Passive microwave remote sensing gap filling method, system, equipment and medium
By constructing a dual-model strategy and feature importance assessment, combined with micro-topography correction coefficients, the problems of data consistency and error propagation in passive microwave remote sensing orbit gap filling are solved, achieving high-precision orbit gap filling applicable to various underlying surface types.
Patent Information
- Application Number
- CN202510815002.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-18
- Publication Date
- 2025-11-21
AI Technical Summary
Existing passive microwave remote sensing methods suffer from poor data consistency, spatiotemporal differences, and error propagation issues in filling orbital gaps. Furthermore, they do not fully utilize real-time monitoring data from power grid icing observation stations, resulting in limited accuracy in estimating local ice distribution.
A dual-model strategy was constructed, using ice-covered observation data and auxiliary data. The first and second models were trained using a random forest regression algorithm. The optimal model was selected for missing value imputation by combining micro-topography correction coefficients and feature importance assessment, and the imputation accuracy was evaluated through multi-dimensional evaluation indicators.
The model's stability and applicability have been improved, the accuracy and robustness of track gap filling have been enhanced, it is applicable to various underlying surface types, errors and spatiotemporal differences have been reduced, and the adaptability and scalability of the method have been strengthened.
Smart Images
Figure CN120997534A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of passive microwave remote sensing technology, specifically to a method, system, device, and medium for filling gaps in passive microwave remote sensing. Background Technology
[0002] Passive microwave remote sensing of brightness and temperature data is obtained by receiving microwave signals from the Earth's surface (rather than actively emitting them), typically in the 1-90 GHz band. Compared to visible light and infrared remote sensing, it offers advantages such as all-weather operation, all-time observation, sensitivity to the Earth's surface, and strong penetration, making it irreplaceable in many fields, including Earth science, meteorology, hydrology, agriculture, and environmental monitoring. Currently, spaceborne passive microwave sensors are usually deployed in polar sun-synchronous orbits. Each satellite pass can only cover one orbital strip, especially in low-latitude regions where the spatial overlap between orbital strips is low, easily creating observation gaps. Secondly, the sensor scanning method and observation swath are limited. Passive microwave sensors generally use fan-shaped or conical scanning, and each scan can only observe a strip of ground of a certain width. The "blank band" formed between adjacent orbits is the orbital gap.
[0003] Currently, methods for imputing missing values in remote sensing data can be broadly categorized into three types: The first type uses data with the same name, which is data from a different source but with the same physical meaning to impute missing values in the target data. However, data from the same source suffers from poor consistency, spatiotemporal differences, error propagation, and complex processing, leading to a decrease in the accuracy of the imputed data. The second type uses auxiliary data for inference, typically establishing the relationship between auxiliary data and target data through machine learning or regression algorithms. Then, the missing values in the target data are inferred from the auxiliary data with higher data completeness, thus achieving the purpose of imputation. However, this method is limited by the quality of the auxiliary data and must consider the correlation between the auxiliary data and the target data, resulting in poor scalability. The third type uses direct interpolation of the target data. Interpolation methods are divided into three categories: temporal interpolation, spatial interpolation, and spatiotemporal interpolation. However, the accuracy of the interpolated missing values decreases when there are too many missing values in the target data. The aforementioned method combines elements of the second and third categories, but uses only the underlying surface as auxiliary data. This limits its applicability to specific underlying surfaces (snow cover, permafrost) and prevents its general applicability to all types of surfaces. Furthermore, due to orbital gaps, the annual pixel brightness temperature sequence is incomplete, requiring time interpolation to fill in missing values, which leads to error accumulation. In addition, the existing method does not fully utilize real-time monitoring data from power grid icing observation stations. Summary of the Invention
[0004] In view of the above-mentioned problems, the present invention is proposed.
[0005] Therefore, the technical problem solved by this invention is that the output method of a single model has poor accuracy in filling missing values.
