Remote sensing data quality automatic grading method and system

Abstract

The present application relates to remote sensing image processing technical field, especially related to a kind of remote sensing data quality automatic grading method and system, obtain the measurement of multiple quality evaluation indexes of target remote sensing data of visible light;According to the scene where target remote sensing data is located, determine the weight of each quality evaluation index;Set the target value corresponding to each quality evaluation index, and quality score is carried out to measurement value in combination with weight;According to quality score, grade division is carried out to target remote sensing data.The present application determines weight based on multidimensional scene characteristics, so that the evaluation model has environmental perception ability, cooperates with score, realizes quantitative synthesis of multidimensional index, and also retains the veto effect of single index deterioration on overall result.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing image processing technology, and in particular relates to an automatic classification method and system for remote sensing data quality. Background Technology

[0002] With the increasing number of remote sensing satellites and the diversification of imaging modes, the amount of remote sensing data acquired is exploding. How to quickly and accurately assess the quality of this data is a prerequisite for subsequent applications (such as mapping, environmental monitoring, and target identification). Currently, commonly used quality assessment indicators include modulation transfer function (MTF) and signal-to-noise ratio (SNR). However, existing assessment methods mostly use single indicators or weighted combinations with fixed weights, which cannot adapt to the dynamic changes in the importance of each indicator under different satellite platforms, imaging modes, and land cover types. For example, for high-resolution detailed survey modes, the weight of MTF should be higher than that of SNR; while for imaging under low-light conditions, SNR is more critical. Evaluation results with fixed weights often deviate from the actual situation and are difficult to directly use for on-orbit autonomous decision-making or large-scale automatic data screening.

[0003] Therefore, there is an urgent need for an automatic remote sensing data quality grading technology that can adaptively adjust weights according to the imaging scene. Although existing studies have attempted to fuse multiple indicators through weighted methods, these methods often rely on fixed weights or require manual intervention for adjustment, and cannot achieve real-time, autonomous weight configuration for dynamically changing imaging scenes (such as switching from urban to ocean observation) in environments with limited on-board resources. Summary of the Invention

[0004] In view of this, the present invention aims to provide a method and system for automatic classification of remote sensing data quality, which dynamically adjusts the weights of each evaluation index according to scene conditions and automatically classifies the data into different quality levels, providing accurate and efficient quantitative basis for on-orbit mission planning and ground data screening.

[0005] To achieve the above objectives, the technical solution created by this invention is implemented as follows: An automatic quality grading method for remote sensing data, characterized by comprising: S1: Measured values ​​of multiple quality evaluation indicators for target remote sensing data in visible light; S2: Determine the weight of each quality evaluation indicator in step S1 based on the scene where the target remote sensing data is located. S3: Set the target value for each quality evaluation indicator in step S1, and score the quality of the measured values ​​obtained in step S1 by combining the weights obtained in step S2. S4: Based on the quality score obtained in step S3, classify the target remote sensing data in step S1 into different grades.

[0006] Furthermore, the measured values ​​of multiple quality evaluation indicators in step S1 include at least the panchromatic MTF measurement value of the target remote sensing data, the multispectral MTF measurement value of the target remote sensing data, and the panchromatic SNR measurement value of the target remote sensing data.

[0007] Furthermore, in step S2: multiple sets of weights are pre-calibrated based on historical data or expert experience to obtain a scene-weight mapping library. By identifying the current scene (such as satellite platform, imaging mode, and ground cover type), the optimal weight combination of multiple quality evaluation indicators is output, that is, the optimal weight of each quality evaluation indicator. If the optimal weight combination cannot be found in the scene-weight mapping library based on the scene, the optimal weight of each quality evaluation indicator under the scene is updated in real time based on the application feedback of similar scenes in historical data (such as subsequent image classification accuracy).

[0008] Furthermore, in step S3: the first quality score of the panchromatic band MTF measurement is calculated using the following formula: ; in, The first quality score, MTF. pan This represents the panchromatic MTF measurement value, w pan Indicates the corresponding weight. This represents the corresponding target value; The sub-mass score corresponding to each band of the multispectral MTF measurement is calculated using the following formula: ; in, This represents the MTF measurement value for the i-th band. w represents the corresponding sub-quality score. mul Indicates the corresponding weight. This represents the corresponding target value; The second quality score of the multispectral MTF measurement is obtained using the following formula: ; Where N represents the total number of multispectral bands; The third quality score of the panchromatic SNR measurement is calculated using the following formula: ; in, SNR represents the third quality score. pan This represents the panchromatic SNR measurement value, w snr Indicates the corresponding weight. This represents the corresponding target value.

