值的绝对值和高空间分辨率地表温度数据逐类别变化率，输出标签为目标时刻的地表温度。
本发明第二方面提供的一种地表温度遥感产品降尺度系统，包括：
数据获取模块，用于获取目标时刻T的T 前、T 后的原始低空间分辨率地表温度数据和原始高空间分辨率地表温度数据；其中，T 前为目标时刻T之前的任意有数据时刻，T 后为目标时刻T之后的任意有数据时刻；
数据处理模块，用于将T 前、T 后的原始低空间分辨率地表温度数据重采样到与原始高空间分辨率地表温度数据一致的空间分辨率，获得T 前、T 后的采样后低空间分辨率地表温度数据以及采样后每个像素对应于原始像素的ID值；
变化率获取模块，基于T 前、T 后的采样后的低空间分辨率地表温度数据，按照T 前、T、T 后三个时间点计算逐像素的变化率，获得低空间分辨率地表温度数据逐像素变化率；
预测模块，用于将低空间分辨率地表温度数据逐像素变化率、采样后每个像素对应于原始像素的ID值以及预获取的地表同质单元ID作为输入，利用预训练好的动态多层感知网络DyNet预测获得高空间分辨率地表温度数据逐类别变化率；将T 前、T 后的原始高空间分辨率地表温度数据的差值的绝对值和获得的高空间分辨率地表温度数据逐类别变化率作为输入，利用预训练好的权重感知网络WNet预测获得目标时刻T的地表温度。
本发明第三方面提供的一种电子设备，包括：
至少一个处理器；以及与所述至少一个处理器通信连接的存储器；其中，所述存储器存储有可被所述至少一个处理器执行的指令，所述指令被所述至少一个处理器执行，以使所述至少一个处理器能够执行如本发明任一项上述的地表温度遥感产品降尺度方法。
本发明第四方面提供的一种计算机可读存储介质，存储有计算机程序，其特征在于，所述计算机程序被处理器执行时实现本发明任一项上述的地表温度遥感产品降尺度方法。
与现有技术相比，本发明具有以下有益效果：
本发明利用多层感知网络的非线性拟合能力强以及在变化率为-1区间中的预测稳定度高的特点，提供了一种新的地表温度遥感产品降尺度方法；本发明技术方案通过动态多层感知网络来构建非线性解混模型，可实现低分辨率卫星像元到高分辨率卫星像元的动态解混；本发明技术方案通过权重感知网络，可有效避免现有权重函数在变化率-1附近的失效问题。综上，本发明提供的地表温度遥感产品降尺度方法，可对地表温度进行有效的降尺度估计，能够提高地表温度产品精度。
附图说明
为了更清楚地说明本发明实施例或现有技术中的技术方案，下面对实施例或现有技术描述中所需要使用的附图做简单的介绍；显而易见地，下面描述中的附图是本发明的一些实施例， 对于本领域普通技术人员来说，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1是本发明实施例提供的一种地表温度遥感产品降尺度方法的流程示意图；
图2是本发明实施例中，获取训练样本数据集的流程示意图；
图3是本发明实施例中，训练动态多层感知网络(DyNet)的流程示意图；
图4是本发明实施例中，训练权重感知网络(WNet)的流程示意图；
图5是本发明实施例中，预测目标日期数据的流程示意图；
图6是本发明实施例中，数据后处理的流程示意图。
具体实施方式
为了使本技术领域的人员更好地理解本发明方案，下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本发明一部分的实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都应当属于本发明保护的范围。
需要说明的是，本发明的说明书和权利要求书及上述附图中的术语“第一”、“第二”等是用于区别类似的对象，而不必用于描述特定的顺序或先后次序。应该理解这样使用的数据在适当情况下可以互换，以便这里描述的本发明的实施例能够以除了在这里图示或描述的那些以外的顺序实施。此外，术语“包括”和“具有”以及他们的任何变形，意图在于覆盖不排他的包含，例如，包含了一系列步骤或单元的过程、方法、系统、产品或设备不必限于清楚地列出的那些步骤或单元，而是可包括没有清楚地列出的或对于这些过程、方法、产品或设备固有的其它步骤或单元。
下面结合附图对本发明做进一步详细描述：
请参阅图1，本发明实施例提供的一种地表温度遥感产品降尺度方法，包括以下步骤：
步骤S1，获取目标时刻T的T 前、T 后的原始低空间分辨率地表温度数据和原始高空间分辨率地表温度数据；其中，T 前为目标时刻T的前一时刻，T 后为目标时刻T的后一时刻；
步骤S2，将T 前、T 后的原始低空间分辨率地表温度数据重采样到与原始高空间分辨率地表温度数据一致的空间分辨率，获得T 前、T 后的采样后低空间分辨率地表温度数据以及采样后每个像素对应于原始像素的ID值；
步骤S3，基于T 前、T 后的采样后低空间分辨率地表温度数据，按照T 前、T、T 后三个时间点计算逐像素的变化率，获得低空间分辨率地表温度数据逐像素变化率；