[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a passive microwave remote sensing gap-filling method, which includes the following steps:
[0007] Acquire remote sensing data, icing observation data, and auxiliary data;
[0008] The remote sensing data, icing observation data, and auxiliary data are preprocessed, and icing features are constructed based on the icing observation data.
[0009] The first and second models were constructed using preprocessed remote sensing data, auxiliary data, and icing features.
[0010] The remote sensing data is evaluated pixel by pixel. If a remote sensing data with the same name exists at the corresponding location, the second model is used to fill in the missing value. If a remote sensing data with the same name does not exist at the corresponding location, the first model is used to fill in the missing value.
[0011] Multiple evaluation indicators are used to quantitatively evaluate the filling accuracy.
[0012] As a preferred embodiment of the passive microwave remote sensing gap-filling method described in this invention, the preprocessing step includes:
[0013] The remote sensing data, icing observation data, and auxiliary data are processed to achieve spatial scale unification.
[0014] As a preferred embodiment of the passive microwave remote sensing gap filling method of the present invention, the icing feature construction step includes: icing type feature construction and icing thickness feature construction;
[0015] The ice-covering type feature is constructed by generating multiple binary features through one-hot encoding, and then performing spatial scale unification processing after nearest neighbor interpolation gridding.
[0016] The ice thickness feature is constructed by using inverse distance weighted interpolation gridding for spatial scale unification and then enhancing it by combining micro-topography correction coefficients.
[0017] The advantages of this preferred technical solution are: unique thermal coding avoids false sorting relationships in the numerical bands of icing type; the enhanced micro-topography correction coefficient makes the icing thickness characteristics more consistent with the actual physical distribution, improving the accuracy of local area filling.
[0018] As a preferred embodiment of the passive microwave remote sensing gap filling method described in this invention, the micro-topography correction coefficient is set to different numerical ranges according to different terrain types;
[0019] The ice thickness feature is calculated by multiplying the original ice thickness by the corresponding micro-topography correction coefficient.
[0020] As a preferred embodiment of the passive microwave remote sensing gap-filling method of the present invention, the auxiliary data selects a preset number of key features from all auxiliary data features through feature importance evaluation;
[0021] The missing value is filled by determining the missing area pixel by pixel, and selecting either the first model or the second model for prediction and filling based on the existence of remote sensing data with the same name.
[0022] The beneficial effects of this preferred technical solution are: feature importance assessment selects the key features with the highest correlation, improving model training efficiency; the dual-model strategy makes full use of available data and avoids imputation failure caused by missing data with the same name.
[0023] As a preferred embodiment of the passive microwave remote sensing gap filling method described in this invention, both the first model and the second model are trained using the random forest regression algorithm.
[0024] The random forest model consists of multiple decision trees, and the prediction output is the average of the outputs of the multiple trees;
[0025] The input to the first model is the filtered auxiliary data and icing features, while the input to the second model is the filtered auxiliary data, icing features, and remote sensing data with the same name.
[0026] The beneficial effects of this preferred technical solution are: the random forest with multiple decision trees provides more stable prediction results; the second model introduces data with the same name as additional features, which significantly improves the accuracy of filling in regions with data with the same name.
[0027] As a preferred embodiment of the passive microwave remote sensing gap-filling method described in this invention, the various evaluation indicators include, but are not limited to, coefficient of determination, mean absolute error, mean square error, and root mean square error.
[0028] The accuracy of the fill is assessed by calculating the difference between the actual observed values and the predicted values.
[0029] The beneficial effects of this preferred technical solution are: multi-dimensional evaluation indicators comprehensively reflect the filling effect, and the assessment of correlation, MAE and RMSE reflects the magnitude of absolute error, providing a quantitative basis for method optimization.
[0030] Another objective of this invention is to provide a system for filling gaps in passive microwave remote sensing.
