Reservoir level and reservoir capacity rechecking method based on multi-source remote sensing image and deep learning

Through multi-source remote sensing imaging and deep learning technology, the efficiency and accuracy of the traditional water level-reservoir volume relationship review method is solved, and efficient and accurate review of the water level and reservoir volume relationship is achieved to adapt to the impact of climate change and human activities on the reservoir.

CN119090672APending Publication Date: 2024-12-06SPACE STAR TECH CO LTD
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202411112968.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-14
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

The traditional water level-reservoir volume relationship review method has a long test cycle, a cumbersome process, time-consuming and labor-intensive, and difficult to update data, making it difficult to effectively respond to the impact of climate change and human activities on reservoirs.

Method used

Using a method based on multi-source remote sensing image and deep learning, by obtaining water level monitoring data and multi-source remote sensing image data, using deep learning technology to eliminate cloud interference, extract water area values, build a relationship between water level and area, and fit water level and storage capacity. Equations and curves.

Benefits of technology

The efficiency of reservoir water level and storage capacity review has been improved, high-precision water level and storage capacity review has been achieved, simplified the data update process, and adapted to environmental changes.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119090672A_ABST
    Figure CN119090672A_ABST
Patent Text Reader

Abstract

The invention provides a reservoir level and storage capacity rechecking method based on a multi-source remote sensing image and deep learning, and the method comprises the steps: obtaining the water level monitoring data and the multi-source remote sensing image data of a target reservoir in a preset time period, and screening out the water level monitoring data and the multi-source remote sensing image data corresponding to the same time; aiming at the screened multi-source remote sensing image data, utilizing a deep learning technology to eliminate cloud interference data, and extracting a water body area value in the image; and constructing an association relationship between the corresponding water level and the water body area value in the image, performing equation and curve fitting on the basis, and finally calculating a water level and reservoir capacity equation and fitting a water level and reservoir capacity curve. Therefore, the efficiency of rechecking the water level and the storage capacity of the reservoir based on the multi-source remote sensing image is improved by utilizing a deep learning technology.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of remote sensing application technology, and in particular to a reservoir water level and storage capacity verification method based on multi-source remote sensing images and deep learning. Background Art

[0002] Reservoirs are an important part of the basin flood control engineering system and water resources allocation system, and are important infrastructure for ensuring national water security. Reservoir capacity refers to the amount of water that a reservoir can hold. It is the basic parameter for reservoir planning and management, and an important guarantee for the reservoir to effectively play its functions such as water resources allocation, flood control, and power generation. The water level-capacity relationship refers to the reservoir capacity value under various water level conditions. It is an important tool in reservoir management and water resources management, and provides strong support and reference for the regulation and operation of reservoirs. Therefore, reviewing the reservoir water level-capacity relationship is a necessary requirement for the effective management of water resources and the safe operation of reservoirs.

[0003] With the influence of climate change and human activities, hydrological conditions have changed. Reservoirs will also be affected by factors such as siltation and land cover changes during long-term operation, resulting in changes in the water level-reservoir capacity relationship. The original design parameters of the reservoir need to be re-evaluated and verified to adapt to the new environmental background. Therefore, regular review of the water level-reservoir capacity relationship can better cope with these changes and ensure the continuous and stable operation of the reservoir.

[0004] The traditional method of verifying the relationship between water level and reservoir capacity mainly uses ground measurement technology to obtain the water level-reservoir capacity curve by measuring the water level, water depth and other parameters of the reservoir on the spot. This method has a long measurement cycle, cumbersome process, time-consuming and labor-intensive, and difficult data update.

[0005] Therefore, there is an urgent need for a method that can achieve efficient and high-precision verification of reservoir water level and storage capacity. Summary of the invention

[0006] The purpose of the present invention is to provide a reservoir water level and storage capacity verification method based on multi-source remote sensing images and deep learning, which effectively improves the efficiency of reservoir water level and storage capacity verification by using deep learning technology.

[0007] In order to achieve the above object, the present invention provides a reservoir water level and storage capacity verification method based on multi-source remote sensing images and deep learning, comprising the steps of:

[0008] Acquire water level monitoring data and multi-source remote sensing image data of a target reservoir within a predetermined time period; wherein the water level monitoring data includes the water level and its acquisition time;

[0009] According to the water level monitoring data, the time information corresponding to each water level is counted;

[0010] Comparing the time information corresponding to each of the water levels with the shooting time of the multi-source remote sensing image data to select the target water level and the first remote sensing image corresponding to the same time;

[0011] Preprocessing the first remote sensing images, and performing image cloud detection on the first remote sensing images by using deep learning technology to obtain the image cloud content of each of the first remote sensing images; wherein the image cloud content represents the proportion of cloud features in the remote sensing image;

[0012] Assessing the quality of the first remote sensing image according to the cloud content of the image to select a second remote sensing image that meets the predetermined quality requirements; and resampling the second remote sensing image to obtain a third remote sensing image with a preset grid accuracy;

[0013] Extracting the water area of ​​the reservoir in the third remote sensing image by using deep learning technology, and counting the water area values ​​corresponding to each of the third remote sensing images based on the water area;

[0014] According to the water body area value at the same water level, the corresponding average water body area value is calculated, and the correlation data between the water level and the area is constructed based on the average water body area value;

[0015] Performing equation and curve fitting on the correlation data, and discretizing the fitted equation;

[0016] The water level and reservoir capacity equation is derived based on the discrete data of the equation, and the water level and reservoir capacity curve is fitted.

[0017] Furthermore, before counting the time information corresponding to each water level according to the water level monitoring data, the method further includes:

[0018] Sorting the water level monitoring data according to water level, and merging the water level monitoring data of the same water level;

[0019] The counting of time information corresponding to each water level according to the water level monitoring data comprises:

[0020] Based on the combined water level monitoring data, time information corresponding to each water level is counted.

[0021] Furthermore, after the water level and reservoir capacity equation is derived according to the discrete data of the equation and the water level and reservoir capacity curve is fitted, the method further includes:

[0022] The water level and reservoir capacity curve is compared with the water level and reservoir capacity historical curve, and the water level and reservoir capacity curve is verified based on the comparison result.