步骤S4，将低空间分辨率地表温度数据逐像素变化率、采样后每个像素对应于原始像素的ID值以及预获取的地表同质单元作为输入，利用预训练好的动态多层感知网络预测获得高空间分辨率地表温度数据逐类别变化率；将T 前、T 后的原始高空间分辨率地表温度数据的差值的绝对值和预测获得的高空间分辨率地表温度数据逐类别变化率作为输入，利用预训练好的权重感知网络预测获得目标时刻T的地表温度。
本发明实施例提供的方法，基于预训练好的动态多层感知网络和权重感知网络，可对地表温度进行有效的降尺度估计，将原始的逐日低空间分辨率数据降尺度为逐日高空间分辨率数据，从而提高地表温度产品精度，服务于地表温度高精度、高频次卫星遥感监测。
本发明实施例提供的技术方案中，获取预训练好的动态多层感知网络和权重感知网络的步骤可概述为：
步骤1，数据预处理，包括：对原始的低空间分辨率地表温度数据和高空间分辨率地表温度数据进行必要的空间配准、同质单元划分、变化率计算和像素索引计算等操作，为后续模型训练提供充分的训练数据；
步骤2，模型训练，包括：利用步骤1获得的数据生成训练样本，并对多层感知网络(DyNet)和权重感知网络(WNet)进行训练。
本发明实施例示例性的，在保留步骤2中最优模型的基础上，针对目标日期，开展地表温度预测。
本发明实施例优选的，获取地表温度预测值后还包括数据后处理，包括：固定目标日期，在时间序列中选取多组观测数据对，对目标日期的地表温度进行重复预测，按照一定规则整合多次预测结果，合成最终数据产品。
请参阅图2，本发明实施例具体示例性的，获取用于训练动态多层感知网络和权重感知网络的训练样本数据的步骤具体包括：
步骤1)，对历史低空间分辨率影像(示例性的，例如MODIS 1km尺度地表温度产品)按照最邻近采样的原则，重采样到与高空间分辨率影像(示例性的，例如Landsat 30m尺度地表温度产品)一致的空间分辨率，形成采样后的低空间分辨率影像(可记作M30)；和采样后每个像素对应于原始MODIS像素的ID值(可记作MID)；
步骤2)，将历史高空间分辨率影像(示例性的，例如Landsat 30m尺度地表温度产品)做为多个特征，利用非监督聚类模型ISOData算法，聚类为多个地表同质单元，记作ClassID；
步骤3)，按照时间顺序，采样后的低空间分辨率影像和高空间分辨率影像按照每三个时间点(T前，T中，T后)计算逐像素的变化率，最终获得低空间分辨率影像逐像素变化率(可记作M_ratio)和高空间分辨率影像逐像素变化率(可记作L_ratio)；
其中，变化率计算表达式为，
Figure PCTCN2022139124-appb-000007
式中，ΔL ke(i,j)和ΔL ok(i,j)分别是两个周期[t k,t e]和[t 0,t k]中Landsat遥感反射率的差值；L e(i,j)是结束日期te的Landsat遥感反射率；L k(i,j)是目标日期tk的Landsat遥感反射率；L 0(i,j)是开始日期t0.的Landsat遥感反射率；下标0是开始日期，k是目标日期；e是结束日期；(i，j)代表像素第i行j列；
步骤4)，利用步骤2)生成的地表同质单元(ClassID)对步骤3)生成的高空间分辨率影像逐像素变化率进行均值计算，得到高空间分辨率逐类别变化率(记作L_ratio_C)；
步骤5)，构建动态多层感知网络(DyNet)训练数据集；数据集的输入为三通道特征影像，三个通道分别为低空间分辨率影像逐像素变化率(M_ratio)、采样后每个像素对应于原始MODIS像素的ID值(MID)和地表同质单元(ClassID)；数据集的输出为单通道特征影像，为高空间分辨率影像逐类别变化率(L_ratio_C)；
步骤6)，构建权重网络WNet训练数据集；数据集的输入为两通道特征影像，两个通道分别为T前时刻和T后时刻高分辨率影像的差值绝对值和高空间分辨率影像逐类别变化率；数据的输出为单通道特征影像，为目标时刻T中的真实地表温度(T中LST)。
请参阅图3，本发明实施例具体示例性的，训练动态多层感知网络的步骤具体包括：
所述动态多层感知网络中，设置有用于输入数据的暂退层，暂退层(DropOut)用于利用输入数据中的MID，只激活当前输入数据中存在的神经节点，屏蔽当前输入数据中没有的神经节点；在暂退层后连续通过五个全连接层(FC)和Relu激活函数层，具体示例性的，每个全连接层的神经元可为128个；在第五个全连接层(FC)和Relu激活函数层后，再设置一个暂退层，该暂退层中的神经元激活与否按照输入数据中地表同质单元ClassID来决定；
训练时，基于训练样本对于预测的逐类别变化率(P_ratio_C)和真值(L_ratio_C)计算均方根损失(MSE Loss)，反向传播，达到预设收敛条件后获得训练好的动态多层感知网络。
请参阅图4.本发明实施例具体示例性的，训练权重感知网络的步骤具体包括：
所述权重感知网络中，设置有数据映射层；在数据映射层后连续设置五个全连接层(FC)和Relu激活函数层，具体示例性的，每个全连接层的神经元可为40个；
训练时，输入参数通过数据映射层，其中数据映射依据下面公式进行映射：
对真值数据的变换为：ΔLST＝LST-LST per；