[0031] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a passive microwave remote sensing gap filling system, comprising a data acquisition module, a data preprocessing module, a model building module, and a pixel judgment and filling module;
[0032] The data acquisition module is used to acquire remote sensing data, icing observation data, and auxiliary data;
[0033] The data preprocessing module is responsible for preprocessing the acquired data;
[0034] The model building module uses preprocessed remote sensing data, icing observation data, auxiliary data, and icing characteristics to construct the first and second models;
[0035] The pixel judgment and filling module performs pixel-by-pixel judgment on the remote sensing data and selects to use the first model or the second model to fill in the missing values based on the judgment results.
[0036] The present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the passive microwave remote sensing gap filling method.
[0037] The present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements the steps of the passive microwave remote sensing gap-filling method.
[0038] The beneficial effects of this invention are as follows: By fusing data from power grid icing observation stations and passive microwave remote sensing data, a dual-model strategy gap-filling method is constructed, effectively solving the problems of poor data consistency, spatiotemporal differences, and error propagation inherent in traditional interpolation methods for track gap filling. This method uses underlying surface information as feature input, making the model applicable to various underlying surface types and overcoming the limitation of existing methods that are only applicable to specific underlying surfaces. By evaluating the importance of features and selecting key auxiliary data, the scalability and prediction accuracy of the model are improved.
[0039] This invention, in accordance with the national standard GB / T 35706—2017 for processing icing observation data, improves the accuracy of data filling in complex terrain conditions such as mountain passes and watersheds by introducing inverse distance weighted interpolation and micro-topography correction coefficients. The dual-model architecture intelligently selects the optimal prediction model based on the availability of data with the same name, avoiding the prediction failure problem of a single model when data is missing in some areas, thus enhancing the robustness and adaptability of the method.
[0040] By unifying multi-source data to the same spatial resolution, the technical challenge of spatial scale mismatch between observation station data and remote sensing data was solved, achieving effective fusion of different data sources. The nonlinear modeling capability of the random forest regression algorithm can better capture complex surface-atmosphere interactions, exhibiting significant advantages over traditional linear interpolation methods in terms of spatiotemporal continuity and accuracy of ground cover response. Attached Figure Description
[0041] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0042] Figure 1 The above is a general flowchart of a passive microwave remote sensing gap-filling method provided in one embodiment of the present invention.
[0043] Figure 2 This is a structural diagram of a passive microwave remote sensing gap-filling method provided in one embodiment of the present invention.
[0044] Figure 3 This is an overall operation diagram of a passive microwave remote sensing gap-filling method provided in one embodiment of the present invention. Detailed Implementation
[0045] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.
[0046] Example 1, referring to Figures 1-3 As an embodiment of the present invention, a passive microwave remote sensing gap-filling method is provided, comprising the following steps S1 to S5:
[0047] S1. Acquire remote sensing data, icing observation data, and auxiliary data;
[0048] S2. Preprocess the remote sensing data, icing observation data and auxiliary data, and construct icing characteristics based on the icing observation data;
[0049] S3. Construct the first and second models using preprocessed remote sensing data, auxiliary data, and icing characteristics;
[0050] S4. Perform pixel-by-pixel judgment on the remote sensing data. If there is a remote sensing data with the same name at the corresponding location, use the second model to fill in the missing value. If there is no remote sensing data with the same name at the corresponding location, use the first model to fill in the missing value.
[0051] S5. Use multiple evaluation indicators to quantitatively evaluate the filling accuracy.
[0052] It should be noted that spaceborne passive microwave sensors are typically deployed in polar sun-synchronous orbits. Each satellite transit can only cover one orbital strip, especially in low-latitude regions where the spatial overlap between orbital strips is low, easily creating observation gaps. The sensor's scanning method and observation swath are limited; passive microwave sensors generally use fan-shaped or conical scanning, and each scan can only observe a strip-shaped area of a certain width on the ground. The "blank band" formed between adjacent orbits is the orbital gap. Existing methods for filling missing remote sensing values suffer from poor data consistency, spatiotemporal differences, and error propagation, and do not fully utilize real-time monitoring data from power grid icing observation stations, resulting in limited accuracy in estimating local ice distribution.