[0023] Furthermore, the step of comparing the time information corresponding to each of the water levels with the shooting time of the multi-source remote sensing image data to select the target water level and the first remote sensing image corresponding to the same time includes:

[0024] Obtaining the shooting time of the multi-source remote sensing image data;

[0025] Comparing the time information corresponding to each of the water levels with the shooting time;

[0026] According to the comparison result, the target water level and the first remote sensing image having the same acquisition time and shooting time are screened out.

[0027] Furthermore, the preprocessing of the first remote sensing image includes:

[0028] The original images of the first remote sensing image are batch-radiated and calibrated, atmospherically corrected, image fused and geometrically precisely corrected using ENVI software.

[0029] Furthermore, the performing of image cloud content detection on the first remote sensing images by using deep learning technology to obtain the image cloud content of each of the first remote sensing images includes:

[0030] Based on the preprocessed first remote sensing image, outlining and marking a first sample characterized by clouds;

[0031] Performing sample cleaning and data enhancement on the first sample to expand the number of samples and form a first sample set;

[0032] Dividing the first sample set into a first training set, a first validation set and a first test set according to a first preset ratio;

[0033] Constructing a cloud detection model by using the U-NET algorithm, training the cloud detection model with the first training set, and minimizing the loss function in the training by using the original gradient descent algorithm to adjust the weights and parameters of the cloud detection model;

[0034] Verifying the accuracy of the cloud detection model using the first verification set, and returning to retrain the cloud detection model if the verification does not meet the preset requirements;

[0035] Testing the cloud detection model using the first test set and outputting the test result;

[0036] Binarizing the first remote sensing image after cloud recognition is completed by the cloud detection model, calculating the cloud content of the first remote sensing image, and counting the cloud content rate of each of the first remote sensing images;

[0037] The cloud content is calculated based on the following formula:

[0038]

[0039] Among them, Cloud rate is the cloud content rate, Cloud pix is the number of pixels occupied by the cloud, Image pix is the total number of pixels in the image.

[0040] Furthermore, the method of extracting the water area of ​​the reservoir in the third remote sensing image by using deep learning technology includes:

[0041] Based on the third remote sensing image, outlining and marking a second sample characterized by a water body;

[0042] Performing sample cleaning, data enhancement, and sample expansion on the second sample to form a second sample set;

[0043] Dividing the second sample set into a second training set, a second validation set and a third test set according to a second preset ratio;

[0044] Constructing a water body detection model by using the DeepLab V3 algorithm, training the water body detection model with the second training set, and minimizing the loss function in the training by using the original gradient descent algorithm to adjust the weights and parameters of the model;

[0045] Performing accuracy verification on the water body detection model using the second verification set, and returning to retrain the water body detection model if the verification does not meet the preset requirements;

[0046] Testing the water body detection model using the second test set and outputting the test result;

[0047] Based on the tested water body detection model, the water body area of ​​the reservoir area in the third remote sensing image is extracted.

[0048] Further, performing equation and curve fitting on the correlation data includes:

[0049] Taking the original water level and area curve as reference, the cubic function relationship between water level and area is obtained by SPSS software analysis; wherein, the original water level and area curve is a relationship curve based on historical data analysis;

[0050] Using the SPSS software to fit the correlation data to determine the parameters in the cubic function relationship;

[0051] The coefficient of determination is used as a regression evaluation index to perform regression verification on the fitted data equation. If the verification does not meet the requirements, the process of fitting the associated data using the SPSS software is returned to be re-executed to determine the parameters in the cubic function relationship until the verification meets the requirements.

[0052] Furthermore, discretizing the fitted equation includes:

[0053] The target water level is discretized and valued from the lowest water level in increments of 0.05 m per level to obtain a discrete value of the water level;

[0054] The water area corresponding to the discrete value of the water level is back-calculated through the fitted equation to complete the equation discretization.

[0055] Furthermore, the step of deducing the water level and reservoir capacity equation according to the discrete data of the equation and fitting the water level and reservoir capacity curve comprises:

[0056] Sort the discrete data of the equation from low to high according to the water level;

[0057] Based on the sorted discrete data, reservoir capacity calculation is performed to obtain the reservoir capacity corresponding to each water level of the discrete water level value;

[0058] The calculated storage capacity and its water level relationship are fitted with a cubic polynomial relationship by SPSS software to obtain a water level and storage capacity equation, a water level and storage capacity curve, and a determination coefficient; if the determination coefficient does not meet the preset requirements, it is returned for refitting;

[0059] Wherein, the storage capacity is calculated based on the following formula:

[0060]

[0061] Where ΔV is the volume of the water layer between two adjacent contour lines; s′ 下 , S′ 上 represents the reservoir water area contained in each of the adjacent water levels; ΔZ′ represents the height difference between adjacent water levels; Z min is the lowest water level in the discrete data.

[0062] The reservoir water level and storage capacity verification method based on multi-source remote sensing images and deep learning described in the present invention obtains the water level monitoring data and multi-source remote sensing image data of the target reservoir within a predetermined time period, and screens out the water level monitoring data and multi-source remote sensing image data corresponding to the same time, and uses deep learning technology to eliminate cloud interference data for the screened multi-source remote sensing image data, and extracts the water area value in the image; constructs a correlation between the corresponding water level and the water area value in the image, and uses this to perform equation and curve fitting, and finally derives the water level and storage capacity equation and fits the water level and storage capacity curve. The present invention uses deep learning technology to improve the efficiency of reservoir water level and storage capacity verification based on multi-source remote sensing images. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 A flowchart of the steps of the reservoir water level and storage capacity verification method based on multi-source remote sensing images and deep learning provided in one embodiment of the present invention;

[0064] Figure 2 A specific flow chart of the reservoir water level and storage capacity verification method based on multi-source remote sensing images and deep learning provided in one embodiment of the present invention for screening water level data and images at the same time;

[0065] Figure 3 A specific flow chart of the reservoir water level and storage capacity verification method based on multi-source remote sensing images and deep learning provided in one embodiment of the present invention, which uses deep learning technology to extract the cloud content of images;

[0066] Figure 4 A specific flow chart of the reservoir water level and storage capacity verification method based on multi-source remote sensing images and deep learning provided in one embodiment of the present invention, which uses deep learning technology to extract the water area of ​​the image;

[0067] Figure 5 A specific flow chart of the reservoir water level and storage capacity verification method based on multi-source remote sensing images and deep learning provided in one embodiment of the present invention for use in water level and water body area equation and curve fitting;