当T后时刻地表温度值减去T前地表温度的值小于0时，对高空间分辨率影像逐类别变化率(L_ratio_C)进行变换为：
Figure PCTCN2022139124-appb-000008
当T后时刻地表温度值减去T前地表温度的值大于0时，对高空间分辨率影像逐类别变化率(L_ratio_C)进行变换为：
Figure PCTCN2022139124-appb-000009
将模型输出换算为地表温度的预测值的表达式为：
Figure PCTCN2022139124-appb-000010
其中，对于预测的地表温度
Figure PCTCN2022139124-appb-000011
和目标日期的地表温度(真值)计算均方根损失(MSE Loss)，反向传播，达到预设收敛条件后获得训练好的权重感知网络。
本发明实施例中的动态多层感知网络，可由其他包含多个全连接层结构和暂退层结构的深度网络或深度卷积网络结构代替，本发明中的权重感知网络可由其他包含多个全连接层结构的深度网络或深度卷积网络结构代替。
请参阅图5，本发明实施例提供的方法的具体应用过程包括：在动态多层感知网络DyNet和权重感知网络WNet充分训练的基础上，利用前一时刻(T前)和后一时刻(T后)的高分辨率和低分辨率数影像数据，可开展对目标日期(T中时刻)数据的预测；具体流程包括以下步骤：
对当前影像对(T前-T中-T后)中的低空间分辨率影像按照最邻近采样的原则，重采样到与高空间分辨率影像一致的空间分辨率，形成采样后的低空间分辨率影像和采样后每个像素对应于原始MODIS像素的ID值；
计算当前影像对(T前-T中-T后)低空间分辨率影像逐像素的变化率；
构建当前影像对的动态多层感知网络(DyNet)数据集，数据集为三通道特征影像，三个通道分别为低空间分辨率影像逐像素变化率(M_ratio)、采样后每个像素对应于原始MODIS像素的ID值(MID)和地表同质单元(ClassID)；用训练好的动态多层感知网络(DyNet)进行同质单元变化率的预测，获得预测结果；
构建当前影像对的权重感知网络(WNet)数据集，数据集的输入为两通道特征影像，两个通道分别为T前时刻和T后时刻高分辨率影像(L30)的差值的绝对值和得到的同质单元变化率的预测值；利用训练好的权重感知网络(WNet)进行目标日期(T中时刻)地表温度的预测。
请参阅图6，本发明实施例优选的，还包括数据后处理，是用多组日期对影像对同一天进行预测，将多次预测值进行合并，达到降低预测误差的目的；具体流程包括：
选取同一目标日期下的多组日期对数据，预测生成多组预测结果；
依据地表温度的实际合理区间，对多组预测结果中的奇异值进行掩膜；
利用前后日期地表温度差异的大小，来设定阈值，掩膜掉差异较小的像素，预测可信度较低的像素；
对多组预测结果，取时间序列中的中值作为目标日期最终的高分辨率地表温度的预测结果。
综上所述，本发明实施例提供的技术方案通过设计全新的动态多层感知网络，来构建非线性解混模型，实现了低分辨率卫星像元(本发明以MODIS卫星为例)到高分辨率卫星像元(本发明以Landsat卫星为例)的动态解混；建立的动态多层感知网络可以有效解决现有线性解混模型无法有效估计地表温度的问题，从本质上避免了地表同质单元数量与可用求解方程数量之间相互矛盾所造成的无法求解的难题。本发明实施例通过设计全新的多层权重感知网络，来有效避免现有权重函数在变化率-1附近的失效问题。本发明建立的网络模型基于数学机理构建，在求解条件不变的前提下，可以实现跨区域，跨时间的模型泛化。
本发明实施例中，在深圳试验区进行模拟和实验，其中以国际通用的评价指标进行测定，对比了现有STARFM模型和ESTARFM模型，对比实验结果如表1所示。
表1.对比实验结果
Figure PCTCN2022139124-appb-000012
分析表1可知，在多个预测日期的实验组中，本发明所构建的系统，在定量指标峰值信噪比(PNSR)，结构相似度(SSIM)，相关系数(CC)，均方根误差(RMSE)和均值误差(MAE)都优于目前国际上通用的两组模型(STARFM和ESTARFM)。
下述为本发明的装置实施例，可以用于执行本发明方法实施例。对于装置实施例中未纰漏的细节，请参照本发明方法实施例。
本发明实施例提供的一种地表温度遥感产品降尺度系统，包括：
数据获取模块，用于获取目标时刻T的T 前、T 后的原始低空间分辨率地表温度数据和原始高空间分辨率地表温度数据；其中，T 前为目标时刻T的前一时刻，T 后为目标时刻T的后一时刻；
数据处理模块，用于将T 前、T 后的原始低空间分辨率地表温度数据重采样到与原始高空间 分辨率地表温度数据一致的空间分辨率，获得T 前、T 后的采样后低空间分辨率地表温度数据以及采样后每个像素对应于原始像素的ID值；
变化率获取模块，用于基于T 前、T 后的采样后低空间分辨率地表温度数据，按照T 前、T、T 后三个时间点计算逐像素的变化率，获得低空间分辨率地表温度数据逐像素变化率；