[0053] Therefore, to address the aforementioned issue of filling orbital gaps, steps S1-S5 are employed to fuse data from power grid icing observation stations and passive microwave remote sensing data, strictly adhering to the GB / T35706—2017 standard for data processing. A dual-model strategy using both the first and second models effectively resolves the problem of missing data with the same name in some areas, improving the model's stability and applicability. Combined with micro-topography correction coefficients, the RMSE in areas such as mountain passes and watersheds is reduced to below 15%. Multi-source data fusion, incorporating icing observation data from the national standard GB / T35706—2017, fills in missing information in local areas. A unified resolution of 6.25km grid interpolation reduces spatial scale differences and improves the model's generalization ability. Using four southern provinces of China as experimental areas, experiments demonstrate that the F17 brightness temperature data filled by the random forest model exhibits good fidelity and detail preservation in missing areas, achieving high-precision filling of passive microwave remote sensing orbital gaps.
[0054] Example 2, refer to Figures 1-3 As an embodiment of the present invention, a passive microwave remote sensing gap filling method is provided based on the above embodiment.
[0055] In this embodiment of the application, in step S1, remote sensing data, icing observation data, and auxiliary data are acquired;
[0056] In this embodiment, when acquiring remote sensing data, F17_SSMIS passive microwave brightness temperature data is used as the target data, with the 37GHz vertical polarization (37V) band selected, covering the period from January 2021 to December 2022, and an original spatial resolution of 6.25km. This data suffers from orbital gaps due to the low spatial overlap between satellite orbital strips.
[0057] When acquiring icing observation data, daily monitoring data from power grid icing observation stations should be collected in accordance with the specifications in Appendices C and D of the national standard GB / T35706—2017, including but not limited to rime, hoarfrost, and mixed rime, and coded according to Appendix C of GB / T35706—2017; measured data in mm; latitude and longitude coordinates and altitude of the observation point; micro-topographic information including but not limited to windward slopes, leeward slopes, mountain passes, watersheds, and micro-topographic categories such as increased water vapor.
[0058] When acquiring auxiliary data, it includes, but is not limited to, ERA5 meteorological data, MODIS products such as Normalized Difference Vegetation Index (NDVI), Land Surface Temperature (LST), SMAP soil moisture data, DEM elevation data, and land use type data.
[0059] In this embodiment of the application, step S2, the preprocessing step includes A1:
[0060] Spatial scale unification processing is performed on remote sensing data, icing observation data, and auxiliary data;
[0061] Specifically, the spatial scale unification process is represented as follows: all data are resampled to a 6.25km resolution using bilinear interpolation or inverse distance-weighted interpolation to ensure a spatial registration error ≤ 0.1 pixels. Specifically, the target remote sensing data F17_SSMIS brightness temperature data has an original resolution of 6.25km, which serves as the baseline resolution; the corresponding remote sensing data FY-3DMWRI is resampled to a 6.25km resolution using bilinear interpolation; auxiliary data, including ERA5 meteorological data, MODIS products, and SMAP soil moisture data, are all unified to a 6.25km resolution using bilinear interpolation; and underlying surface classification data is resampled to the target resolution using nearest neighbor interpolation.
[0062] It should be noted that the inverse distance-weighted interpolation method was chosen because it can effectively handle the problem of uneven spatial distribution of data from icing observation stations. By using distance weights, it ensures that nearby observation stations have higher influence weights, while maintaining the spatial continuity of the interpolation results.
[0063] In an optional implementation, in step S2, the spatial scale unification processing can also employ the Kriging interpolation method, which constructs a variogram by analyzing spatial autocorrelation, thereby achieving the optimal unbiased estimation of icing observation data, and is particularly suitable for areas with sparse observation stations.
[0064] In another optional implementation, step S2 may also include time registration processing, which unifies the time scale of all data to daily values, filters time-matching dates, removes invalid data, and ensures the consistency of multi-source data in the time dimension.