[0009] Furthermore, in step S4: the geometric mean of the quality scores obtained in step S3 is taken as the comprehensive score of the target remote sensing data; a threshold range for the comprehensive score corresponding to each level is set, and the level to which the target remote sensing data belongs is determined based on the comprehensive score falling within the threshold range.

[0010] Furthermore, the methods for obtaining the segmentation threshold include: S41: Define the lower bounds for excellent and good grades in the overall score, and determine the utility function as follows: U=α×P high (T best ) β×R discard (T good )-γ×C imbalance ; Where U represents the utility function, T best and T good P represents the lower limit of excellence and the lower limit of goodness, respectively. high R represents the average performance of data classified as superior in downstream tasks at the current threshold. discard C represents the proportion of data that is classified as poor and discarded. imbalance The terms represent the penalty for imbalance in data volume at each level, and α, β, and γ represent the preset hyperparameters. S42: Set initial values ​​for the superior and good lower limits, and apply positive and negative perturbations to either of the two lower limit thresholds, then substitute them into the utility function of step S41: ; ; in, and Let represent the utility functions after adding a positive perturbation and a negative perturbation respectively during the k-th update of the current lower threshold. Let T1 represent the current lower bound threshold after the k-th update, and T2 represent the lower bound threshold before the update. This represents the perturbation in the k-th update; S43: Based on the utility functions after applying the positive and negative perturbations in step S42, determine the update momentum term of the current lower limit threshold using the following formula: ; Among them, v k Let μ represent the update momentum term for the k-th update, and μ represent the momentum coefficient. S44: Based on the updated momentum term of the current lower threshold obtained in step S43, update the current lower threshold using the following formula: ; Where η represents the learning rate, which controls the step size of a single adjustment, and A and B represent the physical boundaries of the clip function, respectively. S45: If the difference between the current lower limit threshold after the update in step S44 and the current lower limit threshold before the update does not exceed the set change threshold, or the number of updates exceeds the preset upper limit, then stop updating and repeat the operation of steps S42 to S45 with the output lower limit threshold to update the lower limit threshold that has not been updated until the current optimal combination of superior and good lower limits is output.

[0011] Furthermore, the imbalance penalty term is as follows: ; Where, p j This represents the percentage of data at the j-th level in the total number of data points, and M represents the total number of levels.

[0012] An automatic remote sensing data quality grading system, according to the automatic remote sensing data quality grading method provided by the present invention, includes: a data acquisition module, which receives target remote sensing data from a sensor and extracts measured values ​​of multiple quality evaluation indicators from the target remote sensing data; an adaptive weight configuration module, which determines the weight of each quality evaluation indicator according to the scene in which the target remote sensing data is located; a comprehensive score calculation module, which calculates the geometric mean of all quality scores based on the set target value corresponding to each quality evaluation indicator and in combination with the weight, to obtain a comprehensive score of the target remote sensing data; and a quality level classification module, which classifies the target remote sensing data into levels according to the comprehensive score and can attach the level information to the image metadata.

[0013] Furthermore, the measured values ​​of multiple quality evaluation indicators in step S1 include at least the panchromatic MTF measurement value of the target remote sensing data, the multispectral MTF measurement value of the target remote sensing data, and the panchromatic SNR measurement value of the target remote sensing data.

