Multi-temporal remote sensing image uncontrolled water depth inversion method based on adaptive weight fusion

Abstract

The invention relates to the technical field of water depth information extraction, in particular to a multi-temporal remote sensing image uncontrolled water depth inversion method based on adaptive weight fusion, which comprises the following steps: S1, receiving multi-temporal remote sensing image data and actually measured water depth data, and carrying out primary data processing; s2, inverting the initial water depth of each single-time-phase image by using a dual-band logarithmic linear model; s3, tide correction and deep water area fitting completion are carried out based on the result of the initial water depth in the S2; s4, according to the reliability of the water depth of the same name point relative to the fusion coefficient, adaptively calculating the fusion weight of the same name point at different time; and S5, based on an adaptive weight fusion algorithm, carrying out weighted fusion to generate an accurate water depth result, and solving the problem that the water depth inversion precision is influenced by reflectivity abnormity caused by single-temporal noise and insufficiency of a traditional multi-temporal fusion method.

Description

Technical Field

[0001] The present invention relates to the technical field of water depth information extraction, and in particular to an uncontrolled water depth inversion method for multi-temporal remote sensing images based on adaptive weight fusion. Background Art

[0002] Shallow water depth measurement is fundamental to ocean mapping. Accurate shallow water depth information plays a vital role in nearshore marine engineering construction, shipping route selection, and coastal ecological research. Shallow water depth inversion based on remote sensing imagery offers advantages such as multi-temporal observation, a wide inversion range, and low measurement costs. It can also capture information on islands and reefs in disputed, dangerous, or remote waters, making it a widely used and important method for obtaining shallow water depth data.

[0003] Uncontrolled depth inversion, however, can infer depth from remote sensing imagery even in the absence of measured depth data. Its algorithm is unaffected by locally trained models and exhibits greater transferability, thus gradually attracting attention from researchers. Existing technologies for uncontrolled depth inversion include the hyperspectral optimization model and its improved algorithms, the adaptive bathymetry estimation algorithm, and the physical dual-band log-linear analysis model (P-DLA). These models have been used to perform depth inversion in different ocean areas. Research has shown that the hyperspectral optimization model and its improved algorithms make numerous assumptions about water optical properties, resulting in numerous unknown parameters and instability. The adaptive bathymetry estimation method suffers from poor accuracy in deep waters. Compared to other algorithms, the P-DLA model achieves superior inversion results across diverse ocean areas, depth ranges, and satellite imagery resolutions. However, existing uncontrolled depth inversion research has primarily utilized single-temporal satellite data. Due to the inherent complexity of the ocean, depth inversion results based on single-temporal remote sensing imagery are often incomplete. Furthermore, for wide-field-of-view images, single-phase images may be affected by noise, resulting in poor image quality or missing information, which in turn affects the accuracy of uncontrolled depth inversion. Therefore, uncontrolled inversion using only single-phase images cannot obtain stable and reliable depth information.

[0004] Using multi-temporal imagery to jointly invert water depth provides an effective approach to addressing the challenges of single-temporal imagery and obtaining more reliable water depth information. For example, researchers have attempted to synthesize multi-temporal images using methods such as median synthesis, maximum outlier removal, and weighted synthesis, and then perform water depth inversion based on these synthesized images. However, image-based fusion can easily increase reflectance differences between adjacent pixels and introduce noise. In the absence of measured data, researchers have used dual-temporal hyperspectral optimization algorithms to solve for parameter errors and obtain water depth information. However, the instability of hyperspectral optimization algorithms can lead to unstable water depth results. In controlled water depth inversion studies, researchers have used the median of multi-temporal water depth inversion results, the average relative water depth composition (i.e., the logarithmic ratio of the blue and green bands), and the median of the logarithmic ratios as the final inversion result. This approach somewhat avoids outliers but does not guarantee that the fused result is the optimal water depth. Although existing research has shifted the perspective of water depth inversion from a single-temporal to a multi-temporal approach, most fusion methods are based on controlled water depth inversion and fail to fully consider the weighting of water depth inversion results from different temporal phases. The uncontrolled water depth inversion method is based on the physical transmission model. The process is more complicated and has more uncertain factors. The use of traditional multi-temporal fusion will inevitably lead to unstable and inaccurate fusion results, and it cannot be guaranteed that the obtained results are the optimal solutions.

