A precision evaluation method and system for lake area remote sensing extraction and a storage medium

Abstract

The application discloses a lake area remote sensing extraction precision evaluation method and system and a storage medium, and relates to the technical field of remote sensing monitoring. The method comprises the following steps: acquiring long-time sequence remote sensing images of a lake; extracting a lake area time sequence according to the long-time sequence remote sensing images; constructing a water level-simulated area relationship curve according to the lake area time sequence; constructing a water level-observed area relationship curve according to the water level values of a lake control hydrological station at the same moment of the long-time sequence remote sensing images and the lake areas under each water level; and evaluating the precision of lake area remote sensing extraction according to the difference between the water level-simulated area relationship curve and the water level-observed area relationship curve. The application can improve the evaluation efficiency of the lake area extraction precision.

Description

Technical Field

[0001] This invention relates to the field of remote sensing monitoring technology, and in particular to a method, system, and storage medium for assessing the accuracy of remote sensing extraction of lake area. Background Technology

[0002] Lakes, as vital carriers of Earth's water resources, serve as important indicators of the extent to which climate change and human activities have an impact. Scientific and effective monitoring of floods and droughts is crucial for flood control, drought relief, and disaster recovery. Satellite remote sensing technology, with its significant advantages such as wide coverage, short cycles, and high timeliness, has gradually become a primary means of modern flood and drought disaster monitoring and risk assessment, attracting widespread attention and research from scholars both domestically and internationally.

[0003] For lakes lacking topographic data, long-term remote sensing imagery provides a valuable data source for studying lake water evolution patterns. Accurate lake area extraction is crucial for monitoring lake water resource changes and assessing flood and drought risks. However, current lake area extraction based on remote sensing imagery primarily employs visual interpretation methods. This method uses visually interpreted (interpreted) sample data for each land cover category as standard data and compares it with machine extraction results. By establishing a confusion matrix, it calculates the overall classification accuracy Pc and Kappa coefficient for each land cover category, thereby quantifying the accuracy of remote sensing image interpretation and comprehensively evaluating the accuracy of lake area extraction. Visual interpretation methods are labor-intensive (consuming significant manpower and time), requiring manual interpretation of each land cover category sample point in every image. The interpretation results are subjective, affecting the accuracy of remote sensing lake area extraction.

[0004] Therefore, there is an urgent need to propose a fast, simple, and effective method for assessing the accuracy of remote sensing extraction of lake area. Summary of the Invention

[0005] The purpose of this invention is to provide a method, system, and storage medium for assessing the accuracy of remote sensing extraction of lake area. This method can improve the efficiency of assessing the accuracy of lake area extraction and overcome the limitations of subjective classification accuracy evaluation and local optima in the evaluation results of visual interpretation methods. It solves the problem of human error interference from a global perspective of different water levels, improves the reliability and accuracy of remote sensing extraction accuracy assessment of lake area, and provides a scientific basis for selecting more accurate and applicable lake water body inversion methods.

[0006] To achieve the above objectives, the present invention provides the following solution:

[0007] A method for assessing the accuracy of remote sensing extraction of lake area includes:

[0008] Acquire long-term remote sensing images of lakes;

[0009] Extracting the time series of lake area from long-term remote sensing images;

[0010] Construct a water level-simulated area relationship curve based on the time series of lake area;

[0011] Based on the measured water level values ​​of lake control hydrological stations at the same moment in long-term remote sensing images and the lake area at each water level, a water level-measured area relationship curve is constructed.

[0012] The accuracy of remote sensing extraction of lake area is evaluated based on the difference between the water level-simulated area relationship curve and the water level-measured area relationship curve.

[0013] Optionally, the acquisition of long-term remote sensing images of the lake further includes:

[0014] Preprocessing of long-term remote sensing images; the preprocessing includes: radiometric calibration, atmospheric correction, image mosaicking, and cropping;

[0015] Multi-scale segmentation of preprocessed long-term remote sensing images is performed based on object-oriented classification.

[0016] Optionally, the step of extracting the lake area time series from long-term remote sensing images specifically includes:

[0017] Automatic extraction of lake water time series from segmented long-term remote sensing images was performed using the normalized water body NDWI index and the improved normalized water body MNDWI index.

[0018] The first group of lake area values ​​was determined based on the time series of lake water bodies extracted from the Normalized Difference Water Index (NDWI).

[0019] The second set of lake area values ​​was determined based on the lake water time series extracted from the improved normalized water body MNDWI index.