预测模块，用于将低空间分辨率地表温度数据逐像素变化率、采样后每个像素对应于原始像素的ID值以及预获取的地表同质单元作为输入，利用预训练好的动态多层感知网络预测获得高空间分辨率地表温度数据逐类别变化率；将T 前、T 后的原始高空间分辨率地表温度数据的差值的绝对值和预测获得高空间分辨率地表温度数据逐类别变化率作为输入，利用预训练好的权重感知网络预测获得目标时刻T的地表温度。
现有技术中，基于权重函数的方法是假定地表反射率的时间变化在空间尺度上的一致性，通过权重函数组合所有输入影像的信息来预测目标影像高空间分辨率的像素值。Gao等人最早提出时空自适应反射融合模型(Spatial and Temporal Adaptive Reflectance Fusion Model，STARFM)，该模型假定粗分辨率影像中的像素为纯净像元，引入移动窗口和相似像元的概念，从光谱差异性、时间差异性以及空间距离来计算权重；对于复杂地貌中土地类型发生变化的区域，STARFM方法无法准确预测出未知时相真实的地物类型，导致融合结果产生较大误差。针对STARFM方法不足，Hilker等提出了一种新的时空自适应反射率变化检测算法(Spatial Temporal Adaptive Algorithm for Mapping Reflectance Change，STAARCH)，使用缨帽变换检测反射率变化，提高从低分辨率影像上检测出土地覆盖时空变化的能力；但STAARCH中的变化检测仅适用于植被地表，因此Zhu等提出的增强型时空自适应反射融合模型(Enhanced Spatial and Temporal Adaptive Reference Fusion Model，ESTARFM)，引入转换系数和线性光谱解混理论，用光谱相关系数来代替空间距离，保留更多的空间细节的同时提高了在复杂地貌中的预测精度。针对ESTARFM方法预测有形状变化的地物覆被随时间变化的准确性，Zhao等 人提出了鲁棒自适应时空融合模型(Robust Adaptive Spatial and Temporal Fusion Model，RASTFM)，由基于非局部线性回归的加权平均模块和基于非局部线性回归的图像超分辨率模块组成，分别用于预测地表非形状变化(包括物候变化和没有形状变化的土地覆盖变化)和形状变化(具有形状变化的土地覆盖变化)。STARFM系列模型参数简单，计算效率高，但在地物覆盖复杂、不同时相地物类型变化的预测仍具有局限性。基于学习的方法通过机器学习算法用非线性的方式对高、低分辨率影像对建模，预测出高时空分辨率影像。Huang等提出基于稀疏表示的时空反射融合模型(SParse-Representation-Based SpatioTemporal Reflectance Fusion Model，SPSTFM)，利用字典对学习建立低空间分辨率影像和高空间分辨率影像的对应反射率变化关系，并通过时间加权预测目标高分辨率影像。针对字典对中扰动问题，Wu等提出误差约束正则化的半耦合字典学习(Error-Bound-Regularized Semi-Coupled Dictionary Learning，EBSCDL)，利用误差约束的正则化方法解决字典扰动，构建优化的半耦合字典解决低空间分辨率影像和高分辨率影像之间差异性，提高融合精度。然而上述基于稀疏表示的融合方法需要人为设计字典基元，在算法实现过程中，字典学习、稀疏编码和影像重建等步骤是分离的，提升算法的不稳定性和复杂性。对此，Song等提出深度卷积神经网络时空融合模型(Spatiotemporal Fusion Method Based on Deep Convolutional Neural Networks，STFDCNN)，结合卷积神经网络、非线性映射模型设计了一个双重卷积神经网络，实现了自动提取影像特征并提高预测精度，有效解决稀疏表示方法类的缺陷。基于学习的方法可以捕获到更多的地表空间细节，适用于异质区域等，但该类方法过度依赖训练样本和模型参数，具有较高的时间复杂性。基于解混的方法很好的弥补上述方法的不足，该类方法基于线性光谱混合理论，从高空间分辨率影像中提取地物类别和丰度来分解低空间分辨率的像元获得类别的光谱值。其实现过程一般包括端元选取、丰度计算、解混等内容。此类算法最早由Zhukov等提出，该方法假设同类地物反射率相等，在移动窗口内解混低空间分辨率影像反射率，然后将解混结果分配给未知 时相的高空间分辨率影像。但该方法解混过程中会存在严重误差，且算法中的假设对于现实中地物类型随时间变化区域是不成立的。对此，Zurita-Milla等在线性解混过程中引入约束条件，用来处理解混结果的负值和异常值问题；Wu等提出时空数据融合模型(Spatial Temporal Data Fusion Approach，STDFA)，基于各类地物覆盖时相变化总是相同的假设，引入时间变化信息，提高预测结果准确性；针对STDFA算法没有考虑到影像端元反射率的空间异质性，Zhang等基于STDFA算法，结合多尺度分割算法和ISODATA算法生成分类图，并利用移动窗口方法对低空间分辨率影像解混，最后引入时间权重概念预测未知时相影像；为了提升在地物类型变化区域模型预测准确度，Huang等提出基于解混的时空反射融合模型(Unmixing-based Spatio-Temporal reflectance Fusion Model，U-STFM)，该模型假设地物像元反射率在同质变化单元(homogeneous change regions，HCRs)上变化率相等，利用多尺度分割算法得到的超像素进行解混，从而提高了不同时相遥感影像在地表类型空间变化的融合效果。