[0068] Figure 6 A specific flow chart of the reservoir water level and storage capacity verification method based on multi-source remote sensing images and deep learning provided in one embodiment of the present invention for use in water level and storage capacity equation and curve fitting;

[0069] Figure 7 A schematic diagram of the overall process of the reservoir water level and storage capacity verification method based on multi-source remote sensing images and deep learning provided in one embodiment of the present invention. DETAILED DESCRIPTION

[0070] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0071] It should be noted that references to "one embodiment", "embodiment", "example embodiment", etc. in this specification refer to the embodiment described, which may include specific features, structures or characteristics, but not every embodiment must include these specific features, structures or characteristics. In addition, such expressions do not refer to the same embodiment. Furthermore, when describing specific features, structures or characteristics in conjunction with an embodiment, whether or not there is an explicit description, it has been shown that it is within the knowledge of those skilled in the art to combine such features, structures or characteristics into other embodiments.

[0072] In addition, certain words are used in the specification and subsequent claims to refer to specific components or parts. Those with ordinary knowledge in the relevant field should understand that manufacturers can use different nouns or terms to refer to the same component or part. This specification and subsequent claims do not use differences in names as a way to distinguish components or parts, but use differences in the functions of components or parts as the criteria for distinction. "Including" and "including" mentioned throughout the specification and subsequent claims are open-ended terms and should be interpreted as "including but not limited to". In addition, the word "connected" here includes any direct and indirect electrical connection means. Indirect electrical connection means include connection through other devices.

[0073] In order to solve the problems of long measurement cycle, cumbersome process, time-consuming and labor-intensive, and difficult data updating in traditional water level-reservoir capacity relationship verification methods, the present invention provides a reservoir water level and reservoir capacity verification method based on multi-source remote sensing images and deep learning, which specifically uses deep learning technology to analyze multi-source remote sensing images, thereby verifying the changes in the relationship between the water level of the target reservoir and its reservoir capacity.

[0074] Figure 1 A reservoir water level and storage capacity verification method based on multi-source remote sensing images and deep learning provided by an embodiment of the present invention is shown, and the method includes the following steps:

[0075] S101: Acquire the water level monitoring data and multi-source remote sensing image data of the target reservoir within a predetermined time period; the preset time period can be set according to the specific situation, such as within one year, two years or three years; the preset time period of this embodiment is preferably within one year, that is, in step S101, the water level monitoring data and multi-source remote sensing image data of the target reservoir within one year are acquired; in the specific implementation process, the water level monitoring data and the multi-source remote sensing image data are acquired through different channels respectively, the water level monitoring data can be specifically acquired from the recorded data of the target reservoir, and the multi-source remote sensing image data comes from satellites; the water level monitoring data of this embodiment includes water level Z i and its acquisition time T i , i is the number of the corresponding water level.

[0076] Since the same water level may have multiple different collection times at the same time, that is, different collection times may have the same water level record; in this regard, it is preferred that before step S102, the following steps are further included:

[0077] The water level monitoring data are sorted according to the water level, and the water level monitoring data of the same water level are merged; specifically, the acquired water level monitoring data are sorted in order from low to high, and the data of the same water level are merged; for example, the sorting result is: the lowest water level monitoring data, the second lowest water level monitoring data... the second highest water level monitoring data, the highest water level monitoring data; each water level monitoring data after sorting corresponds to one or more collection times, such as the merged data is Z i '-T i , at this time a water level Z i 'May correspond to multiple times T i , respectively denoted as T i1 ,…,T in .

[0078] S102: According to the water level monitoring data, the time information corresponding to each water level is counted; that is, the corresponding collection time is counted and sorted into the time information of the corresponding water level from the water level monitoring data; the time information records the collection time of each corresponding water level. In combination with the above embodiment, step S102 specifically includes: based on the water level monitoring data after the merging process, the time information corresponding to each water level is counted.

[0079] S103: Compare the time information corresponding to each water level with the shooting time of the multi-source remote sensing image data to select the target water level and the first remote sensing image corresponding to the same time. That is, in this step S103, the time information of the water level counted in the above step S102 is matched with the shooting time of the multi-source remote sensing image data to determine whether there are water levels and multi-source remote sensing images recorded at the same time. If there are the same time, the corresponding target water level and the first remote sensing image are output; at the same time, in this step S103, if the time information of a certain water level does not match the shooting time of the multi-source remote sensing image, or the shooting time of a certain multi-source remote sensing image does not match the time information of any water level, these data are regarded as invalid data and are preferentially eliminated; through step S103, only the water level records and the first remote sensing image with the same time can be retained.

[0080] See also Figure 2 In an optional implementation, step S103 includes:

[0081] S1031: Obtain the shooting time of the multi-source remote sensing image data; specifically, the shooting time of each multi-source remote sensing image data is counted and recorded as T ri .

[0082] S1032: Compare the time information corresponding to each water level with the shooting time; respectively compare the time information T corresponding to each water level in and the shooting time T of each multi-source remote sensing image data ri The time precision of the comparison is minute level; that is, when the compared times are the same in minute precision, the two times are considered to have the same time, and the difference of seconds below minutes is negligible.

[0083] S1033: According to the comparison result, the target water level and the first remote sensing image with the same acquisition time and shooting time are selected. in =T ri , then the time constraint is satisfied if T in ≠T ri , the time constraint is not met; and some data (including image data and water level monitoring data) that do not meet the time constraint will be discarded.

[0084] Although the target water level and the first remote sensing image with the same time are screened out through the above steps, the first remote sensing image can be used to reflect the actual picture image of the reservoir water surface under the water level record at this time. However, the first remote sensing image at this time may have noise interference, such as cloud feature interference in cloudy climate; therefore, the interference data caused by cloud features needs to be eliminated in subsequent steps.

[0085] S104: Preprocess the first remote sensing images, and perform image cloud detection on the first remote sensing images using deep learning technology to obtain the image cloud content of each first remote sensing image; wherein the image cloud content represents the proportion of cloud features in the remote sensing image.

[0086] In specific implementation, the preprocessing of the first remote sensing image includes: performing radiation calibration, atmospheric correction, image fusion and geometric precision correction on the original images of the first remote sensing image in batches through ENVI (The Environment for Visualizing Images, remote sensing image processing platform) software to complete the preprocessing of the image.