[0020] Among them, using the formula Determine the normalized water quality index (NDWI); use the formula The improved normalized water body MNDWI index was determined; NDWI is the normalized water body NDWI index, MNDWI is the improved normalized water body MNDWI index, GREEN is the green light band brightness value, NIR is the near-infrared band brightness value, and SWIR is the mid-infrared band brightness value.

[0021] Optionally, the step of constructing the water level-simulated area relationship curve based on the lake area time series specifically includes:

[0022] The time series of the first group of lake area values ​​and the time series of the second group of lake area values ​​are arranged in ascending order according to water level, and the first relationship curve of water level-simulated area and the second relationship curve of water level-simulated area are constructed.

[0023] Optionally, the evaluation of the accuracy of remote sensing extraction of lake area based on the difference between the water level-simulated area relationship curve and the water level-measured area relationship curve specifically includes:

[0024] Based on the first relationship curve of water level-simulated area, the second relationship curve of water level-simulated area, and the relationship curve of water level-measured area, the accuracy of lake area extraction was evaluated using observation and statistical methods.

[0025] A system for assessing the accuracy of remote sensing extraction of lake area includes:

[0026] The image acquisition module is used to acquire long-term remote sensing images of lakes;

[0027] The lake area extraction module is used to extract the time series of lake area based on long-term remote sensing images;

[0028] The water level-simulated area relationship curve construction module is used to construct water level-simulated area relationship curves based on the time series of lake areas.

[0029] The water level-measured area relationship curve construction module is used to construct the water level-measured area relationship curve based on the measured water level values ​​of the lake's control hydrological stations at the same moment in long-term remote sensing images and the lake area at each water level.

[0030] The evaluation module is used to evaluate the accuracy of remote sensing extraction of lake area based on the difference between the water level-simulated area relationship curve and the water level-measured area relationship curve.

[0031] A storage medium having computer program instructions stored thereon, which, when executed by a processor, implement the method.

[0032] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0033] This invention provides a method, system, and storage medium for assessing the accuracy of remote sensing extraction of lake area. It constructs a water level-simulated area relationship curve based on the lake area time series, and a water level-measured area relationship curve based on the measured water level values ​​of key hydrological stations at the same time in long-term remote sensing images and the lake area at each water level. By comparing the differences between the water level-simulated area relationship curve and the water level-measured area relationship curve, the accuracy of remote sensing extraction of lake area is quickly and comprehensively assessed. This invention overcomes the limitations of subjective classification accuracy evaluation and the local optima of evaluation results in visual interpretation methods, improves the efficiency of accuracy assessment, solves the problem of human error interference from a global perspective at different water levels, and enhances the reliability and accuracy of remote sensing extraction accuracy assessment of lake area. It provides a scientific basis for selecting more accurate and applicable lake water body inversion methods. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 This is a schematic diagram of the process for extracting lake area based on remote sensing images provided by the present invention;

[0036] Figure 2 The image shows a time-series water body map of a lake extracted using the NDWI method as an example.

[0037] Figure 3 The following is an example of a time-series water body map of a lake extracted using the MNDWI method;

[0038] Figure 4 This is an example of a curve showing the water level-area relationship of a lake during the dry season. Detailed Implementation

[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] The purpose of this invention is to provide a method, system, and storage medium for assessing the accuracy of remote sensing extraction of lake area. This method can quickly and comprehensively assess the accuracy of remote sensing extraction of lake area, and overcomes the limitations of subjective classification accuracy evaluation and local optima in the evaluation results of visual interpretation methods. It improves the efficiency of accuracy assessment, solves the problem of human error interference from a global perspective of different water levels, and improves the reliability and accuracy of remote sensing extraction accuracy assessment of lake area.

[0041] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0042] like Figure 1 As shown, the present invention provides a method for evaluating the accuracy of remote sensing extraction of lake area, comprising:

[0043] S101, acquire long-term remote sensing images of the lake;

[0044] Following S101 are:

[0045] S1, preprocessing long-term remote sensing images; the preprocessing includes: radiometric calibration, atmospheric correction, image mosaicking and cropping.

[0046] S1 specifically includes:

[0047] S11. When the lake area is too large to obtain continuous images at the same time, images with similar satellite transit times are selected as fillers to ensure data quality and the reliability of the inversion results.

[0048] S12. Considering that remote sensing images use DN values ​​to record information, and that DN values ​​are dimensionless numbers, radiometric calibration is performed on each image used for mosaicking to convert the DN values ​​into actual radiometric values.