[0065] In addition, step S2 also includes the construction of icing features, including the construction of icing type features and the construction of icing thickness features;
[0066] Ice-covering type features are constructed by generating multi-class binary features through one-hot encoding;
[0067] Icing type feature construction: Three types of binary features are generated through one-hot encoding: rain frost is encoded as 100, haze frost as 010, and mixed frost as 001, to realize the numerical representation of icing type.
[0068] Ice thickness feature construction: Missing values are filled using inverse distance weighted interpolation and enhanced by combining micro-topography correction coefficients. The formula is as follows:
[0069] B grid =B×T;
[0070] In the formula, B grid B is the corrected ice thickness; T is the original ice thickness; and T is the micro-topography correction factor.
[0071] It is important to know that the micro-topography correction coefficient is set with different numerical ranges according to different terrain types; the ice thickness characteristic is corrected by multiplying the original ice thickness by the corresponding micro-topography correction coefficient. The correction coefficient is 1.0 for general terrain, 2.0 for mountain pass micro-topography, 1.5 for high mountain watershed micro-topography, 1.2 for water vapor increase micro-topography, 1.2 for windward slope, and 0.5 to 0.8 for leeward slope.
[0072] In this embodiment, the auxiliary data selects a preset number of key features from all auxiliary data features through feature importance assessment; missing values are filled by judging the missing area pixel by pixel, and selecting a first model or a second model for prediction and filling based on the existence of remote sensing data with the same name.
[0073] Secondly, both the first and second models are trained using the random forest regression algorithm;
[0074] The random forest model consists of multiple decision trees, and the predicted output is the average of the outputs of multiple trees. The input of the first model is the filtered auxiliary data and icing features, and the input of the second model is the filtered auxiliary data, icing features, and remote sensing data with the same name.
[0075] Specifically, the feature importance assessment uses a random forest feature importance algorithm to select the top 10 key features from the auxiliary data. Feature importance is obtained by accumulating the error reduction during tree splits at nodes:
[0076] I j =∑ t∈split(j) ΔMSE·f(split=j);
[0077] In the formula, I j ΔMSE is the importance score of the j-th feature; ΔMSE is the reduction in mean squared error when a node is split; f(split=j) is the frequency of splitting using feature j; split(j) represents the set of all nodes split using feature j.
[0078] For example, the feature selection process can be represented as follows: From all auxiliary data features such as ERA5 meteorological data (2-meter air temperature, relative humidity, precipitation), MODIS products (NDVI, LST surface temperature), SMAP soil moisture, DEM elevation data, and land use types, calculate the importance score of each feature, sort them in descending order, and select the top 10 features. j The feature with the largest value is used to construct the set of input variables.
[0079] Specifically, the pixel-by-pixel judgment process for filling in missing values includes the following steps B1-B3:
[0080] B1. Perform pixel-by-pixel scanning of the target remote sensing data to identify missing areas with orbital gaps;
[0081] B2. For each missing pixel, determine whether there is FY-3D remote sensing data with the same name at the corresponding spatial location;
[0082] B3. Select the appropriate model for prediction and filling based on the existence of data with the same name.
[0083] Specifically, the model selection and prediction imputation in step B3 are represented as follows:
[0084] If remote sensing data with the same name exists at the corresponding location, the second model is used to fill in the gap:
[0085] Y missing =f B (X aux ,X surf ,X ice_type ,X ice_thick ,XFY );
[0086] If no remote sensing data with the same name exists at the corresponding location, the first model is used to fill the gap:
[0087] Y missing =f A (X aux ,X surf ,X ice_type ,X ice_thick );
[0088] In the formula, Y missing For the predicted brightness temperature value of the missing pixels; f A f B Let X represent the prediction functions of the first model and the second model, respectively; aux X represents the features of the filtered auxiliary data; surf Indicates the characteristics of the underlying surface; X ice_type Indicates the characteristics of icing type; X ice_thick Indicates the characteristics of ice thickness; X FY This refers to the brightness temperature data of the same name, FY-3D.