[0087] Then, deep learning technology is used to detect image cloud cover on the preprocessed first remote sensing image.

[0088] See also Figure 3 In an optional implementation, the performing of image cloud content detection on the first remote sensing image by using deep learning technology to obtain the image cloud content of each of the first remote sensing images includes:

[0089] S1041: based on the preprocessed first remote sensing image, outlining and labeling a first sample featuring clouds; that is, in the preprocessed first remote sensing image, outlining and labeling samples featuring clouds; of course, in other implementations, samples may also be outlined and labeled from images from other sources, or existing cloud feature sample data may be directly used.

[0090] S1042: Perform sample cleaning and data enhancement on the first sample to expand the number of samples and form a first sample set; specifically, after obtaining the first sample, perform sample cleaning on it, and then use methods such as rotation, flipping, mirroring and reverse data enhancement to expand the sample, and finally form the first sample set.

[0091] S1043: Divide the first sample set into a first training set, a first validation set and a first test set according to a first preset ratio; the first preset ratio can be set arbitrarily, but for model training, it is necessary to ensure that the first training set occupies a larger proportion; the preferred first preset ratio of this embodiment is 7:2:1, that is, this embodiment randomly extracts seven-tenths of the data from the first sample set as the first training set, randomly extracts two-tenths of the data as the first validation set, and the remaining data is used as the first test set.

[0092] S1044: constructing a cloud detection model by using a U-NET algorithm, training the cloud detection model with a first training set, and minimizing a loss function in training by using an original gradient descent algorithm to adjust weights and parameters of the cloud detection model; wherein the accuracy of the cloud detection model is evaluated by using the loss function in training;

[0093] S1045: Verify the accuracy of the cloud detection model through the first verification set. If the verification does not meet the preset requirements, return to retrain the cloud detection model; that is, use the first verification set to verify the accuracy of the model and parameters to determine whether the relevant accuracy meets the preset requirements. If so, the verification is qualified, otherwise the verification is unqualified; when the verification fails, return to repeat steps S1044 to S1045 until the verification is qualified.

[0094] S1046: Testing the cloud detection model through the first test set and outputting the test results; detecting the accuracy of model recognition through the test results, and further determining whether the model needs to be further trained or the model parameters need to be adjusted based on the test results.

[0095] S1047: Binarize the first remote sensing image that has completed cloud recognition through the cloud detection model, calculate the cloud content of the first remote sensing image, and count the cloud content rate of each first remote sensing image; the cloud recognition through the cloud detection model specifically refers to taking the first remote sensing image as a model input and inputting it into the above-mentioned trained and qualified cloud detection model to obtain the corresponding output result (i.e., cloud recognition result); then binarize the first remote sensing image that has completed cloud recognition, and calculate the cloud content rate corresponding to the image from the binarized image.

[0096] Furthermore, the cloud content in this embodiment is calculated based on the following formula:

[0097]

[0098] Among them, Cloud rate is the cloud content rate, Cloud pix is the number of pixels occupied by the cloud, Image pix is the total number of pixels in the image.

[0099] S105: Assess the quality of the first remote sensing image according to the cloud content of the image to select a second remote sensing image that meets the predetermined quality requirements; and resample the second remote sensing image to obtain a third remote sensing image with a preset grid accuracy; in this embodiment, the quality assessment of the first remote sensing image is specifically to determine whether the cloud content in the first remote sensing image is less than 5%. If the cloud content in an image is less than 5%, it means that the image quality is qualified and can be used for subsequent analysis; conversely, if the cloud content in an image is higher than 5%, it means that the image is greatly interfered by clouds and is not suitable for subsequent analysis. Therefore, in this embodiment, it is necessary to remove the image data with a cloud content higher than 5% and its corresponding water level monitoring data, so as to retain the image with qualified image quality as the second remote sensing image; and further resample the second remote sensing image to obtain the third remote sensing image with a preset grid accuracy.

[0100] During specific implementation, ENVI software may be used to resample the second remote sensing image to obtain a third remote sensing image with a grid accuracy of 0.005 m; of course, this embodiment does not limit the value of the preset grid accuracy, which may be adjusted according to specific application conditions.

[0101] S106: Using deep learning technology to extract the water area of ​​the reservoir in the third remote sensing image, and based on the water area, respectively count the water area values ​​corresponding to each third remote sensing image. The water area refers to the area of ​​the target reservoir water surface in the image. This embodiment uses deep learning technology to obtain a detection model for water body features through model training, automatically identifies and extracts the water area in the third remote sensing image; and then calculates the corresponding water area value for the extracted water area image.

[0102] See also Figure 4 In an optional implementation, the extracting of the water area of ​​the reservoir area in the third remote sensing image by using deep learning technology includes:

[0103] S1061: Based on the third remote sensing image, outline and mark the second sample with water body as the feature; that is, from a number of third remote sensing images, outline and mark the second sample with water body as the feature; of course, in other implementations, samples can also be outlined and marked from images from other sources, or existing sample data of water body features can be directly used.

[0104] S1062: performing sample cleaning, data enhancement and sample expansion on the second sample to form a second sample set; specifically, after obtaining the second sample, performing sample cleaning on it, and then using methods such as rotation, flipping, mirroring and reverse data enhancement, and sample expansion to finally form the second sample set.

[0105] S1063: Divide the second sample set into a second training set, a second validation set and a third test set according to a second preset ratio; the second preset ratio can also be set arbitrarily, but for model training, it is necessary to ensure that the second training set occupies a larger proportion; in this regard, the preferred second preset ratio of this embodiment is 7:2:1, that is, this embodiment randomly extracts seven-tenths of the data from the second sample set as the second training set, randomly extracts two-tenths of the data as the second validation set, and the remaining data is used as the second test set.

[0106] S1064: Construct a water body detection model through the DeepLab V3 algorithm, and train the water body detection model with the second training set, and minimize the loss function in the training through the original gradient descent algorithm to adjust the weights and parameters of the model; wherein, the loss function in the training is used to evaluate its accuracy, and the original gradient descent algorithm is used to minimize the loss function to adjust the weights and parameters of the model. The original gradient descent algorithm is a prior art and will not be repeated here.