该类算法计算量小、可操作性强，提高了对于地物类型随时间变化的遥感影像融合精度，常用于反射率数据中。综上，对于时间变化波动性强、光谱特征不丰富的LST数据，使权重函数类方法的参数更加敏感，也影响学习类方法训练量的有效性，导致这俩种方法总体融合效果较低；而基于解混的方法对于时刻发生变化的LST，在LST变化趋势相同的区域上结合不同传感器变化趋势比相等的假设，提高了不同时相LST的融合效果。在此基础上，本发明实施例提供的技术方案通过动态多层感知网络来构建非线性解混模型，可实现低分辨率卫星像元到高分辨率卫星像元的动态解混；本发明实施例提供的技术方案通过权重感知网络，可有效避免现有权重函数在变化率-1附近的失效问题。
本发明再一个实施例中，提供一种计算机设备，该计算机设备包括处理器以及存储器，所述存储器用于存储计算机程序，所述计算机程序包括程序指令，所述处理器用于执行所述计算机存储介质存储的程序指令。处理器可能是中央处理单元(Central Processing Unit，CPU)，还 可以是其他通用处理器、数字信号处理器(Digital Signal Processor，DSP)、专用集成电路(Application Specific Integrated Circuit，ASIC)、现成可编程门阵列(Field-Programmable Gate Array，FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件等，其是终端的计算核心以及控制核心，其适于实现一条或一条以上指令，具体适于加载并执行计算机存储介质内一条或一条以上指令从而实现相应方法流程或相应功能；本发明实施例所述的处理器可以用于地表温度遥感产品降尺度方法的操作。
本发明再一个实施例中，本发明还提供了一种存储介质，具体为计算机可读存储介质(Memory)，所述计算机可读存储介质是计算机设备中的记忆设备，用于存放程序和数据。可以理解的是，此处的计算机可读存储介质既可以包括计算机设备中的内置存储介质，当然也可以包括计算机设备所支持的扩展存储介质。计算机可读存储介质提供存储空间，该存储空间存储了终端的操作系统。并且，在该存储空间中还存放了适于被处理器加载并执行的一条或一条以上的指令，这些指令可以是一个或一个以上的计算机程序(包括程序代码)。需要说明的是，此处的计算机可读存储介质可以是高速RAM存储器，也可以是非不稳定的存储器(non-volatile memory)，例如至少一个磁盘存储器。可由处理器加载并执行计算机可读存储介质中存放的一条或一条以上指令，以实现上述实施例中有关地表温度遥感产品降尺度方法的相应步骤。
本领域内的技术人员应明白，本申请的实施例可提供为方法、系统、或计算机程序产品。因此，本申请可采用完全硬件实施例、完全软件实施例、或结合软件和硬件方面的实施例的形式。而且，本申请可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。
本申请是参照根据本申请实施例的方法、设备(系统)、和计算机程序产品的流程图和/或方框图来描述的。应理解可由计算机程序指令实现流程图和/或方框图中的每一流程和/或方框、以及流程图和/或方框图中的流程和/或方框的结合。可提供这些计算机程序指令到通 用计算机、专用计算机、嵌入式处理机或其他可编程数据处理设备的处理器以产生一个机器，使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的装置。
这些计算机程序指令也可存储在能引导计算机或其他可编程数据处理设备以特定方式工作的计算机可读存储器中，使得存储在该计算机可读存储器中的指令产生包括指令装置的制造品，该指令装置实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能。
这些计算机程序指令也可装载到计算机或其他可编程数据处理设备上，使得在计算机或其他可编程设备上执行一系列操作步骤以产生计算机实现的处理，从而在计算机或其他可编程设备上执行的指令提供用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的步骤。
最后应当说明的是：以上实施例仅用以说明本发明的技术方案而非对其限制，尽管参照上述实施例对本发明进行了详细的说明，所属领域的普通技术人员应当理解：依然可以对本发明的具体实施方式进行修改或者等同替换，而未脱离本发明精神和范围的任何修改或者等同替换，其均应涵盖在本发明的权利要求保护范围之内。