[0005] Publication number CN119206534A discloses a "Multi-temporal Fusion Uncontrolled Water Depth Inversion Method Based on Remote Sensing Image Sediment Identification". This method is based on remote sensing image sediment identification, applies the P-DLA method to pre-processed remote sensing images to invert water depth and perform multi-temporal fusion. Although the fusion method has been extended to uncontrolled water depth inversion research, its essence is still to take the average of the multi-temporal water depth results as the final result, and to fuse the inversion results of different temporal phases with the same weight. It does not take into account the irrationality of applying the same weight to inversion results with different differences, does not evaluate the water depths of homonymous points from different data sources, and does not flexibly and adaptively assign weights based on the reliability of each homonymous point. This reduces the fault tolerance of the inversion process and cannot guarantee that the fusion result is the optimal water depth. Therefore, the current uncontrolled water depth inversion research has the problem that the reflectivity anomaly caused by single-temporal noise makes the inversion result one-sided and inaccurate, and the traditional multi-temporal fusion method has the problem of insufficient fusion and the fusion result is not the optimal solution. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to overcome the shortcomings of the existing technology and provide an uncontrolled water depth inversion method for multi-phase remote sensing images based on adaptive weight fusion to solve the problems mentioned in the above background technology of reflectivity anomaly caused by single-phase noise and the insufficiency of traditional multi-phase fusion methods that affect the accuracy of water depth inversion.

[0007] The technical solution adopted by this application to solve the technical problem is: a multi-temporal remote sensing image uncontrolled water depth inversion method based on adaptive weight fusion includes the following steps:

[0008] S1: Receive multi-temporal remote sensing image data and measured water depth data, and perform preliminary data processing;

[0009] S2: Invert the preliminary water depth of each single-temporal image using a dual-band log-linear model;

[0010] S3: Tidal correction and deep water area fitting completion are performed based on the preliminary water depth results in S2;

[0011] S4: Adaptively calculate the fusion weights of the same-named points at different times based on the reliability of the water depth of the same-named points relative to the fusion coefficient;

[0012] S5: Based on the adaptive weight fusion algorithm, weighted fusion generates accurate water depth results.

[0013] In said S1, multi-temporal satellite remote sensing image data is received and pre-processed to obtain true reflectivity information;

[0014] The measured water depth data are standardized to verify the accuracy of the water depth inversion results.

[0015] In the S2, different bottom textures are identified by the brightness of shallow water areas in remote sensing images. Based on a dual-band log-linear model, characteristic pixels with different water depths and bottom textures are extracted from multi-temporal remote sensing images to invert the preliminary water depth of each single-temporal image.

[0016] The S3 includes the following sub-steps:

[0017] S3-1: Tidal correction is performed on all inversion results, and the instantaneous water depth of each single phase is corrected to the steady-state water depth of the same reference level;

[0018] S3-2: The missing data are completed by linear fitting of the logarithmic ratio of the blue and green bands.

[0019] The S3-2 includes the following sub-steps:

[0020] S3-2-1: Select the temporal images with complete data in deep water areas from multiple remote sensing images as the basic data for establishing the relationship;

[0021] S3-2-2: For each pixel in the selected complete data phase image, calculate the logarithmic ratio of the blue band and green band reflectance, associate it with the inverted water depth value of the corresponding pixel, and establish a linear relationship between the two through linear regression method;

[0022] S3-2-3: For phase images with missing data in deep water areas, obtain the blue-green band reflectance at the same-name point, substitute the reflectance into the established linear relationship, and calculate the water depth value corresponding to the point, thereby completing the missing data and making the reverse evolution results of each phase complete.

[0023] In the S4, based on a dynamic adaptive weighting strategy, the water depths of the synonymous points from different data sources are evaluated and weights are adaptively assigned according to the reliability of each synonymous point to improve the fault tolerance in the inversion process, eliminate the interference of system errors and environmental factors, and achieve higher inversion accuracy.

[0024] The S4 includes the following sub-steps:

[0025] S4-1: Based on the data processed in step S3, the coordinates of the multi-temporal remote sensing image are used to match the nearest neighboring points of the same name pixel by pixel. For each group of points of the same name, the water depth data of each temporal phase is extracted.