[0049] S13 performs atmospheric correction on each image after radiometric calibration in S12, based on header information (including imaging time, sensor type, band information, and data latitude and longitude) acquired using the FLAASH tool and satellite remote sensing data. S13 aims to eliminate the effects of atmospheric absorption and scattering on surface reflectivity, as well as radiometric errors caused by atmospheric influences.

[0050] S14 uses the Seamless Mosaic tool to mosaic the atmospherically corrected image data from S13, stitching together all images within the lake area.

[0051] S15: Draw the vector boundary of the lake, and use the vector boundary to crop the mosaicked remote sensing image in S14.

[0052] S2, based on object-oriented classification, performs multi-scale segmentation on preprocessed long-term remote sensing images; that is, homogeneous pixels are classified into each segmentation unit according to spectral, texture and geometric features, and each segmentation unit is treated as an object.

[0053] S102, Extract the time series of lake area from long-term remote sensing images;

[0054] S102 specifically includes:

[0055] Automatic extraction of lake water time series from segmented long-term remote sensing images was performed using the normalized water body NDWI index and the improved normalized water body MNDWI index.

[0056] The first group of lake area values ​​was determined based on the time series of lake water bodies extracted from the Normalized Difference Water Index (NDWI).

[0057] The second set of lake area values ​​was determined based on the lake water time series extracted from the improved normalized water body MNDWI index.

[0058] Among them, using the formula Determine the normalized water quality index (NDWI); use the formula The improved normalized water body MNDWI index was determined; NDWI is the normalized water body NDWI index, MNDWI is the improved normalized water body MNDWI index, GREEN is the green light band brightness value, NIR is the near-infrared band brightness value, and SWIR is the mid-infrared band brightness value.

[0059] The water body time series extracted by the normalized water body NDWI index and the water body time series extracted by the improved normalized water body MNDWI index were imported into the GIS platform. Using the computational geometry function in the layer attribute table, the sum of the areas of all lake water objects was calculated to obtain the first set of lake area values ​​{A1} and the second set of lake area values ​​{A2}.

[0060] S103, construct the water level-simulated area relationship curve based on the lake area time series;

[0061] S103 specifically includes:

[0062] The time series of the first group of lake area values ​​and the time series of the second group of lake area values ​​are arranged in ascending order according to water level, and the first relationship curve C1 and the second relationship curve C2 between water level and simulated area are constructed.

[0063] S104. Based on the measured water level values ​​of the lake's control hydrological stations at the same moment in the long-term remote sensing imagery and the lake area {A3} at each water level, construct the water level-measured area relationship curve C3. For images lacking measured lake areas, calculate the corresponding area of ​​a specific lake at a given water level based on C3 interpolation, and insert it into the sequence {A3}.

[0064] S105. The accuracy of remote sensing extraction of lake area is evaluated based on the difference between the water level-simulated area relationship curve and the water level-measured area relationship curve.

[0065] S105 specifically includes:

[0066] Based on the first relationship curve of water level-simulated area, the second relationship curve of water level-simulated area, and the relationship curve of water level-measured area, the accuracy of remote sensing extraction of lake area is evaluated using observation and statistical methods.

[0067] The accuracy of lake area extraction was assessed macroscopically using an observational method. By observing and comparing the trends, shapes, and distances of the measured and simulated water level-area relationship curves, the rise and fall of the measured and simulated lake area with the water level was analyzed. The degree of agreement between the water level-simulated area relationship curves C1 and C2 and the C3 curve was determined, thereby assessing the overall accuracy of lake area extraction based on methods such as NDWI and MNDWI.

[0068] Statistical methods were employed to precisely evaluate the accuracy of lake area extraction. The root mean square errors RMSE1 and RMSE2 of sequences {A1} and {A2} were calculated respectively. These indicators are used to measure the accuracy and stability of the simulated area. The smaller the RMSE, the higher the accuracy of the simulated area. This allows us to determine the magnitude of the error between the lake area values ​​extracted by methods such as NDWI and MNDWI and the measured values, and further assess their accuracy.

[0069] The root mean square error (RMSE) is specifically as follows: In the formula, n represents the number of extractions, i.e., the number of data points contained in the sequence {A1} or {A2}; y i The representative value is the simulated lake area in the sequence {A1} or {A2}; y i * This represents the measured value, that is, the measured value of the lake area in the sequence {A3} at the corresponding water level.