Claims (10)

  1. 一种地表温度遥感产品降尺度方法，其特征在于，包括以下步骤：
    获取目标时刻T的T 前、T 后的原始低空间分辨率地表温度数据和原始高空间分辨率地表温度数据；其中，T 前为目标时刻T之前的任意有数据时刻，T 后为目标时刻T之后的任意有数据时刻；
    将T 前、T 后的原始低空间分辨率地表温度数据重采样到与原始高空间分辨率地表温度数据一致的空间分辨率，获得T 前、T 后的采样后低空间分辨率地表温度数据以及采样后每个像素对应于原始像素的ID值；
    基于T 前、T 后的采样后的低空间分辨率地表温度数据，按照T 前、T、T 后三个时间点计算逐像素的变化率，获得低空间分辨率地表温度数据逐像素变化率；
    将低空间分辨率地表温度数据逐像素变化率、采样后每个像素对应于原始像素的ID值以及预获取的地表同质单元ID作为输入，利用预训练好的动态多层感知网络DyNet预测获得高空间分辨率地表温度数据逐类别变化率；将T 前、T 后的原始高空间分辨率地表温度数据的差值的绝对值和获得的高空间分辨率地表温度数据逐类别变化率作为输入，利用预训练好的权重感知网络WNet预测获得目标时刻T的地表温度。
  2. 根据权利要求1所述的一种地表温度遥感产品降尺度方法，其特征在于，所述预训练好的动态多层感知网络DyNet的获取步骤包括：
    获取第一训练样本集；其中，所述第一训练样本集中的每个训练样本均包括作为输入的低空间分辨率地表温度数据逐像素变化率、采样后每个像素对应于原始像素的ID值和地表同质单元ID，以及作为标签的高空间分辨率地表温度数据逐类别变化率；其中，所述高空间分辨率地表温度类别变化率是基于T 前、T、T 后的原始高空间分辨率地表温度数据，按照T 前、T、T 后三个时间点逐像素计算，并利用地表同质单元ID求取逐个类别的均值得到；地表同质单元ID由时间序列地表温度数据聚类或土地覆被类型数据获得；
    基于所述第一训练样本集对预构建的动态多层感知网络，采用MSE Loss损失函数进行训练，达到预设收敛条件后，获得所述预训练好的动态多层感知网络。
  3. 根据权利要求2所述的一种地表温度遥感产品降尺度方法，其特征在于，所述预构建的动态多层感知网络DyNet包括：用于屏蔽部分输入神经元的暂退层，设置于暂退层后串接的多个全连接层和Relu激活函数层，以及用于屏蔽部分输出神经元的暂退层。
  4. 根据权利要求2所述的一种地表温度遥感产品降尺度方法，其特征在于，所述预训练好的权重感知网络WNet的获取步骤包括：
    获取第二训练样本集；其中，所述第二训练样本集中的每个训练样本均包括作为输入的高空间分辨率地表温度数据T 前和T 后的差值的绝对值和所述第一训练样本集中所述高空间分辨率地表温度数据逐类别变化率，以及作为标签的目标日期的高空间分辨率地表温度；
    基于所述第二训练样本集对预构建的权重感知网络，采用MSE Loss损失函数进行训练，达到预设收敛条件后，获得所述预训练好的权重感知网络。
  5. 根据权利要求4所述的一种地表温度遥感产品降尺度方法，其特征在于，所述预构建的权重感知网络包括：用于输入数据的数据映射层，以及设置于数据映射层后串接的多个全连接层和Relu激活函数层。
  6. 根据权利要求4所述的一种地表温度遥感产品降尺度方法，其特征在于，所述预构建的权重感知网络WNet中，所述数据映射层进行数据映射的步骤包括：
    对标签真值数据进行变换，变换表达式为，ΔLST＝LST-LST 前；式中，LST为目标时刻T真实的地表温度LST，LST 前为T 前时刻的地表温度LST，ΔLST为温度差；
    当T 后时刻的地表温度值小于T 前时刻的地表温度值时，高空间分辨率影像逐类别变化率的变换表达式为，
    Figure PCTCN2022139124-appb-100001
    式中，
    Figure PCTCN2022139124-appb-100002
    为高空间分辨率影像逐类别变化率，
    Figure PCTCN2022139124-appb-100003
    为变换后的逐类别变化率；
    当T 后时刻的地表温度值大于T 前时刻的地表温度值时，高空间分辨率影像逐类别变化率的变换表达式为，
    Figure PCTCN2022139124-appb-100004
    权重感知网络输出的地表温度预测值的表达式为，
    Figure PCTCN2022139124-appb-100005
