Satellite imaging quality prediction method and device, electronic equipment and medium

Abstract

This application discloses a method, apparatus, electronic device, and medium for predicting satellite imaging quality. The method includes: acquiring a dataset of the area to be predicted; the dataset includes location information, time information, and meteorological data; evaluating the quality of the dataset using a pre-built classification prediction model to determine its data quality level; and inputting the dataset into a pre-built imaging prediction model of the corresponding level to obtain an imaging prediction score; the higher the imaging prediction score, the higher the imaging suitability of the area to be predicted. Therefore, through a two-stage prediction mechanism, based on the data to be predicted, the imaging suitability of the area to be predicted can be predicted, allowing for the early elimination of mission requests with low imaging suitability, improving the spatiotemporal matching and success rate of imaging plans, reducing invalid imaging and data transmission, and improving satellite imaging efficiency and ground station resource utilization.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of remote sensing images, and in particular to a satellite imaging quality prediction method and device, electronic equipment and a medium. BACKGROUND

[0002] Satellite remote sensing data is widely used in environmental monitoring, resource investigation, urban planning, disaster warning, agricultural yield estimation and many other fields due to its wide coverage, strong timeliness, objective and accurate observation results and other advantages. However, the current satellite remote sensing data acquisition process still faces many challenges.

[0003] Specifically, the satellite remote sensing imaging process is affected by environmental factors. Cloud and fog obstruction, rain and snow interference, atmospheric disturbance, changes in lighting conditions, and low visibility meteorological factors can cause the quality of the acquired images to decrease, thereby affecting the success rate of the imaging task and the usability of the data.

[0004] In addition, the current remote sensing satellites generally use a "sky sensing and ground calculation" working mode, that is, the satellite completes observation on orbit and then downlinks the original data to the ground receiving station for decoding, radiation correction, geometric correction and other image processing. In this way, only a small fraction of the data acquired by the satellite can be timely downlinked, and a large amount of observation data is lost due to the inability to be timely returned. Moreover, the limited bandwidth of satellite-ground communication makes it difficult to support real-time transmission of massive remote sensing data, further reducing the usability and information value of remote sensing image data.

[0005] Therefore, it is of great significance to accurately evaluate the image quality and predict the imaging usability before image acquisition, in order to improve the utilization efficiency of satellite data, reduce resource waste, and ensure the success rate of imaging tasks. SUMMARY

[0006] Therefore, one aspect of the present application provides a satellite imaging quality prediction method, which comprises:

[0007] acquiring a to-be-predicted data set of a to-be-predicted area; the to-be-predicted data set comprises position information, time information and meteorological data;

[0008] performing quality evaluation on the to-be-predicted data set by a pre-constructed classification prediction model to determine a data quality level of the to-be-predicted data set;

[0009] inputting the to-be-predicted data set into a pre-constructed imaging prediction model of a corresponding level according to the data quality level to obtain an imaging prediction score; the higher the imaging prediction score, the higher the imaging suitability of the to-be-predicted area.

[0010] Optionally, the classification prediction model is constructed, comprising:

[0011] acquire historical remote sensing images and an initial classification prediction model constructed in advance;

[0012] construct a training data set according to the historical remote sensing images; the training data set includes a plurality of sub-data sets, different sub-data sets correspond to different data quality level labels, and the sub-data sets include position information, time information and meteorological data corresponding to historical imaging;

[0013] train the initial classification prediction model through the plurality of sub-data sets to obtain the classification prediction model.

[0014] Optionally, constructing the imaging prediction model comprises:

[0015] constructing an initial imaging prediction model; the number of the initial imaging prediction models is the same as the number of the sub-data sets;

[0016] performing one-to-one iterative training of the initial imaging prediction model through the sub-data sets to obtain the imaging prediction model.

[0017] Optionally, the iterative training comprises:

[0018] determining a classification loss function of the initial classification prediction model and a regression loss function of the initial imaging prediction model;

[0019] fusing the classification loss function and the regression loss function to obtain a target loss function;

[0020] adjusting parameters of the initial classification prediction model and the initial imaging prediction model according to the target loss function to minimize the target loss function.

[0021] Optionally, the constructing a training data set according to the historical remote sensing images comprises:

[0022] extracting target image metadata in the historical remote sensing images; the target image metadata includes position information and time information of the historical imaging;

[0023] acquiring initial meteorological data corresponding to the target image metadata;

[0024] screening the initial meteorological data to obtain target meteorological data;

[0025] constructing the training data set through the target image metadata and the target meteorological data.

[0026] Optionally, the screening the initial meteorological data to obtain target meteorological data comprises:

[0027] The initial meteorological data is preprocessed to obtain preprocessed meteorological data; the preprocessing includes missing value filling and normalization processing;

[0028] Meteorological features included in the preprocessed meteorological data are determined;

[0029] A first correlation coefficient of the meteorological features and a target quality score is determined; the target quality score is used to represent the quality of satellite imaging under a specified meteorological feature condition;

[0030] Among the meteorological features, a to-be-removed feature with a first correlation coefficient less than a first threshold value is obtained;

[0031] After the data corresponding to the to-be-removed feature in the preprocessed meteorological data is removed, the target meteorological data is obtained.

[0032] Optionally, before the data corresponding to the to-be-removed feature in the preprocessed meteorological data is removed, the method further includes:

[0033] A second correlation coefficient between any two of the meteorological features is determined; and a meteorological feature pair with a second correlation coefficient greater than a second threshold value is extracted; the second threshold value is greater than the first threshold value;

[0034] A third correlation coefficient of each meteorological feature in the meteorological feature pair and the target quality score is determined;

[0035] Among the meteorological feature pair, the feature corresponding to the third correlation coefficient with a smaller value is taken as the to-be-removed feature.

[0036] Another aspect of the application provides a satellite imaging quality prediction device, the device comprising:

[0037] A to-be-predicted data acquisition module is configured to acquire a to-be-predicted data set of a to-be-predicted area; the to-be-predicted data set includes position information, time information and meteorological data;

[0038] A quality evaluation module is configured to evaluate the quality of the to-be-predicted data set by using a pre-constructed classification prediction model, to determine the data quality level of the to-be-predicted data set;

[0039] An imaging prediction module is configured to input the to-be-predicted data set into a pre-constructed imaging prediction model of a corresponding level according to the data quality level, to obtain an imaging prediction score; the higher the imaging prediction score, the higher the imaging suitability of the to-be-predicted area.