[0089] It should be noted that both models are trained using the random forest regression algorithm, and the model's predicted output is the average of the outputs of multiple decision trees. Let X be the input feature for each sample point. i The corresponding brightness temperature value is Y i The training objective is to minimize the following squared loss function:
[0090]
[0091] In the formula, T m Let M represent the m-th regression tree; M is the total number of trees; and N is the total number of samples.
[0092] The training parameters are random forest regression (number of trees = 200, depth = 15, minimum number of leaf samples = 5), the loss function is MSE, and 5-fold cross-validation is used for parameter optimization.
[0093] In an alternative implementation, feature importance assessment can also employ a mutual information method, which calculates the mutual information value between each auxiliary data feature and the target brightness temperature data, and selects the feature with the largest information gain. This method is suitable for datasets with strong nonlinear relationships.
[0094] In another alternative implementation, missing values can be filled using a weighted fusion strategy. When the quality of data with the same name is poor, the weights of the first and second models are dynamically adjusted, and the final prediction result is obtained by weighted averaging.
[0095] Y missing =
[0096] α·f A (X aux ,X surf ,X ice_type ,X ice_thick )+(1-α)·
[0097] f B (X aux ,X surf ,X ice_type ,X ice_thick ,X FY );
[0098] In the formula, α is the weighting coefficient, which is dynamically determined based on the quality assessment of the data with the same name, and its value range is [0,1].
[0099] Therefore, the dual-model strategy of the first and second models effectively solved the problem of missing data with the same name in some areas, reduced the computational complexity through feature selection, improved the stability and applicability of the model, and achieved high-precision filling of the gaps in the passive microwave remote sensing orbit.
[0100] In the embodiments of this application, various evaluation indicators include, but are not limited to, coefficient of determination, mean absolute error, mean square error, and root mean square error; the filling accuracy is evaluated by calculating the difference between the actual observed value and the predicted value.
[0101] Among them, the coefficient of determination R 2 The mathematical expression is:
[0102]
[0103] The mathematical expression for the mean absolute error (MAE) is:
[0104]
[0105] The mathematical expression for mean squared error (MSE) is:
[0106]
[0107] The mathematical expression for the root mean square error (RMSE) is:
[0108]
[0109] In the formula, y_i is the actual observed brightness temperature value of the i-th pixel. Let i be the predicted brightness temperature value of the i-th pixel. is the mean of the actual observations; n is the total number of validation samples.
[0110] For example, in this embodiment, the filling effect is evaluated using four southern provinces of China as the experimental area, and the global accuracy is evaluated using the coefficient of determination R. 2 =0.89, indicating that the model can explain 89% of the brightness temperature variability. The root mean square error (RMSE) is 12.3K, which is 28% lower than that of traditional interpolation methods. The mean absolute error (MAE) is 9.6K, and the mean square error (MSE) is 151.29K. 2 .
[0111] Accuracy evaluation of micro-topography areas (mountain pass area): Coefficient of determination R 2 =0.82, a 15% improvement compared to the traditional method; Root Mean Square Error (RMSE) = 9.8K, with the error controlled below 15%; Mean Absolute Error (MAE) = 7.4K; Mean Square Error (MSE) = 96.04K. 2 .
[0112] The spatiotemporal continuity error was reduced by 30%, and the number of anomalous jump points decreased by 40%.
[0113] It should be noted that the evaluation process employs a random sampling validation strategy, randomly selecting 20% of the pixels from the track gap region as the validation set, and the remaining 80% as the training set. The validation set includes samples from different terrain types, seasons, and icing conditions to ensure the representativeness of the evaluation results.
[0114] Specifically, the accuracy evaluation process includes the following steps C1 to C4:
[0115] C1. Construct a verification dataset by artificially creating gaps from the original complete brightness temperature data to simulate the real track gap situation;
[0116] C2. Apply a dual-model strategy to fill the artificial gaps and obtain the predicted brightness temperature value;
[0117] C3. Compare the predicted values with the original true values and calculate various accuracy indicators;
[0118] C4. Accuracy indicators are statistically analyzed according to different terrain types, seasons, and icing conditions, and then classified and evaluated.