[0070] The sequences {A1} and {A3}, and {A2} and {A3} are merged into single datasets (A1, A3) and (A2, A3) respectively. The covariances Cov(A1, A3) and Cov(A2, A3) of the two datasets are calculated. This index is used to measure the degree of linear correlation between two random variables. The larger the Cov, the greater the degree of the same direction between the two variables and the stronger the correlation. This helps to determine which method extracts lake area values ​​with a stronger correlation to the measured values ​​and has higher accuracy.

[0071] The covariance Cov is specifically defined as: Cov(X,Y)=E(XY)-E(X)E(Y), where E is the expected value of the random variable.

[0072] The expected value E of the random variable is specifically: In the formula, n represents the number of data points, i.e., the number of extractions; y i Represents the simulated value, i.e., the sequence {A} i The simulated lake area in}

[0073] Considering the variability of the C3 curve caused by different lake topographic features, in order to further objectively reflect the accuracy of lake area extraction, the simulated C1 and C2 curves and the measured C3 curve can be divided into several segments. Based on the same water level segment, the root mean square error and covariance of the lake area can be calculated in detail to evaluate the lake area extraction accuracy of methods such as NDWI and MNDWI.

[0074] To better illustrate the advantages of this invention compared to existing visual interpretation methods (classification accuracy evaluation methods), the following explanation of the visual interpretation method is provided: This method compares a classification map with standard data, using the percentage of correctly classified data to characterize accuracy. The key is selecting several sample points based on the size of the study area. The land cover categories of the visually interpreted sample points are used as standard data, and compared with the land cover categories extracted by the machine. By establishing a confusion matrix (as shown in Table 1), the overall classification accuracy P of each land cover category is calculated. c The Kappa coefficient is used to assess the accuracy of automatic classification of remote sensing images. The confusion matrix is ​​calculated by comparing the location and category of each land cover's true pixels with the corresponding location and category in the automatically classified image; it is mainly used to compare the classification results with the actual land surface information. The overall classification accuracy P is... c Specifically: In the formula P kk The diagonal elements of the confusion matrix represent the number of sample points whose predicted class (automatically extracted by the machine) matches the actual class (visually interpreted data); p represents the total number of sample points involved in the classification. Overall classification accuracy P0 c The larger the value, the higher the extraction accuracy. The Kappa coefficient is specifically defined as follows: In the formula P +k P represents the number of sample points of actual category (visually interpreted data) k; k+ This represents the number of sample points that the machine predicts as category k. A higher Kappa coefficient indicates higher extraction accuracy. Classification accuracy evaluation methods require visual interpretation of each sample point in every image, which is time-consuming and labor-intensive. Furthermore, visual interpretation results are subject to subjectivity and human error, negatively impacting the accuracy of the assessment.

[0075] Table 1

[0076]

[0077] By analyzing the fit between curves using observation and statistical methods, the accuracy of remote sensing extraction of lake area can be quickly and comprehensively assessed. Taking into full account the physical phenomenon of lake area changes caused by water level fluctuations, the fitted curves of water level-simulated area and water level-measured area can be divided into several segments for comparison. This method overcomes the limitations of subjective classification accuracy evaluation and local optima in visual interpretation methods. It addresses the problem of human error interference from a global perspective across different water levels, improving the accuracy, reliability, and efficiency of remote sensing extraction accuracy assessment of lake area.

[0078] The following is an illustration through specific examples:

[0079] S301. Acquire Landsat 8 remote sensing image data of a certain lake. The image dates are June 1, 2015; June 8, 2015; February 26, 2016; July 5, 2019; March 1, 2020; August 24, 2020; and January 31, 2021, totaling 7 scenes. In this embodiment, because the lake covers a large area, it is not possible to acquire images from the same time point for mosaicking. Therefore, images with similar satellite transit times are selected as fillers to ensure data quality and the reliability of the inversion results. The selected data are mostly images from the lake's dry season.

[0080] S302 performs preprocessing on the acquired remote sensing image data, including radiometric calibration, atmospheric correction, image mosaicking, and cropping.

[0081] S302 specifically includes the following:

[0082] S121. Considering that remote sensing images record information using DN values, and that DN values ​​are dimensionless numbers, radiometric calibration is performed on each image used for mosaicking to convert DN values ​​into actual radiometric values. Given that the input data type for subsequent FLAASH atmospheric correction is BIL, the calibration data output type is set to Float, and the format is set to BIL.