[0014] Compared with the prior art, the present invention can achieve the following beneficial effects: This invention presents an automatic remote sensing data quality grading method and system. By dynamically adjusting the weights of MTF and SNR according to the imaging scene, the evaluation results better reflect the actual image quality requirements of different applications, avoiding the bias caused by fixed weights. Simultaneously, it uses geometric mean instead of arithmetic mean to calculate the comprehensive score, ensuring that severe degradation of any single indicator (such as MTF) leads to a significant decrease in the comprehensive score. This aligns with the physical law of the "weakest link effect" in remote sensing imaging quality, resulting in more stringent evaluation results that better reflect actual physical meaning. Furthermore, the invention employs a linear discount model and geometric mean, resulting in low computational complexity and suitability for real-time operation in onboard embedded environments. Finally, this invention further introduces an imbalance penalty term based on information entropy and combines it with a coordinate descent method to dynamically adjust the grading threshold. This solves the problem of "too few excellent grades" or "too many poor grades" caused by traditional fixed thresholds when data distribution is uneven. Additionally, by introducing a momentum smoothing mechanism, it avoids drastic threshold fluctuations caused by abnormal data from a single onboard scene. This ensures that the grading results reflect both the physical quality (MTF / SNR) of a single scene and the overall data utilization rate, making it particularly suitable for on-orbit autonomous decision-making in onboard embedded environments. Attached Figure Description

[0015] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments and descriptions of the invention are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 A schematic flowchart of the automatic remote sensing data quality classification method described in the embodiments of the present invention; Figure 2 A flowchart illustrating the automatic remote sensing data quality grading method described in the embodiments of the present invention; Figure 3 This is a schematic diagram of the scene-weight mapping library described in an embodiment of the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not constitute a limitation thereof. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0017] In the description of this invention, it should be understood that the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0018] The invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0019] like Figures 1 to 2 As shown in the embodiment of the present invention, the automatic remote sensing data quality grading method includes: S1: Measured values ​​of multiple quality evaluation indicators for target remote sensing data obtained in visible light.

[0020] In some embodiments, the measured values ​​of the multiple quality evaluation indicators in step S1 include at least the panchromatic MTF measurement value of the target remote sensing data, the multispectral MTF measurement value of the target remote sensing data, and the panchromatic SNR measurement value of the target remote sensing data. In this embodiment of the invention, the target remote sensing data is four-band multispectral data, corresponding to four bands of MTF measurement values.

[0021] S2: Determine the weight of each quality evaluation indicator in step S1 based on the scene where the target remote sensing data is located.

[0022] In some embodiments, multiple sets of weights are pre-calibrated based on historical data or expert experience to obtain a scene-weight mapping library. By identifying the current scene (such as satellite platform, imaging mode, and ground cover type), the optimal weight combination of multiple quality evaluation indicators is output, that is, the optimal weight of each quality evaluation indicator. If the optimal weight combination cannot be found in the scene-weight mapping library according to the scene, the best weight of each quality evaluation indicator in the scene is updated in real time based on the application feedback of similar scenes in historical data (such as subsequent image classification accuracy), thereby achieving fully autonomous adaptive weight configuration.

[0023] In this embodiment of the invention, the scene includes at least one of the following: satellite platform type (e.g., Gaofen series, Ziyuan series), imaging mode (e.g., panchromatic multispectral fusion mode, hyperspectral mode, night light mode), and land cover type (e.g., city, water body, vegetation). The scene-weight mapping library set up in this embodiment of the invention is presented in tabular form as follows: Figure 3 As shown.

[0024] In other embodiments, the weights can also be optimized online using machine learning methods. For example, historical images and their subsequent application effects (such as classification accuracy and target recognition confidence) can be collected as feedback to construct a regression model or reinforcement learning model that takes scene features (such as the complexity of ground textures and illumination angles) as input and the optimal weight combination as output. When the satellite is in orbit, real-time scene features can be extracted and input into the model to directly obtain adaptive weight coefficients without manual intervention to update the lookup table.

[0025] S3: Set the target value for each quality evaluation indicator in step S1, and score the measurement value obtained in step S1 by combining the weights obtained in step S2.

[0026] In some embodiments, in step S3: The first quality score of the panchromatic band MTF measurement is calculated using the following formula: ; in, The first quality score, MTF. pan This represents the panchromatic MTF measurement value, w pan Indicates the corresponding weight. This represents the corresponding target value; The sub-mass score corresponding to each band of the multispectral MTF measurement is calculated using the following formula: ; in, This represents the MTF measurement value for the i-th band. w represents the corresponding sub-quality score. mul Indicates the corresponding weight. This represents the corresponding target value; The second quality score of the multispectral MTF measurement is obtained using the following formula: ; Where N represents the total number of multispectral bands; The third quality score of the panchromatic SNR measurement is calculated using the following formula: ; in, SNR represents the third quality score. pan This represents the panchromatic SNR measurement value, w snr Indicates the corresponding weight. This represents the corresponding target value.