[0107] S1065: Verify the accuracy of the water body detection model through the second verification set. If the verification does not meet the preset requirements, return to retrain the water body detection model; that is, use the second verification set to verify the accuracy of the model and parameters to determine whether the relevant accuracy meets the preset requirements. If so, the verification is qualified, otherwise the verification is unqualified; when the verification fails, return to repeat steps S1064 to S1065 until the verification is qualified.

[0108] S1066: Testing the water body detection model through a second test set and outputting the test results; detecting the accuracy of model recognition through the test results, and further determining whether the model needs to be further trained or model parameters need to be adjusted based on the test results.

[0109] S1067: Based on the water body detection model that has completed the test, extract the water body area of ​​the reservoir in the third remote sensing image. After verification and completion of the test, the third remote sensing image is input into the trained water body detection model to detect and identify the water body area in the image through the water body detection model, and output the result information; wherein the result information may include the recognition result of the water body area in the image, the proportion of the water body area occupying the image screen and / or the proportion of the number of pixels of the water body area in the image.

[0110] Furthermore, after identifying the water area in the third remote sensing image, the water area value corresponding to the water area is calculated, which is recorded as S i ; The method for calculating the area of ​​a certain region based on remote sensing images can adopt existing remote sensing detection technology, which will not be described here.

[0111] S107: According to the water body area value of the same water level, the corresponding average water body area value is calculated, and the correlation data between the water level and the area is constructed based on the average water body area value; after acquiring the water body area from the third remote sensing image and calculating the corresponding water body area value, this embodiment further combines the water level record corresponding to the third remote sensing image (the water level record is specifically obtained from the water level monitoring data with the same acquisition time as the image), sorts the water level and its corresponding water body area, and then calculates the water body area value.

[0112] In this embodiment, the average water area corresponding to the same water level is calculated, which is denoted as S i '; It can be calculated based on the following formula:

[0113]

[0114] Among them, S′ i is the average water area corresponding to the same water level i; T in is the shooting time of the third remote sensing image corresponding to water level i; S i is the water area corresponding to each shooting time.

[0115] After obtaining the average water area value corresponding to the water level, the water level record is associated and bound with the corresponding water area average value to generate a corresponding associated data, which is the water level-area data; that is, each water level record has a unique associated data, and the associated data forms a unique correspondence between the water level and the area.

[0116] S108: performing equation and curve fitting on the associated data, and discretizing the fitted equation; in combination with the above description, it can be seen that the associated data reflects the unique corresponding correlation between the water level and the average value of the water body area, so equation and curve fitting can be performed based on this to obtain the corresponding relationship equation between the water level and the area; although the fitted equation is discretized to obtain a number of discrete data.

[0117] See also Figure 5 In an optional implementation, performing equation and curve fitting on the correlation data includes:

[0118] S1081: With reference to the original water level and area curve, the cubic function relationship between water level and area is obtained through SPSS software analysis; wherein, the original water level and area curve is a relationship curve based on historical data analysis; before this, the associated data needs to be input into the SPSS software, that is, the associated data of the water level and its unique corresponding water area need to be input into the SPSS software; the data is analyzed, and with reference to the original water level-area curve, the reservoir water level Z' i - Area S' i The relationship between them satisfies the cubic function relationship, that is:

[0119] S' i =AZ' i 3 +BZ' i 2 +CZ' i +D

[0120] Among them, S i ' is the water area; Z i' is the water level; A, B, C, D are all fitting parameters.

[0121] S1082: Use SPSS software to fit the associated data to determine the parameters in the cubic function relationship; that is, use SPSS software to fit the associated data to determine the fitting parameters A, B, C, D in the above formula, so as to determine the cubic function equation of water level-area.

[0122] S1083: Using the coefficient of determination as a regression evaluation index, perform regression verification on the fitted data equation. If the verification does not meet the requirements, return to re-execute the use of SPSS software to fit the associated data to determine the parameters in the cubic function relationship until the verification meets the requirements. After the fitting parameters are obtained in the above steps, the data equation is further regressed. In this embodiment, the coefficient of determination R is specifically used. 2 To determine the fitting effect, preferably, if R 2 >0.95, it means that the fitting effect meets the requirements, otherwise it returns to repeat step S1082; where the determination coefficient R 2 The calculation formula is as follows:

[0123]

[0124] in, represents the sum of squared differences between the true value and the predicted value. It represents the sum of squared differences between the true value and the mean value.

[0125] Furthermore, in this embodiment, discretizing the fitted equation includes:

[0126] The target water level is discretized and taken from the lowest water level in increments of 0.05m per level to obtain a water level discrete value; the water area corresponding to the water level discrete value is back-calculated through the fitted equation to complete the equation discretization. In specific implementation, for the water level with the same time as the image shooting time, that is, the target water level obtained in the above step S103, the water level is discretized and taken from the lowest water level in increments of 0.05m, and the maximum value does not exceed the highest water level, that is, Z0=Z min , Z i+1 =Z i +0.05; in the water level-area equation completed by the above fitting, that is, S' i =AX' i 3 +BZ' i 2 +CZ' i +D; reverse calculation Z' i The corresponding S' i, the equation can be discretized, so as to obtain the corresponding discrete data in the above equation.

[0127] S109: Derivation of a water level and reservoir capacity equation based on the discrete data of the equation, and fitting of a water level and reservoir capacity curve.

[0128] See also Figure 6 In an optional implementation, step S109 includes:

[0129] S1091: Sort the discrete data of the equation from low to high according to the water level. The discrete data specifically includes the water level Z calculated by inverse calculation. i and its corresponding water area S i ', according to the water level Z i Sort from low to high.

[0130] S1092: Calculate the reservoir capacity based on the sorted discrete data to obtain the reservoir capacity corresponding to each water level of the discrete water level value.

[0131] Wherein, the storage capacity is calculated based on the following formula:

[0132]

[0133] Among them, ΔV is the volume of the water layer between two adjacent contour lines; S′ 下 , S′ 上 represents the reservoir water area contained in each of the adjacent water levels; ΔZ′ represents the height difference between adjacent water levels; Z min is the lowest water level in the discrete data.

[0134] In this embodiment, the water level Z at the bottom min By adding up layer by layer, we can find the reservoir volume V at each water level Z'.