[0083] S122 performs atmospheric correction on each frame of the radiometrically calibrated image in S121, based on header information (including imaging time, sensor type, band information, and data latitude and longitude) acquired using the FLAASH tool and satellite remote sensing data. This aims to eliminate the influence of atmospheric absorption and scattering on surface reflectivity, as well as radiometric errors caused by atmospheric effects.

[0084] In step S123, the atmospherically corrected image data from step S122 is mosaicked using the Seamless Mosaic tool to stitch together all images within the lake area. In this embodiment, histogram matching is selected in the color correction function to reduce color differences between images. Considering that bicubic convolution has higher accuracy and better smoothness, and can effectively handle high-frequency details and image edge information, bicubic convolution is chosen as the resampling method for mosaicking.

[0085] S124: Draw the vector boundary of the lake, and use the vector boundary to clip the mosaicked remote sensing image in S123.

[0086] S303, multi-scale segmentation is performed on the preprocessed image based on object-oriented classification. Homogeneous pixels are grouped into each segmentation unit according to spectral, texture, and geometric features, and each segmentation unit is treated as an object. Table 2 shows the parameter settings for multi-scale segmentation in this embodiment.

[0087] Table 2

[0088] 80 0.9 0.1 0.7 0.3

[0089] S304, using the Normalized Normalized Water Body Index (NDWI) and the improved Normalized Normalized Water Body Index (MNDWI) model indexes, and through repeated experiments to set thresholds, automatically extracts the time series data of a lake's water body. Figure 2 ,and Figure 3 As shown.

[0090] The normalized water quality index (NDWI) is specifically as follows: In the formula, GREEN represents the brightness value of the green light band, which is the B3 band in the Landsat 8 OLI image; NIR represents the brightness value of the near-infrared band, which is the B5 band in the Landsat 8 OLI image.

[0091] The improved Normalized Difference Water Index (MNDWI) is specifically as follows: In the formula, SWIR represents the mid-infrared band brightness value, which is the B6 band in the Landsat 8 OLI image.

[0092] In step S305, the lake water bodies extracted using two water index models in step S304 are imported into the GIS platform. Using the computational geometry function in the layer attribute table, the sum of the areas of all lake water bodies is calculated, resulting in two sets of simulated lake area values ​​{A1} and {A2}. {A1} represents the lake area value extracted by the NDWI method at each time phase, and {A2} represents the lake area value extracted by the MNDWI method at each time phase, as shown in Table 3.

[0093] Table 3

[0094]

[0095]

[0096] S306. Based on measured hydrological data and topographic data, obtain the measured water level values ​​of the lake's control hydrological stations at the same moment in the remote sensing images and the lake area at each water level (i.e., the measured area {A3}).

[0097] S307. The lake area time series {A1} and {A2} are arranged in ascending order according to water level, constructing water level-simulated area relationship curves C1 and C2. C1 is the lake water level-simulated area relationship curve extracted by the NDWI method, with a coefficient of determination R1. 2 =0.9797, indicating a good fit; C2 is the lake water level-simulated area relationship curve extracted by the MNDWI method, R2 2 =0.9819, indicating a good fit, such as Figure 4 As shown.

[0098] S308. To find a reference line for determining the accuracy of the water level-simulated area relationship curves C1 and C2 in S307, the lake area sequence {A3} is arranged in ascending order according to water level, and the water level-measured area relationship curve C3 (A=5.3611H) is constructed. 3 -7.8452H 2 +55.722H+2094.3, where A is the lake area and H is the water level, R3 2 =0.9408, indicating a good fit, such as Figure 4 As shown, the area corresponding to the missing water level of a certain lake is calculated based on C3 interpolation and placed into the sequence {A3}.

[0099] The lake area values ​​of {A1}, {A2}, and {A3} after processing by S307 and S308 are shown in Table 4.

[0100] Table 4

[0101]

[0102]

[0103] S309, the accuracy of lake area extraction is macroscopically evaluated using an observational method. By observing and comparing the trends, shapes, and distances of the measured and simulated water level-area relationship curves, the rise and fall of the measured and simulated lake area with water level is analyzed. The degree of agreement between the C1 and C2 water level-simulated area relationship curves and the C3 curve is determined, thereby assessing the overall accuracy of lake area extraction based on methods such as NDWI and MNDWI. In this embodiment, based on observation... Figure 4It was later found that both the C1 and C2 water level-simulated area relationship curves showed an upward trend, and the C2 curve had a higher degree of agreement with the C3 water level-measured area relationship curve. Therefore, it was determined that the MNDWI water index method was more accurate in extracting the area of ​​a lake.