[0027] The physical meaning of the above three scoring formulas is as follows: based on the design target value, when the actual measured value is lower than the design target, the comprehensive score is discounted and penalized according to the importance (i.e., weight) of the indicator in the current scenario. The more important the indicator (the greater the weight), the greater the score loss due to performance degradation.

[0028] S4: Based on the quality score obtained in step S3, classify the target remote sensing data in step S1 into different grades.

[0029] In some embodiments, the geometric mean of the quality scores obtained in step S3 is taken as the comprehensive score of the target remote sensing data, as shown in the following formula: ; Where Q represents the overall score, it can be seen that if the value of any single quality score is 0 (that is, the measurement value of the indicator is 0), then the overall score Q=0; Set a threshold range for the comprehensive score corresponding to each level, and determine the level of the target remote sensing data based on whether the comprehensive score falls within the threshold range.

[0030] In some embodiments, the method for obtaining the segmentation threshold includes: S41: Define the lower bounds for excellent and good grades in the overall score, and determine the utility function as follows: U=α×P high (T best ) β×R discard (T good )-γ×C imbalance ; Where U represents the utility function, T best and T good P represents the lower limit of excellence and the lower limit of goodness, respectively. high R represents the average performance (such as classification accuracy and signal-to-noise ratio margin) of data classified as superior in downstream tasks under the current threshold. discard This indicates the proportion of data that is classified as poor and discarded. This is a penalty term to prevent overly strict thresholds from causing data waste. (C) imbalance This represents the penalty for data imbalance at each level. α, β, and γ are preset hyperparameters that can be set by the user according to current application preferences (e.g., disaster monitoring prioritizes data coverage, so α is lowered and γ is increased). Further explanation: the imbalance penalty term C... imbalance The physical meaning is that the penalty concentration is such that when the imbalance penalty term C imbalance As entropy increases, negative entropy increases as entropy decreases. This leads to a decrease in the utility function, preventing all data from being grouped into the same level and avoiding a mismatch with the average performance (P) of data classified as superior in downstream tasks. high The goal of maximizing conflict.

[0031] Due to limited onboard resources, this invention does not employ gradient descent, but instead uses a combination of batch coordinate descent and momentum updates for data updates. Specifically: S42: Set initial values ​​for the superiority lower limit and the good lower limit, and apply positive and negative perturbations to either of the two lower limit thresholds (i.e., the superiority lower limit or the good lower limit), and substitute them into the utility function of step S41: ; ; in, and Let represent the utility functions after adding a positive perturbation and a negative perturbation respectively during the k-th update of the current lower threshold. Let T1 represent the current lower bound threshold after the k-th update, and T2 represent the lower bound threshold before the update. This represents the perturbation in the k-th update.

[0032] To avoid drastic threshold fluctuations caused by random errors in the on-orbit environment, this invention introduces a momentum term during the update process, specifically: S43: Based on the utility functions after applying the positive and negative perturbations in step S42, determine the update momentum term of the current lower limit threshold using the following formula: ; Among them, v k This represents the update momentum term for the k-th update, where μ represents the momentum coefficient used to smooth historical trends.

[0033] S44: Based on the updated momentum term of the current lower threshold obtained in step S43, update the current lower threshold using the following formula: ; Where η represents the learning rate, used to control the step size of a single adjustment, and A and B represent the physical boundaries of the clip function, respectively.

[0034] S45: If the difference between the current lower limit threshold after the update in step S44 and the current lower limit threshold before the update does not exceed the set change threshold, or the number of updates exceeds the preset upper limit, then stop updating and repeat the operation of steps S42 to S45 with the output lower limit threshold to update the lower limit threshold that has not been updated until the current optimal combination of superior and good lower limits is output.