[0026] S4-2: Calculate the average water depth of the same-name points Δh as the fusion coefficient, and use it as a reference standard to measure the difference between the inversion results of each image and the overall situation. The principle is:

[0027]

[0028] Where Δh is the fusion coefficient, (x, y) represents the coordinates of the pixel, and h i (x,y) is the water depth of the point with the same coordinate (x,y) in different time phases, and n is the number of multi-temporal images.

[0029] S4-3: Calculate the difference between the water depth and fusion coefficient of the same-name points of the multi-temporal results, and use the inverse of the difference as the preliminary weight;

[0030] S4-4: Calculate the sum of the preliminary weights of all images and normalize the preliminary weights so that the sum of the final weights of the multi-temporal inversion results is 1;

[0031] S4-5: Divide the preliminary weight of each image by the total weight to obtain the final weight of each image.

[0032] The final weight calculation formula in S4-5 is as follows:

[0033]

[0034] Where: Δh(x,y) is the fusion coefficient, h i (x,y) is the depth of the result of the multi-temporal image transformation of the same point with coordinates (x,y), n is the number of multi-temporal images, w i (x,y) is the final weight of the i-th image at position (x,y).

[0035] After determining the fusion weights of the data of different time phases in S4 in S5, the same-name points of each time phase are multiplied by their final weights, and then all weighted water depth inversion results are added together to obtain a weighted fusion result;

[0036] The above steps are repeated pixel by pixel, and finally a fusion result is generated by weighted fusion as the final water depth inversion output.

[0037] The calculation formula in S5 is:

[0038]

[0039] Where: h i (x, y) is the water depth in the inversion results at the position (x, y) in different phases, n is the number of multi-phase images, w i (x, y) is the weight of the inversion results of different phases at the (x, y) position, H i (x, y) is the optimal water depth at the (x, y) position after adaptive weight fusion at different time phases.

[0040] Compared with the prior art, this application has the following beneficial effects:

[0041] This application is based on a dynamic adaptive weighting strategy, which aims to adaptively calculate and assign weights according to the reliability of the results, improve the fault tolerance during the inversion process, and minimize the interference of system errors and environmental factors, thereby achieving higher inversion accuracy.

[0042] This application can suppress single-phase image noise, overcome the impact of insufficient traditional multi-phase image fusion on the accuracy of uncontrolled water depth inversion, and improve the accuracy and robustness of the fusion results by sequentially performing data collection and preprocessing, single-phase preliminary water depth inversion, tidal correction and fitting completion of the preliminary water depth, adaptive calculation of fusion weights, and adaptive weight fusion of multi-phase uncontrolled water depth inversion.

[0043] This uncontrolled water depth inversion method for multi-temporal remote sensing images based on adaptive weight fusion is applicable to various complex environments. The water depth information obtained based on this method can be directly used in fields such as nearshore marine engineering construction, shipping route selection, and coastal ecological research. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 1 is a flow chart of the method for uncontrolled water depth inversion based on multi-temporal remote sensing images based on adaptive weight fusion according to an embodiment of the present invention;

[0045] Figure 2 This is an example of feature pixel extraction in the multi-temporal remote sensing image uncontrolled water depth inversion method based on adaptive weight fusion according to the embodiment of the present invention. Figure I ;

[0046] Figure 3 This is an example of feature pixel extraction in the multi-temporal remote sensing image uncontrolled water depth inversion method based on adaptive weight fusion according to the embodiment of the present invention. Figure II ;

[0047] Figure 4 Comparison of the preliminary water depth inversion results of each single phase and the adaptive weight fusion results in the multi-phase remote sensing image uncontrolled water depth inversion method based on adaptive weight fusion according to the embodiment of the present invention Figure I ;

[0048] Figure 5 Comparison of the preliminary water depth inversion results of each single phase and the adaptive weight fusion results in the multi-phase remote sensing image uncontrolled water depth inversion method based on adaptive weight fusion according to the embodiment of the present invention Figure II ;

[0049] Figure 6 Comparison of the adaptive weight fusion results in the multi-temporal remote sensing image uncontrolled water depth inversion method based on adaptive weight fusion according to the embodiment of the present invention with the traditional median fusion and mean fusion results Figure I ;

[0050] Figure 7 Comparison of the adaptive weight fusion results in the multi-temporal remote sensing image uncontrolled water depth inversion method based on adaptive weight fusion according to the embodiment of the present invention with the traditional median fusion and mean fusion results Figure II . DETAILED DESCRIPTION