[0040] Another aspect of the present application provides an electronic device comprising a memory and a processor, wherein the memory stores a computer program executable on the processor, and the processor implements the steps of the method for predicting satellite imaging quality when executing the computer program.

[0041] Another aspect of the present application provides a computer readable storage medium storing a computer program, wherein the program implements the steps of the method for predicting satellite imaging quality when executed by a processor.

[0042] The method, device, electronic device and medium for predicting satellite imaging quality provided by the present application have the following beneficial effects: before a satellite imaging task is issued, the data quality of a to-be-predicted data set is coarsely predicted by a classification prediction model, and then a fine-grained imaging score prediction is performed by an imaging prediction model, so that the imaging suitability of a to-be-predicted region is predicted and evaluated based on the location information, time information and weather data of the to-be-predicted region by a two-stage prediction mechanism, and a low-quality condition, i.e., a task request with low imaging suitability, can be eliminated in advance, thereby significantly improving the spatio-temporal matching degree and success rate of an imaging plan, reducing invalid imaging and data transmission, and improving the efficiency of satellite imaging and the utilization rate of ground station resources.

[0043] In addition, the phased modeling and prediction can reduce errors caused by mixed data categories, improve imaging prediction accuracy, and realize imaging condition prediction at any location and time, and can be widely applied to imaging suitability prediction of different regions and different satellites, thereby providing reliable data support for remote sensing applications. BRIEF DESCRIPTION OF DRAWINGS

[0044] Figure 1 A flowchart of a method for predicting satellite imaging quality provided by an embodiment of the present application;

[0045] Figure 2 A principle diagram of a method for predicting satellite imaging quality provided by an embodiment of the present application;

[0046] Figure 3 A principle diagram of a model training method provided by an embodiment of the present application;

[0047] Figure 4 A principle diagram of a weather data processing method provided by an embodiment of the present application;

[0048] Figure 5 A structure diagram of a device for predicting satellite imaging quality provided by an embodiment of the present application;

[0049] Figure 6 A structure diagram of an electronic device provided by an embodiment of the present application.

[0050] The reference signs are as follows: 50 is, 51 is, 52 is, 60 is a memory, 61 is a processor, 62 is a display screen, 63 is an input / output interface, 64 is a communication interface, 65 is a power supply, 66 is a communication bus, 601 is a computer program, 602 is an operating system, and 603 is data. DETAILED DESCRIPTION

[0051] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in this application and the appended claims, the singular forms "a," "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0052] It should be understood that, although the terms first, second, third, etc. can be used herein to describe various information, these terms are not intended to denote a particular order of the information. These terms are used only to distinguish one type of information from another. For example, without departing from the scope of the application, the first information can also be referred to as the second information, and similarly, the second information can also be referred to as the first information. Depending on the context, the word "if" as used herein can be interpreted as "when" or "upon" or "in response to determining."

[0053] Figure 1 A flowchart of a satellite imaging quality prediction method provided by an embodiment of the application is shown in FIG. 1, which comprises the following steps. Figure 1

[0054] S10: obtaining a to-be-predicted data set of a to-be-predicted area; the to-be-predicted data set comprises position information, time information and meteorological data;

[0055] It can be understood that, in specific embodiments, the satellite imaging quality can be affected by different times, different locations and different meteorological conditions, etc. Therefore, in order to determine the imaging suitability of the to-be-predicted area, the position information, time information and meteorological data of the to-be-predicted area need to be obtained first to form the to-be-predicted data set.

[0056] The position information can be the central point longitude and latitude coordinates of the to-be-predicted area, or the longitude and latitude value range of the entire to-be-predicted area, which is not limited in the application. In an optional embodiment, the time information can be the imaging suitability in a future period of time, and correspondingly, the meteorological data is the meteorological data in the future period of time, i.e. the meteorological data is the data in the time range represented by the time information.

[0057] ​In an alternative embodiment, the acquired meteorological data at least includes weather code, average cloud cover, sunshine duration, average sea level pressure, precipitation, precipitation duration, average temperature, average dew point temperature, average humidity, and day length. In fact, the application does not make specific limitations on the acquired meteorological data.

[0058] It should be noted that in specific embodiments, the to-be-predicted area can be one or multiple, and a to-be-predicted data set is acquired for different to-be-predicted areas.

[0059] S11: quality evaluation of the to-be-predicted data set by the pre-constructed classification prediction model to determine the data quality level of the to-be-predicted data set;

[0060] Figure 2 The principle diagram of the satellite imaging quality prediction method provided by the embodiments of the application is shown in the figure. In an alternative embodiment, to ensure the prediction accuracy of satellite imaging quality, two-stage prediction models are set. Specifically, as shown in the figure, the first-stage classification prediction model and the second-stage imaging prediction model are pre-constructed. Figure 2

[0061] In specific embodiments, the acquired to-be-predicted data set is first input to the first-stage classification prediction model for processing, thereby determining the data quality level of the to-be-predicted data set. In this stage, it can be understood that the to-be-predicted data set is classified in a coarse-grained manner and divided into data of different quality levels.

[0062] It should be noted that the number of data quality levels is not limited by the application, and the number of levels is determined when the classification prediction model is pre-constructed. Specifically, the classification prediction model can be trained by multiple data of different data quality levels, so that the classification data model has the ability to identify data quality and divide it into corresponding quality levels.

[0063] Of course, in an alternative embodiment, to save computing resources, the data quality level can be divided into two levels. To facilitate understanding, the following will take the data quality level divided into two levels of high quality level and low quality level as an example, and the classification prediction model will be described in detail in combination with the Figure 2

[0064] As shown in the figure, after the to-be-predicted data set is input to the classification prediction model, the classification prediction model processes it, thereby identifying whether the quality level of the current to-be-predicted data set belongs to the low quality level or the high quality level. Figure 2

[0065] ​​​S12: input the to-be-predicted data set into a pre-constructed imaging prediction model corresponding to the data quality level according to the data quality level, to obtain an imaging prediction score; the higher the imaging prediction score, the higher the imaging suitability degree of the to-be-predicted region.