[0035] It should be noted that in actual spaceborne applications, the initial values ​​of the superior and good lower limits can be set based on sensor design specifications or historical data distribution (for example, in this embodiment, the initial values ​​are set to T respectively). best =0.8, T good=0.5). Under the default condition of limited on-board computing resources, a single set of initial values ​​combined with a momentum smoothing mechanism (μ∈[0.7,0.9]) can be used to gradually approach a better combination of thresholds through the cumulative iteration of long-term on-orbit data. If on-board computing resources permit, multiple sets of differentiated initial values ​​(such as T) can also be selected. best ∈{0.7,0.8,0.9}, T good The above coordinate descent optimization is performed in parallel (∈{0.4,0.5,0.6}), and the set of thresholds with the largest utility function is finally selected as the output to further reduce the dependence on a single initial value.

[0036] For example, the process of updating the lower bound for excellence and the lower bound for goodness respectively includes the following: After setting the initial values ​​of the superior grade lower limit and the good grade lower limit to 0.8 and 0.5 respectively, the superior grade lower limit is first subjected to positive and negative perturbations according to the following formula; ; ; in, and The utility function after adding positive and negative perturbations to the lower bound of dominance during the k-th update. This represents the lower bound of the best performance after the k-th update, and the perturbation in the k-th update. The initial value is set to 0.02, and the perturbation increases with the number of updates. Gradually decrease, in this embodiment of the invention, the perturbation is set to be updated every 5 times. Reduce by half; Based on the utility functions after applying positive and negative perturbations and The update momentum term of the priority lower bound is determined by the following formula: ; in, The update momentum term represents the lower bound of the priority level for the k-th update, with the momentum coefficient μ taking values ​​in the interval [0.7, 0.9]. The lower bound of dominance is updated based on the momentum term of the dominance lower bound using the following formula: ; The learning rate η takes values ​​in the range [0.005, 0.01], and the physical boundaries A and B take values ​​of 0.6 and 0.95, respectively. If the updated lower limit of excellence Compared to the previous superior level lower limit If the difference between the two values ​​does not exceed the set change threshold, or the number of updates exceeds the preset upper limit, then the update stops, and the current lower limit of the superior grade is obtained as the optimal lower limit of the superior grade. In this embodiment of the invention, a total of 20 updates are performed, and the change threshold is set to 0.001. That is, within 20 updates, the difference between the lower limit of the superior grade before and after the update and the difference between the lower limit of the good grade before and after the update do not exceed 0.001. The update stops and the optimal lower limit of the superior grade is obtained. In this embodiment of the invention, the optimal lower limit of the superior grade is 0.79.

[0037] Then, keeping the optimal lower limit of good grade at 0.79, apply positive and negative perturbations to the lower limit of good grade according to the following formula; ; ; in, and The utility function after adding positive and negative perturbations to the k-th update of the goodness lower limit This represents the lower bound of the good grade after the k-th update. This represents the lower bound of the optimal order, i.e. ; Based on the utility functions after applying positive and negative perturbations and The update momentum term for the goodness lower limit is determined by the following formula: ; in, The update momentum term represents the goodness lower bound of the k-th update; Based on the updated momentum term of the goodness lower bound, the goodness lower bound is updated using the following formula: ; If the updated lower limit of good grade Compared to the previous good grade lower limit If the difference does not exceed the set change threshold, or the number of updates exceeds the preset upper limit, then the update stops, and the current good grade lower limit is obtained as the best grade lower limit. In this embodiment of the invention, the best grade lower limit is 0.50.

[0038] In some embodiments, the imbalance penalty term takes the form of negative information entropy: ; Where, p j Let p1, p2, and p3 represent the proportion of data at the j-th level in the total number of data points, and M represent the total number of levels. In this embodiment of the invention, the data is divided into excellent, good, and poor levels, where N=3, and p1, p2, and p3 respectively represent the proportion of data at the excellent, good, and poor levels in the total number of data points.

[0039] In some other embodiments, the imbalance penalty term may also take the variance form of the following expression: ; The variance-based imbalance penalty term can reduce on-board computational overhead; In some other embodiments, when the user has a clear preference for data proportions, the target deviation can be further extended to the following form: ; Where, ω j θ represents the weight of the j-th level. j This represents the user's expected proportion for the j-th level. In this embodiment of the invention, the weights of the three levels are set to 1, and the expected proportions of the three levels are 0.4 (excellent), 0.35 (good), and 0.25 (poor).

[0040] In some other embodiments, the quality evaluation indicators are not limited to MTF and SNR, but may also include radiometric calibration accuracy, geometric positioning error, etc., simply by adding the corresponding items to the comprehensive scoring calculation and adjusting the order of the geometric mean.