[0051] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0052] It should be noted that all directional indications in the embodiments of the present invention (such as up, down, left, right, front, back, etc.) are only used to explain the relative position relationship, movement status, etc. between the various components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0053] Reference Figure 1-Figure 7 The uncontrolled water depth inversion method of multi-temporal remote sensing images based on adaptive weight fusion includes the following steps:

[0054] S1: Receive multi-temporal remote sensing image data and measured water depth data, and perform preliminary data processing;

[0055] In said S1, multi-temporal satellite remote sensing image data is received and pre-processed to obtain true reflectivity information;

[0056] The measured water depth data are standardized to verify the accuracy of the water depth inversion results.

[0057] In this step, the collected multi-temporal remote sensing images are preprocessed to obtain accurate remote sensing reflectance information, which is the basis for uncontrolled water depth inversion. The preprocessing process of the collected multi-temporal remote sensing images generally includes: geometric correction, radiometric calibration, atmospheric correction and water-land separation. Geometric correction can eliminate the geometric distortion of the image, ensure the correct correspondence between pixels and actual geographical locations, and ensure consistency during multi-temporal fusion. Radiometric calibration can convert the DN value of the original image into a radiance value. Atmospheric correction can eliminate the interference of factors such as atmospheric scattering and aerosols, and thus obtain a surface reflectance close to the actual value. Water bodies and land can be identified by calculating the Normalized Difference Water Index (NDWI). The NDWI of seawater is positive and that of land is negative. Using 0 as the threshold can remove land and retain water bodies to achieve water-land separation. The principle is:

[0058]

[0059] Where: ρ(green) and ρ(NIR) represent the reflectance of green and near-infrared bands, respectively.

[0060] In this step, according to the tide level information at the time of imaging of the multi-temporal remote sensing image, the measured water depth data is tidal corrected for the corresponding image to verify the accuracy.

[0061] S2: The preliminary water depth of each single-temporal image is inverted using the Dual-Band Log-Linear Analysis Model Based on Physics (P-DLA);

[0062] In the S2, different bottom textures are identified by the brightness of shallow water areas in remote sensing images. Based on a dual-band log-linear model, characteristic pixels with different water depths and bottom textures are extracted from multi-temporal remote sensing images to invert the preliminary water depth of each single-temporal image.

[0063] The basic principle of the Dual-Band Log-Linear Analysis Model (P-DLA) involved in this step is to identify the bottom of the sea based on multispectral images, extract characteristic pixels with different bottom characteristics from the images, and realize water depth inversion based on the remote sensing reflectance information of the characteristic pixels. This method has strong inversion capabilities in different sea areas, different water depths, and satellite images of different resolutions. Its model principle is as follows:

[0064]

[0065] Where: h is the water depth, α1 and α2 are the weight feature vectors of the blue and green bands respectively, is the bottom parameter, g1 / g2 is the ratio of the attenuation coefficients of the blue and green bands, g1 is the sum of the diffuse attenuation coefficients of the blue band, and g2 is the sum of the diffuse attenuation coefficients of the green band.

[0066] Different types of substrate have different spectral reflectances, which appear as different brightness levels in remote sensing images. By comparing the brightness levels on remote sensing images, different types of substrate can be identified and distinguished, which is the basis for solving the various parameters of the P-DLA model. In this application, based on the substrate identification of remote sensing images, adjacent pixel pairs of different substrate types at the same depth are extracted to calculate the optimal band rotation unit vector [α1, α2], making them as independent of different substrate types as possible, ensuring that they are not affected by the substrate during water depth inversion and multi-temporal fusion; pixels with mixed substrates are extracted near the water depth of 0m, taking into account different substrate types as much as possible, and the model parameters are calculated and averaged to determine To further minimize the impact of different substrate types, multiple pixels from the same substrate at different depths are extracted to determine g1 / g2, minimizing the influence of water optical properties on the inversion results. g2 can be solved using the QAA algorithm. This method uses QAA_v6 to solve for the green band attenuation coefficient g2. g1 / g2 can be determined from the regression slope of the X1-X2 dataset collected at different depths for the same substrate, eliminating the need to solve g1 separately.