[0066] Further, in an optional embodiment, in order to further improve the imaging suitability prediction accuracy, for the data of different data quality levels in the first stage prediction, different levels of imaging prediction models are selected for prediction. That is, the number of imaging prediction models is the same as the number of data quality levels. As shown in the embodiment, the data quality levels include two levels of low quality level and high quality level, and the imaging prediction models include a first imaging prediction model and a second imaging prediction model, wherein the first imaging prediction model is used for imaging prediction of low quality level data, and the second imaging prediction model is used for imaging prediction of high quality level data. Figure 2

[0067] Similarly, it should be noted that in specific embodiments, the number of imaging prediction models is not limited in the present application, as long as the number of imaging prediction models is the same as the number of data quality levels, and different quality levels of data can be predicted by one imaging prediction model.

[0068] As shown in the embodiment, on the basis of the first stage coarse-grained prediction, according to different data quality levels, the corresponding imaging prediction model is selected to process the to-be-predicted data set, so as to obtain the imaging prediction score of the to-be-predicted region. The higher the obtained imaging prediction score, the higher the imaging suitability of the to-be-predicted region. Figure 2

[0069] According to the embodiment 2, after obtaining the imaging prediction score, further, as an optional embodiment, the imaging suitability level of the to-be-predicted region can be determined according to the pre-constructed mapping relationship, so that the user can more intuitively determine whether the satellite imaging task can be performed at present.

[0070] The mapping relationship is the corresponding relationship between the imaging score and the suitability level, for example, in an optional embodiment, when the imaging score is 5 points, which is the highest score, the corresponding suitability level is the first level, which represents that the imaging condition is excellent and suitable for performing the imaging task. When the imaging score is 3-4 points, the corresponding suitability level is the second level, the imaging condition is general, the imaging is feasible but the quality is uncertain. When the imaging score is 0-2 points, the corresponding suitability level is the third level, the imaging condition is poor and not suitable for performing the imaging task.

[0071] ​​In specific embodiments, the two-stage imaging prediction method provided in the present application can accurately predict the imaging suitability of different meteorological data and different times. Specifically, Table 1 is a prediction result table of remote sensing image quality under different meteorological data conditions provided by an embodiment of the present application. The imaging accuracy under different meteorological data conditions will be described below in conjunction with Table 1.

[0072] Table 1 is a prediction result table of remote sensing image quality under different meteorological data conditions provided by an embodiment of the present application. The imaging accuracy under different meteorological data conditions will be described below in conjunction with Table 1.

[0073]

[0074] As shown in Table 1, the remote sensing image quality under different weather phenomena is counted. It is found that the accuracy rate is the highest when the weather is sunny, which can reach 97.96%. Rain and snow all have an impact on prediction, and as the rain and snow intensity increases, the prediction accuracy rate decreases, and the accuracy rate is the lowest when it is heavily snowing, which is 78.79%.

[0075] Table 2 is a prediction result table of remote sensing image quality under different time periods provided by an embodiment of the present application. The imaging accuracy under different time periods will be described below in conjunction with Table 2.

[0076] Table 2 is a prediction result table of remote sensing image quality under different time periods provided by an embodiment of the present application. The imaging accuracy under different time periods will be described below in conjunction with Table 2.

[0077]

[0078] As shown in Table 2, the remote sensing image quality of different months is counted. The order of imaging quality from high to low in different seasons is winter > autumn > spring > summer. The accuracy rate is the lowest in July, which is 87.34%. After analysis, it is found that this is because July is the peak of summer, and the rainfall is frequent and strong. Especially in the south and east of China, affected by the monsoon, there are more rainstorms, thunderstorms, thick clouds and rapid changes, which makes it difficult to predict the imaging window. When it is rainy and cloudy, the imaging success rate decreases.

[0079] In an optional embodiment, different regions to be detected can be predicted and statistically analyzed. It is known from the statistics that for the Northeast, Central China and North China regions, due to the flat terrain and stable climate, the weather forecast is relatively accurate, the stability of remote sensing imaging is good, and it is beneficial to continuous observation of optical remote sensing. In East China, South China, especially the coastal areas and complex terrain areas, convective clouds develop quickly, local weather is changeable, and forecasting is difficult, and imaging is affected. In addition, Guangxi and Hainan are located in the humid south, with high humidity and high rainfall all year round, which also leads to a low accuracy rate. From Shaanxi in the northwest to the Qinling Mountains, the Loess Plateau and the Qinling Mountains alternate, with sand blowing in spring, heavy rain in summer and haze in autumn and winter, resulting in low imaging prediction accuracy. For Zhoushan, Shantou and Zhongshan regions, which belong to marine weather, the rapid development of convection will also affect the imaging quality. Overall, the combination of climate zones, terrain and rapid convection will lead to variable weather and pose a certain difficulty to remote sensing imaging.

[0080] In an optional embodiment, imaging prediction can be performed according to the meteorological data of the cloud percentage. Table 3 is a table showing the imaging prediction results for different cloud percentages according to an embodiment of the present application. The imaging prediction results for different cloud percentages will be described below in combination with Table 3.

[0081] Table 3 is a table showing the imaging prediction results for different cloud percentages

[0082]

[0083] As shown in Table 3, for different cloud percentages, the corresponding imaging suitability is different. For no cloud weather conditions, the imaging conditions are best. It should be noted that in specific embodiments, the quality of the cloud distribution is visually determined, and there is a certain degree of freedom for complex situations. Specifically, the range of scattered cloud distribution is calculated, the range of thin clouds is appropriately reduced, the range of cloud layer blocking ground important targets is appropriately increased, and the range of cloud layer blocking land at sea-land boundary is calculated.

[0084] Based on the above description, in an optional embodiment, the prediction method provided by the present application can be used to predict the imaging suitability score for different regions, times and weather conditions before the imaging task is performed. By combining satellite state information and task requirements, the imaging task priority can be dynamically adjusted to avoid task planning failure caused by resource changes or meteorological mutations, and to ensure the stability and response efficiency of the imaging task execution.