[0041] Based on the automatic remote sensing data quality grading method provided by this invention, an automatic remote sensing data quality grading system is also provided, including a data acquisition module, an adaptive weight configuration module, a comprehensive score calculation module, and a quality level classification module. The data acquisition module receives target remote sensing data from sensors and extracts measured values ​​of multiple quality evaluation indicators from the target remote sensing data. The adaptive weight configuration module determines the weight of each quality evaluation indicator based on the scene in which the target remote sensing data is located. The comprehensive score calculation module scores the measured values ​​based on the set target values ​​corresponding to each quality evaluation indicator and in combination with the weights, calculates the geometric mean of all quality scores, and obtains the comprehensive score of the target remote sensing data. The quality level classification module classifies the target remote sensing data according to the comprehensive score and can attach the level information to the image metadata.

[0042] In some embodiments, the measured values ​​of multiple quality evaluation indicators include at least the panchromatic MTF measurement value of the target remote sensing data, the multispectral MTF measurement value of the target remote sensing data, and the panchromatic SNR measurement value of the target remote sensing data, wherein the MTF measurement value is calculated by the edge method or the pulse method, and the SNR measurement value is obtained by uniform scene statistics.

[0043] In different application scenarios, the quality level labels obtained by the method and system provided by this invention can be used to further manipulate the remote sensing data of the target.

[0044] In the application scenario of satellite autonomous on-orbit planning: After imaging, the satellite immediately executes the method and system provided by this invention. If the quality level is found to be "poor," possible reasons include platform vibration causing a decrease in MTF or insufficient illumination causing a low SNR. In this case, the onboard computer adjusts the imaging parameters according to preset rules (such as reducing the pushbroom speed to improve MTF, or increasing the integration time to improve SNR), and retryes on the next imaging opportunity, thereby improving the success rate of data acquisition. It can be seen that in the application scenario of satellite autonomous on-orbit planning, this invention can calculate the quality level of the acquired data in real time. If the level is lower than the threshold, it can automatically adjust subsequent imaging parameters (such as integration time and gain) or replan the observation task, realizing on-board closed-loop control.

[0045] In applications involving the automatic screening of massive amounts of ground-based data: Ground receiving stations receive terabytes of remote sensing data daily. Using the method and system provided by this invention, each data scene is automatically graded. "Excellent" data is directly sent to the advanced product production line; "Good" data is stored in a regular archive; and "Poor" data is stored separately and prompted to the operator for manual inspection or rejection, significantly reducing the workload of manual screening. It is evident that in applications involving the automatic screening of massive amounts of ground-based data, this invention can automatically grade the incoming remote sensing data, prioritizing the distribution of excellent data or its use in high-precision product production, while marking poor data for manual review or disposal, greatly improving data processing efficiency.

[0046] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this invention disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this invention can be achieved, and this is not limited herein.

[0047] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for automatic grading of remote sensing data quality, characterized in that, include: S1: Obtain the measured values ​​of multiple quality evaluation indicators of the target remote sensing data in visible light; the measured values ​​of multiple quality evaluation indicators shall include at least the panchromatic MTF measurement value of the target remote sensing data, the multispectral MTF measurement value of the target remote sensing data, and the panchromatic SNR measurement value of the target remote sensing data. S2: Determine the weight of each quality evaluation indicator in step S1 based on the scene where the target remote sensing data is located. S3: Set the target value for each quality evaluation indicator in step S1, and score the quality of the measured values ​​obtained in step S1 by combining the weights obtained in step S2. In step S3: The first quality score of the panchromatic band MTF measurement is calculated using the following formula: ； in, The first quality score, MTF. pan This represents the panchromatic MTF measurement value, w pan Indicates the corresponding weight. This represents the corresponding target value; The sub-mass score corresponding to each band of the multispectral MTF measurement is calculated using the following formula: ； in, This represents the MTF measurement value for the i-th band. w represents the corresponding sub-quality score. mul Indicates the corresponding weight. This represents the corresponding target value; The second quality score of the multispectral MTF measurement is obtained using the following formula: ； Where N represents the total number of multispectral bands; The third quality score of the panchromatic SNR measurement is calculated using the following formula: ； in, SNR represents the third quality score. pan This represents the panchromatic SNR measurement value, w snr Indicates the corresponding weight. This represents the corresponding target value; S4: Based on the quality score obtained in step S3, classify the target remote sensing data in step S1 into different grades.