    为模型预测的地表温度LST。
  7. 根据权利要求4所述的一种地表温度遥感产品降尺度方法，其特征在于，获取第一训练样本集和获取第二训练样本集的步骤具体包括：
    获取样本原始的低空间分辨率地表温度数据和高空间分辨率地表温度数据；
    将低空间分辨率地表温度数据按照最邻近采样的原则，重采样到与高空间分辨率地表温度数据一致的空间分辨率，形成采样后低空间分辨率地表温度数据以及采样后每个像素对应于原始像素的ID值；
    将时间序列高空间分辨率地表温度数据作为多个特征，聚类生成地表同质单元；
    按照时间顺序，采样后低空间分辨率地表温度数据和高空间分辨率地表温度数据按照目标时刻前一时刻、目标时刻、目标时刻后一时刻三个时间点计算逐像素的变化率，获得低空间分辨率地表温度数据逐像素变化率和高空间分辨率地表温度数据逐像素变化率；其中，变化率计算表达式为，
    Figure PCTCN2022139124-appb-100006
    式中，ΔL ke(i,j)和ΔL ok(i,j)分别是两个周期[t k,t e]和[t 0,t k]中Landsat遥感反射率的差值；L e(i,j)是结束日期te的Landsat遥感反射率；L k(i,j)是目标日期tk的Landsat遥感反射率；L 0(i,j)是开始日期t0.的Landsat遥感反射率；下标0是开始日期，k是目标日期；e是结束日期；(i，j)代表像素第i行j列；
    利用生成的地表同质单元对生成的高空间分辨率地表温度数据逐像素变化率进行均值计算，得到高空间分辨率地表温度数据逐类别变化率；
    构建第一训练集；其中，所述第一训练集的输入特征为低空间分辨率地表温度数据逐像素 变化率、采样后每个像素对应于原始像素的ID值和地表同质单元，输出标签为高空间分辨率地表温度数据逐类别变化率；
    构建第二维码训练集；其中，所述第二训练集的输入特征为目标时刻前一时刻与目标时刻后一时刻高空间分辨率地表温度数据差值的绝对值和高空间分辨率地表温度数据逐类别变化率，输出标签为目标时刻的地表温度。
  8. 一种地表温度遥感产品降尺度系统，其特征在于，包括：
    数据获取模块，用于获取目标时刻T的T 前、T 后的原始低空间分辨率地表温度数据和原始高空间分辨率地表温度数据；其中，T 前为目标时刻T之前的任意有数据时刻，T 后为目标时刻T之后的任意有数据时刻；
    数据处理模块，用于将T 前、T 后的原始低空间分辨率地表温度数据重采样到与原始高空间分辨率地表温度数据一致的空间分辨率，获得T 前、T 后的采样后低空间分辨率地表温度数据以及采样后每个像素对应于原始像素的ID值；
    变化率获取模块，基于T 前、T 后的采样后的低空间分辨率地表温度数据，按照T 前、T、T 后三个时间点计算逐像素的变化率，获得低空间分辨率地表温度数据逐像素变化率；
    预测模块，用于将低空间分辨率地表温度数据逐像素变化率、采样后每个像素对应于原始像素的ID值以及预获取的地表同质单元ID作为输入，利用预训练好的动态多层感知网络DyNet预测获得高空间分辨率地表温度数据逐类别变化率；将T 前、T 后的原始高空间分辨率地表温度数据的差值的绝对值和获得的高空间分辨率地表温度数据逐类别变化率作为输入，利用预训练好的权重感知网络WNet预测获得目标时刻T的地表温度。
  9. 一种电子设备，其特征在于，包括：
    至少一个处理器；以及与所述至少一个处理器通信连接的存储器；其中，所述存储器存储有可被所述至少一个处理器执行的指令，所述指令被所述至少一个处理器执行，以使所述至少 一个处理器能够执行如权利要求1至7中任一项所述的地表温度遥感产品降尺度方法。
  10. 一种计算机可读存储介质，存储有计算机程序，其特征在于，所述计算机程序被处理器执行时实现权利要求1至7中任一项所述的地表温度遥感产品降尺度方法。
PCT/CN2022/139124 2022-12-02 2022-12-14 一种地表温度遥感产品降尺度方法、系统、设备及介质 WO2024113427A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202211543073.6A CN115909087A (zh) 2022-12-02 2022-12-02 一种地表温度遥感产品降尺度方法、系统、设备及介质
CN202211543073.6 2022-12-02