[0085] Specifically, high-risk imaging periods and regions such as atmospheric instability, complex local circulation and frequent strong convection are automatically identified, the imaging task priority is reduced or delayed, and the weight is adjusted according to different seasonal weather patterns, the imaging condition stable period is preferentially arranged, and the proportion of effective data throughout the year is improved.

[0086] In addition, the prediction result can be applied to multi-satellite cooperative scheduling to realize dynamic monitoring, rapid feedback and continuous coverage of a specific area, and significantly improve the spatial coverage and time continuity of disaster emergency, major event monitoring and long-term sequence observation.

[0087] Finally, before performing the imaging task, low-quality data is screened out to reduce satellite, ground link bandwidth occupation and storage pressure, and reduce the redundant burden of ground station reception, storage and processing. In specific embodiments, an API interface call is supported to provide stable data services for third-party imaging planning systems.

[0088] Therefore, the satellite imaging quality prediction method provided by the embodiments of the present application can perform coarse-grained prediction on the data quality of the to-be-predicted data set by the classification prediction model before the satellite imaging task is issued, and further perform fine-grained imaging score prediction by the imaging prediction model. Thus, through the two-stage prediction mechanism, the imaging suitability of the to-be-predicted area is predicted and evaluated based on the location information, time information and weather data of the to-be-predicted area. Low-quality conditions, i.e., imaging suitability low task requests, can be eliminated in advance, the spatio-temporal matching degree and success rate of imaging planning can be significantly improved, invalid imaging and data transmission can be reduced, and satellite imaging efficiency and ground station resource utilization rate can be improved. In addition, the phased modeling prediction can reduce the error caused by mixed data categories, improve the imaging prediction accuracy, realize imaging condition prediction at any place and time, and can be widely applied to imaging suitability prediction of different regions and different satellites to provide reliable data support for remote sensing applications.

[0089] In an optional embodiment, the classification prediction model is constructed, including:

[0090] obtaining historical remote sensing images and a pre-constructed initial classification prediction model;

[0091] constructing a training data set according to the historical remote sensing images; the training data set includes a plurality of sub-data sets, different sub-data sets correspond to different data quality level labels, and the sub-data sets include location information, time information and weather data corresponding to historical imaging;

[0092] iteratively training the initial classification prediction model through the plurality of sub-data sets to obtain the classification prediction model.

[0093] In specific embodiments, the initial classification prediction model is pre-constructed, and at the same time, a large amount of historical remote sensing image data is obtained. It can be understood that the remote sensing images include latitude and longitude and shooting time information, and therefore, the location information and time information of historical imaging can be extracted from the historical remote sensing images. In addition, the corresponding weather data can be obtained based on the location information and time information.

[0094] In one optional embodiment, meteorological information corresponding to the image capture can be obtained by calling a historical weather API interface. This historical weather API interface can support searching historical weather data by latitude, longitude, and date. The query results may include, but are not limited to, weather phenomena (e.g., sunny, rainy, and snowy), temperature, precipitation, air pressure, cloud cover, surface radiation, wind speed and direction, daylight hours, and sunshine duration.

[0095] Based on acquired historical remote sensing imagery, a training dataset can be constructed that includes the location, time, and meteorological data of the historical images. Understandably, this training dataset is used to train an initial classification and prediction model, allowing the model to learn from the data in the training dataset. This enables the trained model to be directly used subsequently to predict the data quality level of the dataset to be predicted.

[0096] Therefore, the training dataset includes multiple subsets corresponding to different data quality level labels. In other words, the data included in different subsets belong to different data quality levels. In order for the initial classification prediction model to obtain data quality levels, level labels are set for different subsets.

[0097] The number of subsets is the same as the number of data quality level labels. A single subset can include data from multiple locations, times, and weather conditions. In other words, for a large amount of historical remote sensing imagery, all data is divided according to different data quality levels. For ease of understanding, the following explanation will use low-quality and high-quality levels as examples.

[0098] For example, the training dataset can be divided into two subsets: one subset representing low-quality data and the other representing high-quality data. It should be noted that the subset division can be done manually or by using a specified large language model; this application does not impose any restrictions on this.

[0099] Figure 3 This is a schematic diagram illustrating the principle of a model training method provided in an embodiment of this application. Further, as... Figure 3 As shown, the initial classification prediction model is iteratively trained using a training dataset to obtain the final classification prediction model. In one optional instance, the classification prediction model can be a LightGBM (Light Gradient Boosting Machine) classification model. The iterative training time can be used to obtain the classification loss function of the classification prediction model. The parameters of the model are optimized with the goal of minimizing the classification loss function until the classification loss function reaches the convergence condition, thereby obtaining the final classification prediction model.

[0100] In the above two quality level examples, the prediction result of the classification prediction model can be represented as wherein, when is represented as a low quality level, and when is represented as a high quality level. In an optional embodiment, after training the classification prediction model, the classification prediction model pkl file is stored for subsequent model deployment and inference.

[0101] On the basis of the above embodiment, as an optional embodiment, the imaging prediction model is constructed, comprising:

[0102] constructing an initial imaging prediction model; the number of initial imaging prediction models is the same as the number of sub-data sets;

[0103] The initial imaging prediction model is trained one by one by the sub-data set to obtain the imaging prediction model.

[0104] In a specific embodiment, first, a plurality of initial imaging prediction models are constructed, and the specific number is the same as the number of sub-data sets obtained in the above embodiment, i.e., the same as the number of data quality level labels in the above embodiment.

[0105] In order to more specifically predict imaging for data of different qualities, the present application sets up a plurality of imaging prediction models corresponding to different quality levels. In fact, for high-quality data, the accuracy of prediction by the model is very high, and after prediction, the imaging task can be directly performed. However, in fact, in a specific embodiment, the weather is very rare, and most of the time the environmental weather condition is not very good, at which time the quality of the data to be predicted is also relatively low, and dividing different imaging prediction models for prediction can obtain data that can realize the imaging task from the low-quality data as much as possible. That is, different quality level data selects different imaging prediction models, and the purpose is not only to obtain high-quality data, but more importantly, to obtain data that can realize the imaging task from low-quality data as much as possible.