[0135] S1093: Perform cubic polynomial fitting on the calculated reservoir capacity and its water level relationship through SPSS software to obtain the water level and reservoir capacity equation, water level and reservoir capacity curve and determination coefficient; if the determination coefficient does not meet the preset requirements, return to refit; where the determination coefficient R 2 The calculation of is as shown above and will not be repeated here; if the determination coefficient R 2 >0.95, the fitting effect meets the requirements. If not, re-fitting is performed, that is, step S1093 is repeated until the determination coefficient R 2 Until the preset requirements are met, the water level and reservoir capacity equation and its water level and reservoir capacity curve for verifying the relationship between water level and reservoir capacity can be obtained.

[0136] This embodiment uses deep learning technology to automatically obtain the water level and reservoir capacity equation and its fitting curve from multi-source remote sensing images and water level monitoring data, thereby improving the efficiency of reservoir water level and reservoir capacity review based on multi-source remote sensing images.

[0137] Preferably, after step S109, the method further includes: comparing the water level and reservoir capacity curve with the historical water level and reservoir capacity curve, and verifying the water level and reservoir capacity curve based on the comparison result. Specifically, the water level and reservoir capacity curve fitted in step S109 is compared with the old water level and reservoir capacity curve recorded in history, and judging whether the error is within a reasonable range from the comparison result; this is equivalent to comparing the shapes of the new and old curves, thereby verifying the water level and reservoir capacity curve fitted in step S109.

[0138] Figure 7 The overall step flow of the reservoir water level and storage capacity verification method based on multi-source remote sensing images and deep learning preferred in the embodiment of the present invention is shown, and the specific implementation process is as follows:

[0139] Step 1: Collect and obtain the water level monitoring data of the target reservoir within one year, including the water level Z i And the acquisition time T i .

[0140] Step 2, based on step 1, sort the collected data according to the water level, and merge the same water level; further, in step 2, specifically including: step 2.1, sort the acquired water level-time data in order from low to high water level; step 2.2, merge the same water level, and the merged data is Z i '-T i , here a water level Z i 'May correspond to multiple times T i , respectively denoted as T i1 ,……,T in .

[0141] Step 3, counting the time corresponding to each water level in step 2.

[0142] Step 4: Obtain multi-source remote sensing image data of the target area within one year.

[0143] Step 5, compare the corresponding time of each water level in step 3 with the shooting time of the remote sensing image in step 4, and select the same time between the two; further, in step 5, specifically including: step 5.1, count the shooting time of the remote sensing image in step 4, and record it as T ri ; Step 5.2, the water level in step 2.2 corresponds to time T in The shooting time T of the remote sensing image in step 5.1 ri Compare the time with the accuracy of minutes.

[0144] Step 5.3, based on step 5.2, if T ri =T in , then the time constraint is satisfied if T ri ≠T in , then the time constraint is not met, and this part of the data, including the image and water level, is discarded.

[0145] Step 6: Based on step 5, the target water level and remote sensing images corresponding to the same time are counted.

[0146] Step 7. Based on step 6, pre-process the remote sensing images that meet the time requirements: use ENVI software to perform radiation calibration, atmospheric correction, image fusion and geometric precision correction on the original images in batches to complete the image pre-processing.

[0147] Step 8: Based on step 7, the cloud content of the images that meet the time requirements is detected based on deep learning technology to obtain the cloud content of each image;

[0148] Furthermore, in step 8, it specifically includes:

[0149] Step 8.1, based on the image preprocessed in step 7, outline and annotate the sample with cloud as the feature;

[0150] Step 8.2: Based on step 8.1, the samples are cleaned, and then rotation, flipping, mirroring and other methods as well as retrograde data enhancement are used to expand the samples, and finally a sample set is formed;

[0151] Step 8.3: Based on step 8.2, divide the sample set into training set, validation set and test set in a ratio of 7:2:1;

[0152] Step 8.4, using the U-NET algorithm to build a cloud detection model, and training the training set in step 8.2, using the loss function in training to evaluate its accuracy, and minimizing the loss function through the original gradient descent algorithm to adjust the weights and parameters of the model;

[0153] Step 8.5: Use the validation set to verify the accuracy of the model and parameters. If the verification is qualified, test it. If it is not qualified, repeat step 8.4;

[0154] Step 8.6, test the model using the test set in step 8.2 and output the test results.

[0155] Step 8.7, binarize the sample images that have completed cloud recognition, calculate the cloud content according to formula (1), and count the cloud content rate of each image.

[0156]

[0157] in:

[0158] Cloud rate : cloud content;

[0159] Cloud pix : The number of pixels occupied by clouds;

[0160] Image pix : Total number of pixels in the image.

[0161] Step 9: According to the cloud content of the image obtained in step 8, the quality of each image data is evaluated, and the image data with a cloud content higher than 5% and its corresponding water level data are removed.

[0162] Step 10: Use ENVI software to resample the remote sensing image in step 9 to obtain an image with a grid accuracy of 0.005m.

[0163] Step 11, based on step 10, using deep learning technology to extract the water body range of the reservoir area;

[0164] Furthermore, in step 11, it specifically includes:

[0165] Step 11.1, based on the image processed in step 1, outlining and labeling samples with water bodies as features;

[0166] Step 11.2: Based on step 11.1, perform sample cleaning, data enhancement and sample expansion to form a sample set;

[0167] Step 11.3: Based on step 11.2, divide the sample set into training set, validation set and test set in a ratio of 7:2:1;

[0168] Step 11.4, using the DeepLab V3 algorithm to build a water body detection model, and train the training set in step 11.2, and use the loss function in training to evaluate its accuracy, and adjust the weights and parameters of the model by minimizing the loss function through the original gradient descent algorithm;

[0169] Step 11.5, use the validation set to verify the accuracy of the model and parameters. If the verification is qualified, test it. If it is not qualified, repeat step 11.4;

[0170] Step 11.6, test the model using the test set in step 11.2 and output the test results.

[0171] Step 12: Binarize the image extracted in step 11 and calculate the water area value S i .

[0172] Step 13, sorting the water level and the corresponding water area in step 6, finding the average of the water area corresponding to the same water level, and forming a unique correspondence between water level and area;

[0173] Find the average water area corresponding to the same water level, denoted as S i ', specifically calculated according to Formula 2:

[0174]

[0175] in:

[0176] S i ': The average water area corresponding to the same water level;

[0177] T in : The time corresponding to the water level;

[0178] S i : The water area corresponding to each time.