[0104] S310 uses statistical methods to precisely evaluate the accuracy of lake area extraction.

[0105] S310 specifically includes the following:

[0106] S321. Calculate the root mean square error (RMSE1) and root mean square error (RMSE2) of sequences {A1} and {A2} respectively. This index is used to measure the accuracy and stability of the simulated area. The smaller the RMSE, the higher the accuracy of the simulated area. This helps to determine the magnitude of the error between the lake area values ​​extracted by methods such as NDWI and MNDWI and the measured values, and further assess their accuracy.

[0107] The root mean square error (RMSE) is specifically as follows: In the formula, n represents the number of extractions, i.e., the number of data points contained in the sequence {A1} or {A2}; y i The representative value is the simulated lake area in the sequence {A1} or {A2}; y i * This represents the measured value, that is, the measured value of the lake area in the sequence {A3} at the corresponding water level.

[0108] In the example, RMSE1 = 455.297 and RMSE2 = 203.238 were obtained. Since RMSE1 > RMSE2, the error of extracting the area value of a lake using the MNDWI water index method is smaller and the accuracy is higher.

[0109] S322: The sequences {A1} and {A3}, and {A2} and {A3} are merged into single datasets (A1, A3) and (A2, A3) respectively, and the covariances Cov(A1, A3) and Cov(A2, A3) of the two datasets are calculated. This index is used to measure the degree of linear correlation between two random variables. The larger the Cov, the greater the degree of same direction between the two variables and the stronger the correlation. This helps to determine which method extracts the lake area value with a stronger correlation to the measured value and has higher accuracy.

[0110] The covariance Cov is specifically defined as: Cov(X,Y)=E(XY)-E(X)E(Y), where E is the expected value of the random variable.

[0111] The expected value E of the random variable is specifically: In the formula, n represents the number of data points, i.e., the number of extractions; y i Represents the simulated value, i.e., the sequence {A} iThe simulated lake area values in

[0112] In the embodiment, Cov(A1, A3) = 137320.985, Cov(A2, A3) = 338190.639, Cov(A1, A3) < Cov(A2, A3). Then, using the MNDWI water body index method to extract the lake area value has a stronger correlation with the measured value and higher accuracy.

[0113] S323. Considering the variability of the C3 curve caused by the topographic characteristics of different lakes, in order to more objectively reflect the accuracy of lake area extraction, the simulated C1 and C2 curves and the measured C3 curve can be divided into several segments. According to the same water level, the root mean square error and covariance of the lake area are calculated in detail for each segment to evaluate the lake area extraction accuracy of methods such as NDWI and MNDWI. In this embodiment, the simulated C1 and C2 curves and the measured C3 curve are evenly divided into three segments (water level 1.18m - 1.48m, 1.48m - 1.61m, 1.61m - 2.67m), and the root mean square error and covariance of the lake area values of each segment of the curve are calculated respectively. The calculation results are shown in Table 5.

[0114] Table 5

[0115]

[0116] From the calculation results in the above table, it can be seen that the root mean square error of using the NDWI method to extract the lake area value is less than the corresponding value of the MNDWI method in the first two segments of the curve, but it is 3.5 times higher than the root mean square error of using the MNDWI method to extract the lake area value in the third segment of the curve; in the three segments of the curve, the covariance of using the NDWI method to extract the lake area value is less than the corresponding value of the MNDWI method, and there are large differences in the first and third segments. Through comprehensive analysis, it is evaluated that the remote sensing extraction accuracy of a certain lake area based on the MNDWI method is higher.

[0117] In this embodiment, by applying the accuracy evaluation method for remote sensing extraction of lake area proposed by the present invention, it is concluded that the remote sensing extraction accuracy of a certain lake area based on the MNDWI method is higher. Thus, it can be inferred that the MNDWI method has stronger applicability for the inversion of the water body of this lake. Compared with the measured area, there are still certain errors in the lake area values extracted by the NDWI and MNDWI water body index methods. The main sources are: (1) The resolution and quality of Landsat8 images have a greater impact on the simulation results. High-resolution image data is more likely to inversely reflect the characteristics of the lake water body in detail; (2) The influence of clouds and fog in optical remote sensing images is complex and changeable, causing greater interference to the extraction of the lake water body; (3) The water body index method belongs to a traditional method of threshold classification. Its disadvantage is that only two band values are involved in the calculation, and the weights of other bands participating in the segmentation are low, and the advantages of multi-spectral information cannot be fully utilized.