[0106] Therefore, in the embodiment of the present application, the sub-data sets of different data quality level labels are respectively input into different initial imaging prediction models for one-to-one iterative training to obtain the imaging prediction model that can be directly used for imaging suitability prediction. In order to facilitate understanding, the following will be described taking high and low two quality levels as examples.

[0107] In an optional embodiment, the data quality level label includes a low quality level label and a high quality level label, and correspondingly, two sub-data sets are included, the initial imaging prediction model is a LightGBM regression model, and the model includes a first initial imaging prediction model for predicting low-quality level data And a second initial imaging prediction model for predicting high-quality data. .

[0108] The prediction results of the initial imaging predictive model can be expressed as follows: , , The two prediction results can be combined as follows: Similarly, the trained imaging prediction model is stored as a pkl file for subsequent model deployment and inference.

[0109] In one alternative embodiment, iterative training includes:

[0110] Determine the classification loss function of the initial classification prediction model and the regression loss function of the initial imaging prediction model;

[0111] The classification loss function and the regression loss function are fused to obtain the target loss function;

[0112] Based on the target loss function, the parameters of the initial classification prediction model and the initial imaging prediction model are iteratively adjusted to minimize the target loss function.

[0113] In a specific embodiment, the loss functions of the classification prediction model and the imaging prediction model can be determined separately, and the parameters of the two models can be optimized separately with the goal of minimizing the loss function to obtain the final model. However, it is understood that there is a close relationship between the two models. In a specific embodiment, the output of the classification prediction model is directly input into the imaging prediction model. Therefore, in an optional embodiment, to further improve the imaging prediction accuracy, the two models are jointly optimized. That is, the loss function adopts a standardized weighted combination of classification and regression losses to improve prediction accuracy.

[0114] Specifically, such as Figure 3 As shown, during model training, the classification loss function of the initial classification model and the regression loss function of the initial imaging prediction model are calculated. In an optional embodiment, the classification loss function can be the cross-entropy loss function, calculated using the following formula: ,in, The total number of samples, Indicates the first The data quality level label of the input training samples, for example, in the two types of data quality level labels in the above embodiment, when the data quality level label is high quality, the representation is a positive class result. When the data quality level is marked as low, the representation is a negative result. It should be noted that the position information, time information and weather data included in a historical image data are an input sample. represent all input samples predicted as positive class, i.e. probability.

[0115] In another alternative embodiment, the regression loss function can be a mean square error loss function, and the calculation formula is wherein, represents the predicted quality level of the th input training sample, represents the real data quality level.

[0116] After obtaining the classification loss function and the regression loss function, as shown in Figure 3 , the two loss functions are fused to obtain the target loss function. Specifically, in an alternative embodiment, the loss functions of the two stages are dimensionally aligned, and the specific calculation formula is , wherein, is the normalized classification loss, is the normalized regression loss, is the average value of the classification loss, is the average value of the regression loss, is the standard deviation of the classification loss, is the standard deviation of the regression loss, is a very small positive number to prevent division by zero, and in a specific embodiment, it can be .

[0117] After the two alignments, in an alternative embodiment, the classification loss and the regression loss index fall within the value interval of , similar to a normal distribution, so as to facilitate subsequent joint optimization training of the model.

[0118] Thus, the target loss function expression can be obtained as follows: wherein, is a weight coefficient, and in an alternative embodiment, the weight coefficient can take a value of 0.5 to maintain the balance between the classification loss and the regression loss.

[0119] As shown in Figure 3 , after obtaining the target loss function, the model parameters of the classification prediction model and the imaging prediction model are continuously iteratively optimized with the goal of minimizing the target loss function until the target loss function reaches the convergence condition, thereby obtaining the prediction models of the two stages, which can be returned through a standardized API interface for calling by a remote sensing imaging planning system or other third-party task planning system.

[0120] In specific embodiments, a two-stage model is loaded, for example, as shown in FIG. 2, a coarse-grained classification is first predicted by using a one-stage classification prediction model, and then a corresponding imaging prediction model is used for accurate prediction according to the classification result. Figure 2

[0121] After predicting the image quality, the predicted value is compared with the true value, and the image quality prediction accuracy of different cities, the image quality prediction accuracy under different meteorological conditions, and the image quality prediction accuracy at different times are calculated respectively.

[0122] In an optional embodiment, a plurality of different prediction methods are compared to analyze the accuracy of different prediction methods. Table 4 is a comparison result table of imaging accuracy of different prediction methods provided by the embodiment of the present application, which will be described below in conjunction with Table 4.

[0123] Table 4 is a comparison result table of imaging accuracy of different prediction methods

[0124]

[0125] The mean square error RMSE and the coefficient of determination R2 of the single-stage regression model and the two-stage model provided by the present application are compared. As shown in Table 4, the two-stage model provided by the present application reduces the average prediction error by 0.144 compared with the single-stage Light_GBM model. The coefficient of determination is improved by 11.2%, and the overall fitting effect of the model is better. It can be seen that the two-stage method provided by the present application reduces the local error by segmenting the model, and effectively enhances the robustness.

[0126] Therefore, the satellite imaging quality prediction method provided by the embodiment of the present application constructs a joint classification and regression prediction model based on historical image quality data, improves the stability and accuracy of the determination of meteorological elements such as cloud cover and humidity, reduces the misjudgment and misfiltration problem of the traditional cloud cover threshold method, and improves the ability to distinguish between good and bad imaging conditions.

[0127] In an optional embodiment, the initial meteorological data is filtered to obtain target meteorological data, including:

[0128] After preprocessing the initial meteorological data, preprocessed meteorological data is obtained; the preprocessing includes missing value filling and normalization processing;

[0129] Determine the meteorological features included in the preprocessed meteorological data;

[0130] Determine the first correlation coefficient of the meteorological features and the target quality score; the target quality score is used to represent the quality of satellite imaging under the condition of a specified meteorological feature;

[0131] ​In acquiring meteorological features, features whose first correlation coefficient is less than the first threshold are to be removed.

[0132] After removing the data corresponding to the features to be removed from the preprocessed meteorological data, the target meteorological data is obtained.

[0133] It is understandable that the initial meteorological data obtained initially includes a lot of incomplete and poor-quality data, which will directly affect the training effect of the model. Therefore, in order to improve the training accuracy of the model, in one optional embodiment, the initial meteorological data is preprocessed after it is obtained.