[0067] S3: Tidal correction and deep water area fitting completion are performed based on the preliminary water depth results in S2;

[0068] This step is the data preparation stage before adaptive weight fusion, and its main purpose is to ensure the consistency of the fused data. In order to ensure the accuracy of the multi-temporal fusion of adaptive weights, it is necessary to perform tidal correction on all inversion results before fusion, correcting the instantaneous water depth of each single temporal phase to the steady-state water depth of the same reference level. The principle is as follows:

[0069]

[0070] …

[0071] h mn =f(α 1-mn ,α 2-mn ,g 1-mn / g 2-mn ,g 2-mn )-T n

[0072] Where: m1, m2...mn represent remote sensing images at different times, T1, T2...T n Obtain the tide height at the time of each single-phase image, h m1 , h m2 ……h mn They are the steady-state water depths of each single time phase regression result corrected to the same reference plane.

[0073] When using remote sensing images for water depth inversion, the characteristics of each single-phase image vary. Due to various factors (such as the optical properties of the water body and imaging conditions), the reflectivity of deep water areas varies in different single-phase images. This results in missing data in some single-phase images in deep water areas, affecting the integrity and accuracy of the water depth inversion results. The present invention proposes to supplement the missing data by linearly fitting the logarithmic ratio of the blue and green bands. The basis for this method is that there is a certain inherent connection between the reflectivity information of the blue and green bands and the water depth. This connection is used to establish a linear relationship, and then infer the missing data.

[0074] Select the time-phase images with complete data in deep water areas from multiple remote sensing images. These images will serve as the basic data for establishing the relationship. For the selected time-phase images with complete data, calculate the logarithmic ratio of the blue band and green band reflectance for each pixel, and associate it with the inverted water depth value of the corresponding pixel. Through the linear regression method, establish a linear relationship between the two. For time-phase images with missing data in deep water areas, obtain the blue and green band reflectance at the same point, substitute the reflectance into the established linear relationship, and calculate the water depth value corresponding to the point, thereby completing the missing data and making the inversion results of each time phase complete. The principle is as follows:

[0075]

[0076] Where: dp represents the missing deep-water data in a single-phase time-phase evolution result, ρ rs-blue ,ρ rs-green The blue and green band reflectances of the pixels in the original remote sensing image that represent the missing deepwater data are used. n is a constant to ensure the correct value of the ratio, which is generally set to 1000. α and β are the regression coefficients of the model, which are constants obtained by fitting the inverted water depth values of the same-name points in each single phase without missing deepwater data and the corresponding logarithmic ratio of remote sensing reflectance.

[0077] S4: Adaptively calculate the fusion weights of the same-named points at different times based on the reliability of the water depth of the same-named points relative to the fusion coefficient;

[0078] In remote sensing water depth inversion, multiple images of the same area taken at different times or under different conditions can be affected by various factors (such as weather and sensor noise), resulting in the inversion results not fully reflecting the underwater information. The purpose of weighted fusion is to integrate the information of these images and, by reasonably assigning weights to each image, generate a fusion result that more accurately reflects the characteristics of the regional features. This step, based on the noise suppression and signal enhancement principles of weighted fusion, proposes a dynamic adaptive weighting strategy. By evaluating the water depths of synonymous points from different data sources, weights are flexibly and adaptively assigned according to the reliability of each synonymous point. This aims to improve the fault tolerance of the inversion process, minimize the interference of systematic errors and environmental factors, and thus achieve higher inversion accuracy. The weight is calculated based on the difference between the inversion result of each image and the fusion coefficient. The smaller the difference between the inversion result and the fusion coefficient, the greater its weight, which means that the inversion result of the image contributes more to the final fusion result.

[0079] Based on the fused data prepared in step S3, the coordinates of the multi-temporal remote sensing images are used to match the nearest neighboring points of the same name pixel by pixel. For each set of points of the same name, the water depth data of each phase is extracted. In order to calculate the difference between the inversion results, the average water depth of the points of the same name Δh is calculated as the fusion coefficient, which is used as a reference standard to measure the difference between the inversion results of each image and the overall situation. The principle is as follows:

[0080]

[0081] Where Δh is the fusion coefficient, (x, y) represents the coordinates of the pixel, and h i (x,y) is the water depth of the point with the same coordinate (x,y) in different time phases, and n is the number of multi-temporal images.