[0179] Step 14, based on step 13, using SPSS software to perform equation and curve fitting on the water level-area data;

[0180] Furthermore, in step 14, it specifically includes:

[0181] Step 14.1, input the water level and the corresponding water area into SPSS software;

[0182] Step 14.2, analyze the data, refer to the original water level-area curve, reservoir water level Z i '- Area S i 'The relationship should satisfy the cubic function relationship, that is:

[0183] S' i =AZ' i 3 +BZ' i 2 +CZ' i +D (3)

[0184] in:

[0185] S i ': water body area;

[0186] Z i ': water level;

[0187] A, B, C, D: fitting parameters;

[0188] Step 14.3, use SPSS software to fit the data to determine the four parameters A, B, C, and D;

[0189] Step 14.4: Based on step 14.3, perform regression verification on the fitted data equation and use the determination coefficient R 2 To determine the fitting effect, R 2 Calculate according to formula 4, if R 2 >0.95, the fitting effect meets the requirements, otherwise repeat step 14.3.

[0190]

[0191] Step 15, discretizing the water level-area equation fitted in step 14;

[0192] Furthermore, in step 15, it specifically includes:

[0193] Step 15.1: For the water level data in step 6, discretize and take values ​​according to 0.05m starting from the lowest water level, and the maximum value does not exceed the highest water level, that is, Z0 = Z min ; Z i+1 =Z i +0.05;

[0194] Step 15.2, use the equation fitted in step 14 to back-calculate Z i The corresponding S i ', complete the discretization of the equation.

[0195] Step 16, deriving the water level-storage capacity equation based on step 15, and fitting the water level-storage capacity curve;

[0196] Furthermore, in step 16, it specifically includes:

[0197] Step 16.1, the water level Z generated in step 15 i and area S i 'Sort by water level from low to high;

[0198] Step 16.2, based on step 16.1, calculate the storage capacity, the equation is shown in formula 5;

[0199]

[0200]

[0201] in:

[0202] S′ 下 , S′ 上 : The water surface area of ​​the reservoir contained between adjacent water levels;

[0203] ΔZ′: height difference between two water levels;

[0204] From the bottom Z minBy adding up layer by layer, the reservoir volume V at each water level Z' can be obtained.

[0205] Step 16.3: Use SPSS software to perform cubic polynomial fitting on the water level-storage capacity relationship to obtain the water level-storage capacity equation, water level-storage capacity relationship curve and determination coefficient R 2 , if R 2 >0.95, the fitting effect meets the requirements; if not, re-fit.

[0206] Step 17, compare the new and old curves, and verify the new curve.

[0207] This embodiment can effectively improve the efficiency of reservoir water level and storage capacity verification based on multi-source remote sensing images.

[0208] It should be noted that the present application can be implemented in software and / or a combination of software and hardware, for example, can be implemented using an application specific integrated circuit (ASIC), a general purpose computer or any other similar hardware device. In one embodiment, the software program of the present application can be executed by a processor to implement the above steps or functions. Similarly, the software program of the present application (including relevant data structures) can be stored in a computer-readable recording medium, for example, a RAM memory, a magnetic or optical drive or a floppy disk and similar devices. In addition, some steps or functions of the present application can be implemented using hardware, for example, as a circuit that cooperates with a processor to perform each step or function.

[0209] The method according to the present invention can be implemented on a computer as a computer-implemented method, or implemented in dedicated hardware, or implemented in a combination of the two. The executable code or part thereof for the method according to the present invention can be stored on a computer program product. The example of a computer program product includes a memory device, an optical storage device, an integrated circuit, a server, online software, etc. Preferably, the computer program product includes a non-temporary program code component stored on a computer-readable medium so as to perform the method according to the present invention when the program product is executed on a computer.

[0210] In a preferred embodiment, the computer program comprises computer program code means adapted to perform all the steps of the method according to the invention when the computer program is run on a computer.Preferably, the computer program is embodied on a computer readable medium.

[0211] In summary, the reservoir water level and storage capacity verification method based on multi-source remote sensing images and deep learning described in the present invention obtains the water level monitoring data and multi-source remote sensing image data of the target reservoir within a predetermined time period, and screens out the water level monitoring data and multi-source remote sensing image data corresponding to the same time, and uses deep learning technology to eliminate cloud interference data for the screened multi-source remote sensing image data, and extracts the water area value in the image; constructs a correlation between the corresponding water level and the water area value in the image, and uses this to perform equation and curve fitting, and finally derives the water level and storage capacity equation and fits the water level and storage capacity curve. In this way, the present invention uses deep learning technology to improve the efficiency of reservoir water level and storage capacity verification based on multi-source remote sensing images.

[0212] Of course, the present invention may have many other embodiments. Without departing from the spirit and essence of the present invention, those skilled in the art may make various corresponding changes and modifications based on the present invention, but these corresponding changes and modifications should all fall within the scope of protection of the claims attached to the present invention.

Claims

1. A reservoir water level and storage capacity verification method based on multi-source remote sensing images and deep learning, It is characterized in that Includes steps: Acquire water level monitoring data and multi-source remote sensing image data of a target reservoir within a predetermined time period; wherein the water level monitoring data includes the water level and its acquisition time; According to the water level monitoring data, the time information corresponding to each water level is counted; Comparing the time information corresponding to each of the water levels with the shooting time of the multi-source remote sensing image data to select the target water level and the first remote sensing image corresponding to the same time; Preprocessing the first remote sensing images, and performing image cloud detection on the first remote sensing images by using deep learning technology to obtain the image cloud content of each of the first remote sensing images; wherein the image cloud content represents the proportion of cloud features in the remote sensing image; Assessing the quality of the first remote sensing image according to the cloud content of the image to select a second remote sensing image that meets the predetermined quality requirements; and resampling the second remote sensing image to obtain a third remote sensing image with a preset grid accuracy; Extracting the water area of ​​the reservoir in the third remote sensing image by using deep learning technology, and counting the water area values ​​corresponding to each of the third remote sensing images based on the water area; According to the water body area value at the same water level, the corresponding average water body area value is calculated, and the correlation data between the water level and the area is constructed based on the average water body area value; Performing equation and curve fitting on the correlation data, and discretizing the fitted equation; The water level and reservoir capacity equation is derived based on the discrete data of the equation, and the water level and reservoir capacity curve is fitted.