[0134] Specifically, for meteorological features with missing values, the data is filled with the average value of the corresponding meteorological feature, and continuous meteorological features are normalized. The processed data can be saved to the PostGIS spatial database, supporting direct data retrieval from the database for model training and validation.

[0135] Figure 4 This is a schematic diagram illustrating the principle of meteorological data processing provided in an embodiment of this application, as shown below. Figure 4 As shown, after preprocessing the initial meteorological data to obtain preprocessed meteorological data, the meteorological data screening stage begins.

[0136] Specifically, in one optional embodiment, the meteorological characteristics included in the preprocessed meteorological data are first determined, wherein the meteorological characteristics may include, but are not limited to, maximum, minimum, and average wind speeds, reference total evapotranspiration, cloud cover, dew point temperature, ambient temperature, sunshine duration, day length, and air pressure.

[0137] It is understandable that, in specific embodiments, the data corresponding to meteorological characteristics such as maximum, minimum, and average wind speeds and reference evapotranspiration have little impact on image quality and are therefore invalid data. In order to improve the efficiency and accuracy of model training, these data need to be screened and removed.

[0138] Therefore, in an alternative embodiment, such as Figure 4 As shown, the first correlation coefficient between each meteorological feature and the target quality score is calculated using the following formula: ,in, The first correlation coefficient, For the first A meteorological feature, for example, temperature. The target quality score is used to characterize the quality of satellite imaging under specified meteorological conditions. It is a specific numerical score. For example, under specified meteorological conditions including cloud cover, precipitation, temperature, humidity and air pressure, the target quality score is 3 points, which reflects the quality of imaging under those meteorological conditions and also the suitability of imaging under those meteorological conditions.

[0139] a weather feature and a target quality score , a covariance between a weather feature and a target quality score , a standard deviation of a weather feature , a standard deviation of a target quality score , a first correlation coefficient , for measuring a degree of linear correlation between a weather feature and a target quality score , when a weather feature and a target quality score are completely positively correlated, when a weather feature and a target quality score are completely negatively correlated, when a weather feature and a target quality score are not correlated.

[0140] Further, after the first correlation coefficient is calculated, in an optional embodiment, a feature with a first correlation coefficient less than a first threshold value is extracted as a feature to be removed from the weather features in the preprocessed weather data. As shown in Figure 4 , after the data corresponding to the feature to be removed is removed from the preprocessed weather data, target weather data is obtained.

[0141] On the basis of the above embodiment, as an optional embodiment, before the data corresponding to the feature to be removed is removed from the preprocessed weather data, the method further comprises:

[0142] determining a second correlation coefficient between any two weather features; and extracting a pair of weather features with a second correlation coefficient greater than a second threshold value; the second threshold value is greater than the first threshold value;

[0143] determining a third correlation coefficient between each weather feature in the pair of weather features and the target quality score;

[0144] taking the feature corresponding to the weather feature with a smaller third correlation coefficient in the pair of weather features as the feature to be removed.

[0145] In an embodiment, the preprocessed weather data can include multiple similar weather features, for example, data of temperature or data of air pressure. For weather features with the same meaning, in order to ensure the calculation efficiency and save the calculation resources, one of the similar weather features can be deleted.

[0146] In fact, when there are too many meteorological features, the trained model is poor in scalability and is not conducive to wide application in various scenarios. In addition, the more meteorological features, the poorer the performance and efficiency of the trained model. At the same time, after screening the meteorological features, the amount of data screened out can also meet the demand.

[0147] Based on the above description, in an optional embodiment, as shown in Figure 4 the first correlation coefficient of the above embodiment is regarded as the first stage screening of preprocessing meteorological data, and the multicollinearity detection provided in the embodiment of the application is regarded as the second stage screening, so as to further improve the training efficiency and accuracy of the model.

[0148] Specifically, as shown in Figure 4 , when performing multicollinearity detection, the correlation matrix of all meteorological features can be calculated, that is, the second correlation coefficient between any two meteorological features is calculated first. The calculation of the second correlation coefficient is the same as the calculation of the first correlation coefficient in the above embodiment, and will not be described here. At the same time, the meteorological feature pair with the second correlation coefficient greater than the second threshold value is extracted, that is, two highly similar meteorological features are screened out. It should be noted that the features screened here are highly similar features, so the second threshold value is greater than the first threshold value.

[0149] After extracting the highly similar meteorological feature pair, further, the third correlation coefficient of each meteorological feature in the meteorological feature pair and the target quality score is calculated, and the feature corresponding to the smaller third correlation coefficient is removed as a to-be-removed feature. That is, of the two highly similar meteorological features, only the feature more similar to the target quality score is retained.

[0150] It should be noted that in specific embodiments, Figure 4 the two screening stages shown in Figure 4 may be performed simultaneously or one stage can be performed first, and the application does not limit , which is only one realizable embodiment. In specific embodiments, after two stages of screening, the obtained target meteorological features can include but are not limited to weather code, average cloud cover, sunshine time, average sea level pressure, precipitation, precipitation duration, average temperature, average dew point temperature, average humidity, and day length.

[0151] It should be noted that in an optional embodiment, the initial meteorological data is screened to facilitate subsequent construction of a training data set. In specific embodiments, different quality data models focus on different meteorological features, so after obtaining the training data set, the meteorological features of different data quality levels can be sorted, and the final target training data can be selected based on the sorting result.

[0152] For example, the ranking result of the meteorological features for a low quality level is pressure, day length, temperature, dew point temperature, sunshine duration and cloud cover. The ranking result of the meteorological features for a high quality level is cloud cover, dew point temperature, ambient temperature, sunshine duration, day length and pressure. When selecting the target training data, data in a preset percentage of the ranking result can be selected, for example, data in the top 70% of the ranking result. The final selection manner is not limited in the present application.

[0153] In an optional embodiment, the latest historical remote sensing image can be acquired every interval of time, and the classification prediction model and the imaging prediction model are retrained and optimized after the data screening in the above embodiments, so as to continuously adapt to environmental changes and ensure the imaging prediction accuracy.