[0118] To better illustrate the advantages of this invention compared to existing visual interpretation methods (classification accuracy evaluation methods), the embodiment also uses the classification accuracy evaluation method to assess the accuracy of extracting the area of ​​a lake: To avoid interference from clouds and fog on remote sensing interpretation, an image from January 22, 2015 (with a cloud cover of 0.57% and good image quality) was selected for lake area extraction. 1500 sample points were randomly generated within the lake area. The land cover categories of the visually interpreted sample points were used as standard data and compared with the land cover categories extracted based on the NDWI and MNDWI water index methods. By establishing a confusion matrix, the overall classification accuracy P of each land cover category was calculated. c And the Kappa coefficient. The overall classification accuracy P is mentioned. c Specifically: In the formula P kk The diagonal elements of the confusion matrix represent the number of sample points whose predicted class (automatically extracted by the machine) matches the actual class (visually interpreted data); p represents the total number of sample points involved in the classification. Overall classification accuracy P0 c The larger the value, the higher the extraction accuracy. The Kappa coefficient is specifically defined as follows: In the formula P +k P represents the number of sample points of actual category (visually interpreted data) k; k+ This represents the number of sample points where the machine predicts category k. A higher Kappa coefficient indicates higher extraction accuracy. In this embodiment, the overall classification accuracy P based on the visual interpretation method is... c The calculation results of the Kappa coefficient are shown in Table 6.

[0119] Table 6

[0120] NDWI method 0.961 0.716 MNDWI method 0.964 0.745

[0121] Calculations show that the overall classification accuracy (Pc) and Kappa coefficient of the MNDWI method are both greater than those of the NDWI method, thus indicating that the MNDWI method has a higher accuracy in extracting lake area than the NDWI method. In the example, the accuracy assessment of remote sensing extraction of lake area from a single image of a lake using visual interpretation took 2 hours, and it was estimated that it would take 14 hours for 7 images. However, the accuracy assessment of remote sensing extraction of lake area from 7 images of a lake using this invention took only 1.2 hours, reducing the working time by 1066.7%, significantly improving work efficiency. Furthermore, this invention can quickly and comprehensively assess the accuracy of remote sensing extraction of lake area. This method overcomes the limitations of subjective classification accuracy evaluation and the local optima of the evaluation results in visual interpretation, and solves the problem of human error interference from a global perspective at different water levels, improving the reliability and accuracy of remote sensing extraction accuracy assessment of lake area.

[0122] Corresponding to the above method, the present invention also provides an accuracy assessment system for remote sensing extraction of lake area, comprising:

[0123] The image acquisition module is used to acquire long-term remote sensing images of lakes.

[0124] The lake area extraction module is used to extract the time series of lake area based on long-term remote sensing images.

[0125] The water level-simulated area relationship curve construction module is used to construct water level-simulated area relationship curves based on the time series of lake area.

[0126] The water level-measured area relationship curve construction module is used to construct the water level-measured area relationship curve based on the measured water level values ​​of the lake's control hydrological stations at the same moment in long-term remote sensing images and the lake area at each water level.

[0127] The evaluation module is used to evaluate the accuracy of remote sensing extraction of lake area based on the difference between the water level-simulated area relationship curve and the water level-measured area relationship curve.

[0128] In order to execute the methods corresponding to the above embodiments and achieve the corresponding functions and technical effects, the present invention also provides a storage medium storing computer program instructions thereon, characterized in that the method is implemented when the computer program instructions are executed by a processor.

[0129] Based on the above description, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods of the various embodiments of the present invention. The aforementioned computer storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks.