[0119] In one optional implementation, the accuracy calculation process for a specific pixel is as follows: Assuming the validation set contains 100 pixels, the actual observed value of the 50th pixel is 245.8K, and the predicted value is 243.2K. Then: the absolute error is expressed as |245.8-243.2|=2.6K, and the squared error is expressed as (245.8-243.2). 2 =6.76K 2 ;
[0120] Cumulative calculations for all 100 pixels: MAE = (2.6 + 1.8 + 3.4 + ... + 2.1) / 100 = 9.6K; MSE = (6.76 + 3.24 + 11.56 + ... + 4.41) / 100 = 151.29K 2 ;RMSE=√151.29=12.3K;
[0121] In one alternative implementation, accuracy evaluation can also employ cross-validation, dividing the dataset into K folds (typically K=5), using K-1 folds for training and 1 fold for validation each time, repeating this process K times and taking the average value to improve the stability of the evaluation results.
[0122] In another alternative implementation, supplementary indicators such as relative error and normalized root mean square error can also be introduced:
[0123] The relative error (RE) is expressed as:
[0124]
[0125] The normalized root mean square error (NRMSE) is expressed as:
[0126]
[0127] It should be noted that the above-mentioned multi-dimensional accuracy evaluation system can comprehensively quantify the performance of the filler method, R 2 The MAE and RMSE reflect the model fit, respectively, while the prediction accuracy is reflected by the comprehensive analysis of different indicators. This study verifies the effectiveness and superiority of the method of the present invention in filling gaps in passive microwave remote sensing.
[0128] In summary, by integrating data from power grid icing observation stations and passive microwave remote sensing data, a dual-model strategy gap-filling method was constructed, effectively solving the problems of poor data consistency, spatiotemporal differences, and error propagation inherent in traditional interpolation methods for track gap filling. This method uses underlying surface information as feature input, making the model applicable to various underlying surface types and overcoming the limitation of existing methods that are only applicable to specific underlying surfaces. By evaluating the importance of features and selecting key auxiliary data, the scalability and prediction accuracy of the model are improved.
[0129] This invention, in accordance with the national standard GB / T 35706—2017 for processing icing observation data, improves the accuracy of data filling in complex terrain conditions such as mountain passes and watersheds by introducing inverse distance weighted interpolation and micro-topography correction coefficients. The dual-model architecture intelligently selects the optimal prediction model based on the availability of data with the same name, avoiding the prediction failure problem of a single model when data is missing in some areas, thus enhancing the robustness and adaptability of the method.
[0130] By unifying multi-source data to the same spatial resolution, the technical challenge of spatial scale mismatch between observation station data and remote sensing data was solved, achieving effective fusion of different data sources. The nonlinear modeling capability of the random forest regression algorithm can better capture complex surface-atmosphere interactions, exhibiting significant advantages over traditional linear interpolation methods in terms of spatiotemporal continuity and accuracy of ground cover response.
[0131] Example 3 illustrates a schematic scheme for a passive microwave remote sensing gap filling method. It should be noted that the technical solution of this passive microwave remote sensing gap filling system belongs to the same concept as the technical solution of the aforementioned passive microwave remote sensing gap filling method. Details not described in detail in this embodiment can be found in the description of the technical solution of the aforementioned passive microwave remote sensing gap filling method.
[0132] This embodiment also provides a passive microwave remote sensing gap filling system, including a data acquisition module, a data preprocessing module, a model building module, and a pixel judgment and filling module;
[0133] The data acquisition module is used to acquire remote sensing data, icing observation data, and auxiliary data;
[0134] The data preprocessing module is responsible for preprocessing the acquired data;
[0135] The model building module uses preprocessed remote sensing data, icing observation data, auxiliary data, and icing characteristics to construct the first and second models;
[0136] The pixel judgment and filling module performs pixel-by-pixel judgment on the remote sensing data and selects to use the first model or the second model to fill in the missing values based on the judgment results.