[0154] Therefore, the satellite imaging quality prediction method provided in the present application supports weather imaging quality prediction for any place and any time, encapsulates an API interface service to provide a third-party system, for example, a remote sensing imaging planning system, calls, and divides into three categories, such as high (for example, 5 points suitable for imaging), medium (for example, 3 to 4 points for general imaging conditions), and low (for example, 0 to 2 points unsuitable for imaging), which significantly reduces invalid imaging tasks and supports intelligent task planning and autonomous decision-making.

[0155] In the above embodiments, the satellite imaging quality prediction method is described in detail, and the present application also provides a corresponding embodiment of a satellite imaging quality prediction device.

[0156] Figure 5 A structure diagram of a satellite imaging quality prediction device provided in an embodiment of the present application is shown in FIG. 1, which includes: Figure 5

[0157] The prediction data acquisition module 50 is configured to acquire a prediction data set of a prediction area; the prediction data set includes position information, time information and meteorological data;

[0158] The quality evaluation module 51 is configured to evaluate the quality of the prediction data set by using a pre-constructed classification prediction model, so as to determine the data quality level of the prediction data set;

[0159] The imaging prediction module 52 is configured to input the prediction data set into a pre-constructed imaging prediction model corresponding to the data quality level, so as to obtain an imaging prediction score; the higher the imaging prediction score is, the higher the imaging suitability of the prediction area is.

[0160] In addition, the satellite imaging quality prediction device provided in the embodiment of the present application further includes:

[0161] ​The first acquisition module is configured to acquire historical remote sensing images and an initial classification prediction model constructed in advance.

[0162] The training data set construction module is configured to construct a training data set according to the historical remote sensing images. The training data set includes a plurality of sub-data sets. Different sub-data sets correspond to different data quality level labels. The sub-data sets include location information, time information and meteorological data corresponding to historical imaging.

[0163] The iterative training module is configured to perform iterative training on the initial classification prediction model through the plurality of sub-data sets to obtain a classification prediction model.

[0164] The initial imaging prediction model construction module is configured to construct an initial imaging prediction model. The number of initial imaging prediction models is the same as the number of sub-data sets.

[0165] The iterative training module is further configured to perform one-to-one iterative training on the initial imaging prediction model through the sub-data sets to obtain an imaging prediction model.

[0166] The loss function determination module is configured to determine a classification loss function of the initial classification prediction model and a regression loss function of the initial imaging prediction model.

[0167] The loss function fusion module is configured to fuse the classification loss function and the regression loss function to obtain a target loss function.

[0168] The parameter optimization module is configured to perform iterative adjustment on parameters of the initial classification prediction model and the initial imaging prediction model according to the target loss function to minimize the target loss function.

[0169] The metadata extraction module is configured to extract target image metadata in the historical remote sensing images. The target image metadata includes location information and time information of historical imaging.

[0170] The initial meteorological data acquisition module is configured to acquire initial meteorological data corresponding to the target image metadata.

[0171] The meteorological data screening module is configured to screen the initial meteorological data to obtain target meteorological data.

[0172] The data set construction module is configured to construct a training data set through the target image metadata and the target meteorological data.

[0173] The preprocessing module is configured to obtain preprocessed meteorological data by preprocessing the initial meteorological data. The preprocessing includes missing value filling and normalization processing.

[0174] The meteorological feature determination module is configured to determine meteorological features included in the preprocessed meteorological data.

[0175] The first correlation coefficient determination module is configured to determine a first correlation coefficient between the meteorological feature and the target quality score, the target quality score being used to represent the quality of satellite imaging under a specified meteorological feature condition.

[0176] The feature to be removed determination module is configured to obtain a feature to be removed from the meteorological features, the feature to be removed being a meteorological feature whose first correlation coefficient is less than the first threshold.

[0177] The data removal module is configured to obtain target meteorological data by removing data corresponding to the feature to be removed from the preprocessed meteorological data.

[0178] The second correlation coefficient determination module is configured to determine a second correlation coefficient between any two meteorological features, and extract a pair of meteorological features whose second correlation coefficient is greater than a second threshold, the second threshold being greater than the first threshold.

[0179] The third correlation coefficient determination module is configured to determine a third correlation coefficient between each meteorological feature in the pair of meteorological features and the target quality score.

[0180] The feature to be removed determination module is further configured to take the feature corresponding to the meteorological feature with a smaller third correlation coefficient in the pair of meteorological features as the feature to be removed.

[0181] Figure 6 A structural schematic diagram of an electronic device provided by an embodiment of the present application is shown in FIG. 1. Figure 6 As shown in FIG. 1, the electronic device includes a memory 60 configured to store a computer program.

[0182] A processor 61 is configured to implement the steps of the satellite imaging quality prediction method mentioned in the above embodiments when executing the computer program.

[0183] The electronic device provided by the embodiment can include but is not limited to a notebook computer or a desktop computer, etc.

[0184] The processor 61 can include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 61 can be implemented in at least one of a hardware form of a digital signal processor (DSP), a field-programmable gate array (FPGA), a programmable logic array (PLA), etc. The processor 61 can also include a main processor and a co-processor. The main processor is a processor for processing data in a wake-up state, also referred to as a central processing unit (CPU). The co-processor is a low-power processor for processing data in a standby state. In some embodiments, the processor 61 can be integrated with a graphics processor (GPU) for rendering and drawing content to be displayed by the display screen. In some embodiments, the processor 61 can further include an artificial intelligence (AI) processor for processing machine learning-related computing operations.

[0185] The memory 60 can include one or more computer-readable storage media that can be non-transitory. The memory 60 can further include a high-speed random access memory, and a nonvolatile memory such as one or more disk storage devices, flash storage devices. In this embodiment, the memory 60 is at least used to store the following computer program 601, wherein the computer program is loaded and executed by the processor 61, and can implement the related steps of the satellite imaging quality prediction method disclosed in any of the preceding embodiments. In addition, the resources stored in the memory 60 can further include an operating system 602 and data 603, etc., and the storage manner can be temporary storage or permanent storage. The operating system 602 can include Windows, Unix, Linux, etc. The data 603 can include, but is not limited to, related data involved in the satellite imaging quality prediction method, etc.