2. The automatic remote sensing data quality grading method according to claim 1, characterized in that, In step S2: Multiple sets of weights are pre-calibrated based on historical data or expert experience to obtain a scenario-weight mapping library. By identifying the current scenario, the optimal weight combination of multiple quality evaluation indicators is output, that is, the optimal weight of each quality evaluation indicator. If the optimal weight combination cannot be found in the scenario-weight mapping library based on the scenario, the optimal weight of each quality evaluation indicator in the scenario will be updated in real time based on the application feedback of similar scenarios in historical data.

3. The automatic remote sensing data quality grading method according to claim 1, characterized in that, In step S4: The geometric mean of the quality scores obtained in step S3 is taken as the comprehensive score of the target remote sensing data; Set a threshold range for the comprehensive score corresponding to each level, and determine the level of the target remote sensing data based on whether the comprehensive score falls within the threshold range.

4. The automatic remote sensing data quality grading method according to claim 3, characterized in that, The methods for obtaining the segmentation threshold include: S41: Define the lower bounds for excellent and good grades in the overall score, and determine the utility function as follows: U=α×P high (T best ) β×R discard (T good )-γ×C imbalance ; Where U represents the utility function, T best and T good P represents the lower limit of excellence and the lower limit of goodness, respectively. high R represents the average performance of data classified as superior in downstream tasks at the current threshold. discard C represents the proportion of data that is classified as poor and discarded. imbalance The terms represent the penalty for imbalance in data volume at each level, and α, β, and γ represent the preset hyperparameters. S42: Set initial values ​​for the superior and good lower limits, and apply positive and negative perturbations to either of the two lower limit thresholds, then substitute them into the utility function of step S41: ； ； in, and Let represent the utility functions after adding a positive perturbation and a negative perturbation respectively during the k-th update of the current lower threshold. Let T1 represent the current lower bound threshold after the k-th update, and T2 represent the lower bound threshold before the update. This represents the perturbation in the k-th update; S43: Based on the utility functions after applying the positive and negative perturbations in step S42, determine the update momentum term of the current lower limit threshold using the following formula: ； Among them, v k Let μ represent the update momentum term for the k-th update, and μ represent the momentum coefficient. S44: Based on the updated momentum term of the current lower threshold obtained in step S43, update the current lower threshold using the following formula: ； Where η represents the learning rate, which controls the step size of a single adjustment, and A and B represent the physical boundaries of the clip function, respectively. S45: If the difference between the current lower limit threshold after the update in step S44 and the current lower limit threshold before the update does not exceed the set change threshold, or the number of updates exceeds the preset upper limit, then stop updating and repeat the operation of steps S42 to S45 with the output lower limit threshold to update the lower limit threshold that has not been updated until the current optimal combination of superior and good lower limits is output.

5. The automatic remote sensing data quality grading method according to claim 4, characterized in that, The imbalance penalty term is: ； Where, p j This represents the percentage of data at the j-th level in the total number of data points, and M represents the total number of levels.

6. An automatic remote sensing data quality grading system, comprising the automatic remote sensing data quality grading method according to any one of claims 1 to 5, characterized in that, include: The data acquisition module receives target remote sensing data from the sensor and extracts the measured values ​​of multiple quality evaluation indicators from the target remote sensing data. The adaptive weight configuration module determines the weight of each quality evaluation indicator based on the scene in which the target remote sensing data is located. The comprehensive scoring calculation module calculates the comprehensive score of the target remote sensing data by scoring the measured values ​​based on the target values ​​corresponding to each set quality evaluation index and combining the weights. The quality grading module classifies target remote sensing data into grades based on comprehensive scores and can attach grade information to image metadata.

7. The automatic remote sensing data quality grading system according to claim 6, characterized in that, The measured values ​​of multiple quality evaluation indicators include at least the panchromatic MTF measurement value of the target remote sensing data, the multispectral MTF measurement value of the target remote sensing data, and the panchromatic SNR measurement value of the target remote sensing data.

Images