[0082] Calculate the difference between the water depth of the same-name points of the multi-temporal results and the fusion coefficient (the absolute value of the difference is used here, and only the size of the difference is concerned, not the direction of the difference). The smaller the difference, the closer the inversion result is to the fusion coefficient, the more reliable the information it contains, and the greater the weight should be given. Therefore, the inverse of the difference is used as the preliminary weight. In order to convert the preliminary weight into the final weight, it is necessary to calculate the sum of the preliminary weights of all images. This sum is used to normalize the preliminary weight so that the sum of the final weights of the multi-temporal inversion results is 1, ensuring that the contribution ratio of each phase in the fusion process is reasonable. The final weight of each image is obtained by dividing the preliminary weight of each image by the sum of the weights. The principle is:

[0083]

[0084] Where: Δh(x,y) is the fusion coefficient, h i (x,y) is the depth of the result of the multi-temporal image transformation of the same point with coordinates (x,y), n is the number of multi-temporal images, w i (x,y) is the final weight of the i-th image at position (x,y). So far, the fusion weights of the same points at different times are adaptively calculated according to the reliability of the water depth of the same points relative to the fusion coefficient.

[0085] S5: Based on the adaptive weight fusion algorithm, weighted fusion generates accurate water depth results.

[0086] After adaptively determining the fusion weights for data from different time phases, the synonymous points for each time phase are multiplied by their final weights. All weighted depth inversion results are then summed to produce a weighted fusion result. This process is repeated pixel by pixel, ultimately generating an accurate and comprehensive fusion result as the final depth inversion output. This result integrates information from all filtered images, with images with smaller differences playing a greater role in the fusion process. The principle is as follows:

[0087]

[0088] Where: h i (x, y) is the water depth in the inversion results at the position (x, y) in different phases, n is the number of multi-phase images, w i (x, y) is the weight of the inversion results of different phases at the (x, y) position, H i (x, y) is the optimal water depth at the (x, y) position after adaptive weight fusion at different time phases.

[0089] In summary, in the multi-temporal remote sensing image uncontrolled water depth inversion method based on adaptive weight fusion provided by the present invention, by executing steps S1 to S5, high-precision uncontrolled water depth inversion in a complex shallow sea environment can be completed; this multi-temporal remote sensing image uncontrolled water depth inversion method based on adaptive weight fusion is suitable for different offshore distances, different sea areas and different satellite data, and the water depth information obtained based on this method can be directly used in fields such as nearshore marine engineering construction, shipping route selection and coastal ecological research.

[0090] It should be noted that the parts that are not described in detail or in detail in the above scheme, such as the specific implementation process of each preprocessing in step S1 and the implementation process of uncontrolled water depth inversion of each single-phase remote sensing image using the P-DLA model in step S2, are all existing technologies, do not belong to the improvements made by the present invention on the existing technology, nor do they belong to the scope of protection of the technical solution of the present invention. Therefore, they will not be repeated in this article.

[0091] The above descriptions are merely optional embodiments of the present invention and do not limit the patent scope of the present invention. All equivalent structural transformations made using the contents of the present invention specification under the concept of the present invention, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present invention.

Claims

1. A method for uncontrolled water depth inversion from multi-temporal remote sensing images based on adaptive weight fusion, characterized in that: The steps include: S1: Receive multi-temporal remote sensing image data and measured water depth data, and perform preliminary data processing; S2: Invert the preliminary water depth of each single-temporal image using a dual-band log-linear model; S3: Tidal correction and deep water area fitting completion are performed based on the preliminary water depth results in S2; S4: Adaptively calculate the fusion weights of the same-named points at different times based on the reliability of the water depth of the same-named points relative to the fusion coefficient; S5: Based on the adaptive weight fusion algorithm, weighted fusion generates accurate water depth results.

2. The method for uncontrolled water depth inversion based on multi-temporal remote sensing images based on adaptive weight fusion according to claim 1 is characterized in that: In said S1, multi-temporal satellite remote sensing image data is received and pre-processed to obtain true reflectivity information; The measured water depth data are standardized to verify the accuracy of the water depth inversion results.

3. The method for uncontrolled water depth inversion based on multi-temporal remote sensing images based on adaptive weight fusion according to claim 1 is characterized in that: In the S2, different bottom textures are identified by the brightness of shallow water areas in remote sensing images. Based on a dual-band log-linear model, characteristic pixels with different water depths and bottom textures are extracted from multi-temporal remote sensing images to invert the preliminary water depth of each single-temporal image.