[0186] In some embodiments, the electronic device can further include a display screen 62, an input / output interface 63, a communication interface 64, a power supply 65, and a communication bus 66.

[0187] Those skilled in the art can understand that, Figure 6 The structure shown in the figure does not constitute a limitation on the electronic device, and can include more or fewer components than those shown.

[0188] The electronic device provided by the embodiment of the present application comprises a memory and a processor, and the processor can realize the satellite imaging quality prediction method in the above embodiment when executing the program stored in the memory.

[0189] It should be noted that, although the operations are depicted in a particular order in the accompanying drawings, this should not be understood as requiring the operations to be performed in the particular order shown or sequentially, or requiring all of the illustrated operations to be performed to achieve a desired result. In some cases, multi-tasking and parallel processing can be advantageous. In addition, the separation of various system modules and components in the above embodiments should not be understood as requiring such separation in all embodiments, and it should be understood that the described program components and systems can generally be integrated together in a single software product, or packaged into multiple software products.

Claims

1. A method of predicting satellite imaging quality, characterized in that, The method comprises: acquiring a to-be-predicted data set of a to-be-predicted area; the to-be-predicted data set comprises position information, time information and meteorological data; performing quality evaluation on the to-be-predicted data set by a pre-constructed classification prediction model, to determine a data quality level of the to-be-predicted data set; inputting the to-be-predicted data set into a pre-constructed imaging prediction model of a corresponding level according to the data quality level, to obtain an imaging prediction score; the higher the imaging prediction score is, the higher the imaging suitability of the to-be-predicted area is; constructing the classification prediction model, comprising: acquiring historical remote sensing images and a pre-constructed initial classification prediction model; constructing a training data set according to the historical remote sensing images; the training data set comprises a plurality of sub-data sets, different sub-data sets correspond to different data quality level labels, and the sub-data sets comprise position information, time information and meteorological data corresponding to historical imaging; performing iterative training on the initial classification prediction model by the plurality of sub-data sets, to obtain the classification prediction model; constructing the imaging prediction model, comprising: constructing an initial imaging prediction model; the number of initial imaging prediction models is the same as the number of sub-data sets; performing one-to-one iterative training on the initial imaging prediction model by the sub-data sets, to obtain the imaging prediction model; the iterative training comprises: determining a classification loss function of the initial classification prediction model and a regression loss function of the initial imaging prediction model; fusing the classification loss function and the regression loss function to obtain a target loss function; performing iterative adjustment on parameters of the initial classification prediction model and the initial imaging prediction model according to the target loss function, to minimize the target loss function.

2. The method of predicting satellite imaging quality according to claim 1, wherein, the constructing a training data set according to the historical remote sensing images comprises: extracting target image metadata in the historical remote sensing images; the target image metadata comprises position information and time information of the historical imaging; acquiring initial meteorological data corresponding to the target image metadata; performing screening on the initial meteorological data to obtain target meteorological data; constructing the training data set by the target image metadata and the target meteorological data.

3. The method of predicting satellite imaging quality according to claim 2, wherein, the performing screening on the initial meteorological data to obtain target meteorological data comprises: obtaining preprocessed meteorological data after preprocessing the initial meteorological data; the preprocessing comprises missing value filling and normalization processing; determining meteorological features included in the preprocessed meteorological data; determining a first correlation coefficient of the meteorological features and a target quality score; the target quality score is used to represent the quality of satellite imaging under a specified meteorological feature condition; acquiring to-be-removed features in the meteorological features, for which the first correlation coefficient is less than a first threshold value; obtaining the target meteorological data by removing data corresponding to the to-be-removed features in the preprocessed meteorological data.

4. The method of predicting satellite imaging quality according to claim 3, wherein, before the removing data corresponding to the to-be-removed features in the preprocessed meteorological data, further comprising: determining a second correlation coefficient between any two of the meteorological features; and extracting a pair of meteorological features with a second correlation coefficient greater than a second threshold value; the second threshold value is greater than the first threshold value; determining a third correlation coefficient between each of the pair of meteorological features and the target quality score; taking the feature corresponding to the smaller third correlation coefficient in the pair of meteorological features as the feature to be removed.

5. A device for predicting the quality of satellite imaging, characterized in that it comprises: The device comprises: a to-be-predicted data acquisition module configured to acquire a to-be-predicted data set of a to-be-predicted area; the to-be-predicted data set comprises position information, time information, and meteorological data; a quality evaluation module configured to evaluate the quality of the to-be-predicted data set by using a pre-constructed classification prediction model, so as to determine the data quality level of the to-be-predicted data set; an imaging prediction module configured to input the to-be-predicted data set into a pre-constructed imaging prediction model corresponding to the data quality level, so as to obtain an imaging prediction score; the higher the imaging prediction score is, the higher the imaging suitability of the to-be-predicted area is; a first acquisition module configured to acquire historical remote sensing images and an initial classification prediction model; a training data set construction module configured to construct a training data set according to the historical remote sensing images; the training data set comprises a plurality of sub-data sets; different sub-data sets correspond to different data quality level labels; and the sub-data sets comprise position information, time information, and meteorological data corresponding to historical imaging; an iterative training module configured to perform iterative training on the initial classification prediction model by using the plurality of sub-data sets, so as to obtain the classification prediction model; an initial imaging prediction model construction module configured to construct an initial imaging prediction model; the number of initial imaging prediction models is the same as the number of sub-data sets; the iterative training module is further configured to perform one-to-one iterative training on the initial imaging prediction model by using the sub-data sets, so as to obtain the imaging prediction model; a loss function determination module configured to determine a classification loss function of the initial classification prediction model and a regression loss function of the initial imaging prediction model; a loss function fusion module configured to fuse the classification loss function and the regression loss function, so as to obtain a target loss function; a parameter optimization module configured to perform iterative adjustment on the parameters of the initial classification prediction model and the initial imaging prediction model according to the target loss function, so as to minimize the target loss function.

6. An electronic device comprising a memory and a processor, said memory having stored thereon a computer program operable to run on said processor, characterized in that, The processor executes the computer program to realize the steps of the satellite imaging quality prediction method in any one of claims 1 to 4.

7. A computer readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to realize the steps of the satellite imaging quality prediction method in any one of claims 1 to 4.

Images