[0137] This embodiment also provides an electronic device suitable for passive microwave remote sensing gap filling, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the passive microwave remote sensing gap filling method proposed in the above embodiment.
[0138] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the passive microwave remote sensing gap-filling method as proposed in the above embodiments.
[0139] The storage medium proposed in this embodiment and the passive microwave remote sensing gap filling method proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.
[0140] Based on the above description of the implementation methods, those skilled in the art will clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.
[0141] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
Claims
1. A passive microwave remote sensing gap-filling method, characterized in that, Includes the following steps: Acquire remote sensing data, icing observation data, and auxiliary data; The remote sensing data, icing observation data, and auxiliary data are preprocessed, and icing features are constructed based on the icing observation data. The first and second models were constructed using preprocessed remote sensing data, auxiliary data, and icing features. The remote sensing data is evaluated pixel by pixel. If a remote sensing data with the same name exists at the corresponding location, the second model is used to fill in the missing value. If a remote sensing data with the same name does not exist at the corresponding location, the first model is used to fill in the missing value. Multiple evaluation indicators are used to quantitatively evaluate the filling accuracy.
2. The passive microwave remote sensing gap-filling method as described in claim 1, characterized in that, The preprocessing steps include: The remote sensing data and auxiliary data are subjected to spatial scale unification processing.
3. The passive microwave remote sensing gap-filling method as described in claim 2, characterized in that, The ice accretion feature construction steps include: ice accretion type feature construction and ice accretion thickness feature construction; The ice-covering type feature is constructed by generating multiple binary features through one-hot encoding, and then performing spatial scale unification processing after nearest neighbor interpolation gridding. The ice thickness feature is constructed by using inverse distance weighted interpolation gridding for spatial scale unification and then enhancing it by combining micro-topography correction coefficients.
4. The passive microwave remote sensing gap-filling method as described in claim 3, characterized in that, The micro-topography correction coefficient is set with different numerical ranges according to different terrain types; The ice thickness feature is calculated by multiplying the original ice thickness by the corresponding micro-topography correction coefficient.
5. A passive microwave remote sensing gap-filling method as described in claim 4, characterized in that, The auxiliary data is used to select a predetermined number of key features from all auxiliary data features through feature importance assessment; The missing value is filled by determining the missing area pixel by pixel, and selecting either the first model or the second model for prediction and filling based on the existence of remote sensing data with the same name.
6. The passive microwave remote sensing gap-filling method as described in claim 5, characterized in that, Both the first and second models are trained using the random forest regression algorithm; The random forest model consists of multiple decision trees, and the prediction output is the average of the outputs of the multiple trees; The input to the first model is the filtered auxiliary data and icing features, while the input to the second model is the filtered auxiliary data, icing features, and remote sensing data with the same name.
7. A passive microwave remote sensing gap-filling method as described in claim 6, characterized in that, The various evaluation indicators include, but are not limited to, coefficient of determination, mean absolute error, mean square error, and root mean square error; The accuracy of the fill is assessed by calculating the difference between the actual observed values and the predicted values.
8. A system for filling gaps in passive microwave remote sensing, employing a passive microwave remote sensing gap-filling method as described in any one of claims 1 to 7, characterized in that, It includes a data acquisition module, a data preprocessing module, a model building module, and a pixel judgment and filling module; The data acquisition module is used to acquire remote sensing data, icing observation data, and auxiliary data; The data preprocessing module is responsible for preprocessing the acquired data; The model building module uses preprocessed remote sensing data, icing observation data, auxiliary data, and icing characteristics to construct the first and second models; The pixel judgment and filling module performs pixel-by-pixel judgment on the remote sensing data and selects to use the first model or the second model to fill in the missing values based on the judgment results.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of a passive microwave remote sensing gap-filling method according to any one of claims 1 to 7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the passive microwave remote sensing gap filling method according to any one of claims 1 to 7.