4. The method for uncontrolled water depth inversion based on multi-temporal remote sensing images based on adaptive weight fusion according to claim 1 is characterized in that: The S3 includes the following sub-steps: S3-1: Tidal correction is performed on all inversion results, and the instantaneous water depth of each single phase is corrected to the steady-state water depth of the same reference level; S3-2: The missing data are completed by linear fitting of the logarithmic ratio of the blue and green bands.

5. The method for uncontrolled water depth inversion based on multi-temporal remote sensing images based on adaptive weight fusion according to claim 4 is characterized in that: The S3-2 includes the following sub-steps: S3-2-1: Select the temporal images with complete data in deep water areas from multiple remote sensing images as the basic data for establishing the relationship; S3-2-2: For each pixel in the selected complete data phase image, calculate the logarithmic ratio of the blue band and green band reflectance, associate it with the inverted water depth value of the corresponding pixel, and establish a linear relationship between the two through linear regression method; S3-2-3: For phase images with missing data in deep water areas, obtain the blue-green band reflectance at the same-name point, substitute the reflectance into the established linear relationship, and calculate the water depth value corresponding to the point, thereby completing the missing data and making the reverse evolution results of each phase complete.

6. The method for uncontrolled water depth inversion based on multi-temporal remote sensing images based on adaptive weight fusion according to claim 1 is characterized in that: In the S4, based on a dynamic adaptive weighting strategy, the water depths of the synonymous points from different data sources are evaluated and weights are adaptively assigned according to the reliability of each synonymous point to improve the fault tolerance in the inversion process, eliminate the interference of system errors and environmental factors, and achieve higher inversion accuracy.

7. The method for uncontrolled water depth inversion based on multi-temporal remote sensing images based on adaptive weight fusion according to claim 6 is characterized in that: The S4 includes the following sub-steps: S4-1: Based on the data processed in step S3, the coordinates of the multi-temporal remote sensing image are used to match the nearest neighboring points of the same name pixel by pixel. For each group of points of the same name, the water depth data of each temporal phase is extracted. S4-2: Calculate the average water depth of the same-name points Δh as the fusion coefficient, and use it as a reference standard to measure the difference between the inversion results of each image and the overall situation. The principle is: Where Δh is the fusion coefficient, (x, y) represents the coordinates of the pixel, and h i (x,y) is the water depth of the point with the same coordinate (x,y) in different time phases, and n is the number of multi-temporal images. S4-3: Calculate the difference between the water depth and fusion coefficient of the same-name points of the multi-temporal results, and use the inverse of the difference as the preliminary weight; S4-4: Calculate the sum of the preliminary weights of all images and normalize the preliminary weights so that the sum of the final weights of the multi-temporal inversion results is 1; S4-5: Divide the preliminary weight of each image by the total weight to obtain the final weight of each image.

8. The method for uncontrolled water depth inversion based on multi-temporal remote sensing images based on adaptive weight fusion according to claim 7 is characterized in that: The final weight calculation formula in S4-5 is as follows: Where: Δh(x,y) is the fusion coefficient, h i (x,y) is the depth of the result of the multi-temporal image transformation of the same point with coordinates (x,y), n is the number of multi-temporal images, w i (x,y) is the final weight of the i-th image at position (x,y).

9. The method for uncontrolled water depth inversion based on multi-temporal remote sensing images based on adaptive weight fusion according to claim 7, characterized in that: After determining the fusion weights of the data of different time phases in S4 in S5, the same-name points of each time phase are multiplied by their final weights, and then all weighted water depth inversion results are added together to obtain a weighted fusion result; The above steps are repeated pixel by pixel, and finally a fusion result is generated by weighted fusion as the final water depth inversion output.

10. The method for uncontrolled water depth inversion based on multi-temporal remote sensing images based on adaptive weight fusion according to claim 9, characterized in that: The calculation formula in S5 is: Where: h i (x, y) is the water depth in the inversion results at the position (x, y) in different phases, n is the number of multi-phase images, w i (x, y) is the weight of the inversion results of different phases at the (x, y) position, H i (x, y) is the optimal water depth at the (x, y) position after adaptive weight fusion at different time phases.

Images