2. The reservoir water level and storage capacity verification method based on multi-source remote sensing images and deep learning according to claim 1, It is characterized in that Before counting the time information corresponding to each water level according to the water level monitoring data, the method further includes: Sorting the water level monitoring data according to water level, and merging the water level monitoring data of the same water level; The counting of time information corresponding to each water level according to the water level monitoring data comprises: Based on the combined water level monitoring data, time information corresponding to each water level is counted.

3. The reservoir water level and storage capacity verification method based on multi-source remote sensing images and deep learning according to claim 1, It is characterized in that After the water level and reservoir capacity equation is derived according to the discrete data of the equation and the water level and reservoir capacity curve is fitted, the method further includes: The water level and reservoir capacity curve is compared with the water level and reservoir capacity historical curve, and the water level and reservoir capacity curve is verified based on the comparison result.

4. The reservoir water level and storage capacity verification method based on multi-source remote sensing images and deep learning according to claim 1, It is characterized in that The step of comparing the time information corresponding to each of the water levels with the shooting time of the multi-source remote sensing image data to select the target water level and the first remote sensing image corresponding to the same time includes: Obtaining the shooting time of the multi-source remote sensing image data; Comparing the time information corresponding to each of the water levels with the shooting time; According to the comparison result, the target water level and the first remote sensing image having the same acquisition time and shooting time are screened out.

5. The reservoir water level and storage capacity verification method based on multi-source remote sensing images and deep learning according to claim 1, It is characterized in that The preprocessing of the first remote sensing image comprises: The original images of the first remote sensing image are batch-radiated and calibrated, atmospherically corrected, image fused and geometrically precisely corrected using ENVI software.

6. The reservoir water level and storage capacity verification method based on multi-source remote sensing images and deep learning according to claim 1, It is characterized in that The performing of image cloud content detection on the first remote sensing images by using the deep learning technology to obtain the image cloud content of each of the first remote sensing images includes: Based on the preprocessed first remote sensing image, outlining and marking a first sample characterized by clouds; Performing sample cleaning and data enhancement on the first sample to expand the number of samples and form a first sample set; Dividing the first sample set into a first training set, a first validation set and a first test set according to a first preset ratio; Constructing a cloud detection model by using the U-NET algorithm, training the cloud detection model with the first training set, and minimizing the loss function in the training by using the original gradient descent algorithm to adjust the weights and parameters of the cloud detection model; Verifying the accuracy of the cloud detection model using the first verification set, and returning to retrain the cloud detection model if the verification does not meet the preset requirements; Testing the cloud detection model using the first test set and outputting the test result; Binarizing the first remote sensing image after cloud recognition is completed by the cloud detection model, calculating the cloud content of the first remote sensing image, and counting the cloud content rate of each of the first remote sensing images; The cloud content is calculated based on the following formula: Among them, Cloud rate is the cloud content rate, Cloud pix is the number of pixels occupied by the cloud, Image pix is the total number of pixels in the image.

7. The reservoir water level and storage capacity verification method based on multi-source remote sensing images and deep learning according to claim 1, It is characterized in that The method of extracting the water area of ​​the reservoir in the third remote sensing image by using deep learning technology includes: Based on the third remote sensing image, outlining and marking a second sample characterized by a water body; Performing sample cleaning, data enhancement, and sample expansion on the second sample to form a second sample set; Dividing the second sample set into a second training set, a second validation set and a third test set according to a second preset ratio; Constructing a water body detection model by using the DeepLab V3 algorithm, training the water body detection model with the second training set, and minimizing the loss function in the training by using the original gradient descent algorithm to adjust the weights and parameters of the model; Performing accuracy verification on the water body detection model using the second verification set, and returning to retrain the water body detection model if the verification does not meet the preset requirements; Testing the water body detection model using the second test set and outputting the test result; Based on the tested water body detection model, the water body area of ​​the reservoir area in the third remote sensing image is extracted.

8. The reservoir water level and storage capacity verification method based on multi-source remote sensing images and deep learning according to claim 1, It is characterized in that The performing equation and curve fitting on the correlation data comprises: Taking the original water level and area curve as reference, the cubic function relationship between water level and area is obtained by SPSS software analysis; wherein, the original water level and area curve is a relationship curve based on historical data analysis; Using the SPSS software to fit the correlation data to determine the parameters in the cubic function relationship; The coefficient of determination is used as a regression evaluation index to perform regression verification on the fitted data equation. If the verification does not meet the requirements, the process of fitting the associated data using the SPSS software is returned to be re-executed to determine the parameters in the cubic function relationship until the verification meets the requirements.

9. The reservoir water level and storage capacity verification method based on multi-source remote sensing images and deep learning according to claim 1, It is characterized in that Discretizing the fitted equation includes: The target water level is discretized and valued from the lowest water level in increments of 0.05 m per level to obtain a discrete value of the water level; The water area corresponding to the discrete value of the water level is back-calculated through the fitted equation to complete the equation discretization.

10. The reservoir water level and storage capacity verification method based on multi-source remote sensing images and deep learning according to claim 1, It is characterized in that The step of deducing the water level and reservoir capacity equation based on the discrete data of the equation and fitting the water level and reservoir capacity curve comprises: Sort the discrete data of the equation from low to high according to the water level; Based on the sorted discrete data, reservoir capacity calculation is performed to obtain the reservoir capacity corresponding to each water level of the discrete water level value; The calculated storage capacity and its water level relationship are fitted with a cubic polynomial relationship by SPSS software to obtain a water level and storage capacity equation, a water level and storage capacity curve, and a determination coefficient; if the determination coefficient does not meet the preset requirements, it is returned for refitting; Wherein, the storage capacity is calculated based on the following formula: Among them, ΔV is the volume of the water layer between two adjacent contour lines; S′ 下 , S′ 上 Indicates the reservoir water area between adjacent water levels; ΔZ ′ Expressed as the height difference between adjacent water levels; Z min is the lowest water level in the discrete data.

Citation Information

Cited By

  • Reservoir capacity calculation and reservoir capacity loss interval identification method for large and medium-sized reservoirs

    CN120597212A

  • Reservoir capacity measuring and calculating method based on multi-temporal remote sensing image

    CN121708491A