[0130] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0131] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A precision evaluation method for lake area remote sensing extraction, characterized in that, include: Acquire long-term remote sensing images of lakes; Extracting the time series of lake area from long-term remote sensing images; Construct a water level-simulated area relationship curve based on the time series of lake area; Based on the measured water level values ​​of lake control hydrological stations at the same moment in long-term remote sensing images and the lake area at each water level, a water level-measured area relationship curve is constructed. The accuracy of remote sensing extraction of lake area is evaluated based on the difference between the water level-simulated area relationship curve and the water level-measured area relationship curve; The acquisition of long-term remote sensing images of the lake further includes: Preprocessing of long-term remote sensing images; the preprocessing includes: radiometric calibration, atmospheric correction, image mosaicking, and cropping; Multi-scale segmentation of preprocessed long-term remote sensing images based on object-oriented classification method; The extraction of lake area time series from long-term remote sensing images specifically includes: Automatic extraction of lake water time series from segmented long-term remote sensing images was performed using the normalized water body NDWI index and the improved normalized water body MNDWI index. The first group of lake area values ​​was determined based on the time series of lake water bodies extracted from the Normalized Difference Water Index (NDWI). The second set of lake area values ​​was determined based on the lake water time series extracted from the improved normalized water body MNDWI index. Among them, using the formula Determine the normalized water quality index (NDWI); use the formula Determine the improved Normalized Difference Water Index (MNDWI); The normalized water NDWI index, For the improved Normalized Difference Water Index (MNDWI), GREEN represents the brightness value in the green light band, NIR represents the brightness value in the near-infrared band, and SWIR represents the brightness value in the mid-infrared band. The construction of the water level-simulated area relationship curve based on the lake area time series specifically includes: The time series of the first group of lake area values ​​and the time series of the second group of lake area values ​​are arranged in ascending order according to water level, and the first relationship curve of water level-simulated area and the second relationship curve of water level-simulated area are constructed. The accuracy of remote sensing extraction of lake area based on the difference between the water level-simulated area relationship curve and the water level-measured area relationship curve is evaluated, specifically including: Based on the first relationship curve of water level-simulated area, the second relationship curve of water level-simulated area, and the relationship curve of water level-measured area, the accuracy of remote sensing extraction of lake area is evaluated using observation and statistical methods. The accuracy of remote sensing extraction of lake area was evaluated using observational and statistical methods, specifically including: The accuracy of lake area extraction is macroscopically evaluated using the observation method. Specifically, by observing and comparing the trends, shapes, and distances of the measured and simulated water level-area relationship curves, the rise and fall of the measured and simulated lake area with the water level is analyzed. The degree of agreement between the first water level-simulated area relationship curve C1 and the second water level-simulated area relationship curve C2 and the water level-measured area relationship curve C3 is determined, thereby determining the overall accuracy of lake area extraction based on the NDWI and MNDWI methods. Statistical methods were employed to meticulously evaluate the accuracy of lake area extraction. Specifically, the root mean square errors (RMSE1) and RMSE2 of sequences {A1} and {A2} were calculated. RMSE1 and RMSE2 are used to measure the accuracy and stability of the simulated area; the smaller the RMSE, the higher the accuracy of the simulated area. This allows for the assessment of the error between the lake area values ​​extracted by the NDWI and MNDWI methods and the measured values, further evaluating their accuracy. Here, {A1} and {A2} represent the lake area values ​​of the first and second groups, respectively. {A1} was obtained using the NDWI method, and {A2} was obtained using the MNDWI method. The sequences {A1} and {A3}, and {A2} and {A3} are merged into single datasets (A1, A3) and (A2, A3) respectively. The covariances Cov(A1, A3) and Cov(A2, A3) of the two datasets are calculated. Covariance is used to measure the degree of linear correlation between two random variables. The larger the covariance, the greater the degree of the same direction between the two variables and the stronger the correlation. This helps to determine which method extracts lake area values ​​with a stronger correlation to the measured values ​​and higher accuracy. The first relationship curve C1 between water level and simulated area, the second relationship curve C2 between water level and simulated area, and the relationship curve C3 between water level and measured area are divided into several segments. Based on the same water level segment, the root mean square error and covariance of lake area are calculated in detail to evaluate the lake area extraction accuracy of the NDWI and MNDWI methods.

2. A precision evaluation system for lake area remote sensing extraction, used for implementing the precision evaluation method for lake area remote sensing extraction in claim 1; characterized in that, include: The image acquisition module is used to acquire long-term remote sensing images of lakes; The lake area extraction module is used to extract the time series of lake area based on long-term remote sensing images; The water level-simulated area relationship curve construction module is used to construct water level-simulated area relationship curves based on the time series of lake areas. The water level-measured area relationship curve construction module is used to construct the water level-measured area relationship curve based on the measured water level values ​​of the lake's control hydrological stations at the same moment in long-term remote sensing images and the lake area at each water level. The evaluation module is used to evaluate the accuracy of remote sensing extraction of lake area based on the difference between the water level-simulated area relationship curve and the water level-measured area relationship curve.

3. A storage medium having stored thereon computer program instructions, characterized in that, When the computer program instructions are executed by the processor, they implement the accuracy assessment method for remote sensing extraction of lake area as described in claim 1.

Images