Publications (1)

Publication Number Publication Date
WO2024113427A1 true WO2024113427A1 (zh) 2024-06-06

Family

ID=86479252

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/139124 WO2024113427A1 (zh) 2022-12-02 2022-12-14 一种地表温度遥感产品降尺度方法、系统、设备及介质

Country Status (2)

Country Link
CN (1) CN115909087A (zh)
WO (1) WO2024113427A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118552336A (zh) * 2024-07-26 2024-08-27 武汉大学 基于遥感的农田生态系统时序净交换估算方法及系统

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109060133A (zh) * 2018-05-31 2018-12-21 北京师范大学 遥感地表温度降尺度算法
CN110866364A (zh) * 2019-11-20 2020-03-06 中国科学院遥感与数字地球研究所 基于机器学习的地表温度降尺度方法
CN113486000A (zh) * 2021-09-08 2021-10-08 中国测绘科学研究院 基于多源数据和深度学习的地表蒸散发数据降尺度方法
CN113779863A (zh) * 2021-08-02 2021-12-10 中国农业科学院农业资源与农业区划研究所 一种基于数据挖掘的地表温度降尺度方法
US20220043182A1 (en) * 2019-10-14 2022-02-10 Guangzhou Institute of Geography, Guangdong Academy of Science Spatial autocorrelation machine learning-based downscaling method and system of satellite precipitation data

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109060133A (zh) * 2018-05-31 2018-12-21 北京师范大学 遥感地表温度降尺度算法
US20220043182A1 (en) * 2019-10-14 2022-02-10 Guangzhou Institute of Geography, Guangdong Academy of Science Spatial autocorrelation machine learning-based downscaling method and system of satellite precipitation data
CN110866364A (zh) * 2019-11-20 2020-03-06 中国科学院遥感与数字地球研究所 基于机器学习的地表温度降尺度方法
CN113779863A (zh) * 2021-08-02 2021-12-10 中国农业科学院农业资源与农业区划研究所 一种基于数据挖掘的地表温度降尺度方法
CN113486000A (zh) * 2021-09-08 2021-10-08 中国测绘科学研究院 基于多源数据和深度学习的地表蒸散发数据降尺度方法

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118552336A (zh) * 2024-07-26 2024-08-27 武汉大学 基于遥感的农田生态系统时序净交换估算方法及系统

Also Published As

Publication number Publication date
CN115909087A (zh) 2023-04-04

Similar Documents

Publication Publication Date Title
Chen et al. 2-D regional short-term wind speed forecast based on CNN-LSTM deep learning model
CN111612243B (zh) 交通速度预测方法、系统及存储介质
Pan et al. Short-term wind speed forecasting based on spatial-temporal graph transformer networks
CN113705880A (zh) 基于时空注意力图卷积网络的交通速度预测方法和装置
US11238307B1 (en) System for performing change detection within a 3D geospatial model based upon semantic change detection using deep learning and related methods
US11747468B2 (en) System using a priori terrain height data for interferometric synthetic aperture radar (IFSAR) phase disambiguation and related methods
CN109522831B (zh) 一种基于微卷积神经网络的车辆实时检测方法
WO2024113427A1 (zh) 一种地表温度遥感产品降尺度方法、系统、设备及介质
CN110689183A (zh) 一种集群光伏功率概率预测方法、系统、介质及电子设备
CN117474781A (zh) 一种基于注意力机制的高光谱与多光谱图像融合方法
Huang et al. A 3D ConvLSTM-CNN network based on multi-channel color extraction for ultra-short-term solar irradiance forecasting
CN116760031A (zh) 基于气象数据的高时间分辨率光伏功率预测方法和装置
Chen et al. Tempee: Temporal-spatial parallel transformer for radar echo extrapolation beyond auto-regression
Lin et al. Context-aware attentional graph U-Net for hyperspectral image classification
Zhang et al. Pqa-cnn: Towards perceptual quality assured single-image super-resolution in remote sensing
Sit et al. TempNet–temporal super-resolution of radar rainfall products with residual CNNs
Wang et al. Siamese spectral attention with channel consistency for hyperspectral image classification
CN117576483A (zh) 基于多尺度卷积自编码器的多源数据融合地物分类方法
Gendy et al. Lightweight image super-resolution network based on extended convolution mixer
Zhang et al. Multi-target Domain Adaptation Building Instance Extraction of Remote Sensing Imagery with Domain-common Approximation learning
CN114692964B (zh) 一种基于交叉注意力机制的时空交通流量预测方法
Luo et al. Hierarchical Neural Operator Transformer with Learnable Frequency-aware Loss Prior for Arbitrary-scale Super-resolution
Bosma et al. Estimating solar and wind power production using computer vision deep learning techniques on weather maps
Shan et al. A deep learning model for multi-modal spatio-temporal irradiance forecast
Li et al. A transformer-CNN parallel network for image guided depth completion

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22967012

Country of ref document: EP

Kind code of ref document: A1