Shallow water depth remote sensing inversion method based on regional self-adaptation

Abstract

The present application provides a shallow sea water depth remote sensing inversion method based on regional self-adaption, comprising the following steps: S1, data collection and processing; S2, according to the data collected and processed in step S1, comprehensive regional demarcation; S3, model library construction, the water depth remote sensing inversion model is summarized as a single-band model, a double-band model, a multi-band model, a band ratio model, a logarithmic conversion ratio model, a neural network model and a machine learning model; S4, regional model optimization according to the model library constructed in step S3; S5, water depth inversion according to the regional model optimized in step S4. The present application has the beneficial effects that the shallow sea water depth remote sensing inversion method based on regional self-adaption, combined with the existing research basis, comprehensively considers the water depth range, the bottom type and the water quality condition, and establishes a regional self-adaptive water depth remote sensing inversion method, so as to maximize the remote sensing inversion precision of the near sea water depth.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing technology, and in particular relates to a method for remote sensing inversion of shallow water depth based on regional adaptation. Background Technology

[0002] Shallow seas are areas of most frequent interaction between land and sea, and are also crucial for engineering development activities such as land reclamation, port construction, mariculture, and maritime transportation. Conducting shallow sea depth measurements to accurately grasp shallow sea depth information and seabed topography is of great significance for national maritime traffic safety, integrated coastal zone management, and military defense. Traditional depth measurement methods use shipborne detection equipment, but these cannot be used in shallow sea areas and sensitive waters, resulting in a large number of gaps in nearshore shallow sea depth measurements. Remote sensing technology, with its advantages of large-area coverage, rapid updates, and non-direct contact, can serve as a valuable supplement to traditional depth measurement techniques for acquiring depth information in shallow sea areas.

[0003] Water depth remote sensing can be active or passive. Active water depth remote sensing currently relies mainly on airborne LiDAR measurements, while this invention utilizes a passive water depth remote sensing method. The theoretical basis for water depth inversion based on remote sensing technology is as follows: Sunlight passes through the atmosphere into the water body, where it is absorbed or scattered by various components. Some scattered light exits directly to the water surface, while some reaches the bottom, is reflected, and then absorbs and scatters again before exiting the surface. Finally, the optical signal containing water body and depth information is received by the remote sensing sensor. By analyzing the remote sensing image containing water depth information and establishing a functional relationship with the measured water depth, water depth remote sensing inversion can be achieved. The mainstream remote sensing water depth inversion models mainly include three types: theoretical analytical models, semi-theoretical / semi-empirical models, and statistical analysis models. Theoretical analytical models are based on the radiative transfer process of water bodies and have high accuracy, but they have many parameters that are difficult to obtain, making practical application challenging. Currently, the most widely used model is the semi-theoretical and semi-empirical model. This model is derived based on existing prior data and the physical relationship between water depth and the radiance value of remote sensing images. Commonly used models include single-band models, multi-band models, log-linear models, and log-transform ratio models.

[0004] Shallow sea environments are complex, with variations in seabed sediment types leading to different levels of light attenuation at the bottom. Furthermore, shallow seas are significantly influenced by terrestrial materials, resulting in complex nearshore water color characteristics due to suspended sediment, yellow substances, and phytoplankton. Existing models generally suffer from high depth inversion errors in shallower areas (0-5m). Moreover, they fail to adequately refine and filter models based on nearshore sediment type, water quality, and imaging conditions, resulting in poor model portability. Summary of the Invention

[0005] In view of this, the present invention aims to propose a regionally adaptive shallow water depth remote sensing inversion method. Based on existing research, and taking into account water depth range, bottom sediment type and water quality, a regionally adaptive water depth remote sensing inversion method is established to maximize the accuracy of nearshore water depth remote sensing inversion.

[0006] The basic idea of ​​regional adaptation is to select the best model and parameters based on the basic conditions of the depth, bottom sediment and water quality of the water depth inversion area, rather than using a single model to calculate the entire water depth inversion area, so as to minimize the inversion error caused by model differences and parameter differences.

[0007] To achieve the above objectives, the technical solution of the present invention is implemented as follows:

[0008] The region-adaptive shallow water depth remote sensing inversion method includes the following steps:

[0009] S1. Data collection and processing;

[0010] S2. Based on the data collected and processed in step S1, the comprehensive area is delineated.

[0011] S3. Model library construction: The water depth remote sensing inversion models are categorized into single-band models, dual-band models, multi-band models, band ratio models, logarithmic transformation ratio models, neural network models, and machine learning models.

[0012] S4. Optimize regional models based on the model library constructed in step S3;

[0013] S5. Perform water depth inversion based on the regional model selected in step S4.

[0014] Furthermore, step S1 includes the following steps:

[0015] A1. Data Collection: Collect all kinds of data around the study area, including remote sensing image data, measured water depth data, nautical chart data, tidal data and seabed type data;

[0016] A2. Remote sensing image processing: Perform geometric correction, radiometric calibration, flare correction, atmospheric correction, and land-water separation processing on remote sensing image data;

[0017] A3. Water depth data processing: including chart correction and information extraction, as well as tidal correction of water depth information.

[0018] Furthermore, step A2 includes the following steps:

[0019] (1) Radiometric calibration: Through radiometric calibration, the image DN values ​​are converted into apparent radiance values ​​L.

[0020] L = Gain × DN + offset

[0021] Where Gain is the gain coefficient and Offset is the offset, which is obtained from the image header file;

[0022] (2) Solar flare correction: Solar flares on water surfaces are caused by the specular reflection of solar radiation on the water surface, appearing as bright white patches that obscure the true radiation characteristics of the target water body. The Hedley method is used for flare removal.

[0023]

[0024] Among them, L′ i Let L be the radiance of the i-th band after flare removal. i Let L be the radiance of the i-th band before flare removal, θ be the inclination angle of the regression line, and L be the radiance of the i-th band before flare removal. NIR The radiance is in the near-infrared band. This represents the minimum radiance value in the near-infrared band.

[0025] (3) Atmospheric correction: used to eliminate the influence of atmospheric and light factors on the reflectance of ground objects and obtain the true reflectance data of ground objects;

[0026] (4) Land-water segmentation: To prevent land pixels from participating in water depth inversion calculations and improve computational efficiency, land-water segmentation is required. The NDWI index is used for land-water segmentation.

[0027]

[0028] Where, p Green and p NIR These are the reflectance values ​​for the green light band and the near-infrared band.

[0029] Furthermore, step A3 includes the following steps:

[0030] (1) Chart calibration: If it is a paper chart, it is first scanned, and then geometric calibration is performed based on information from kilometers away or a reference base map;

[0031] (2) Chart information extraction: Extract thematic information on depth points and contour lines based on the corrected nautical charts;

[0032] (3) Tidal correction: Sea surface height changes constantly due to the influence of tides. All water depth data from all sources need to be unified to the same tide level at the same time before they can be used. The measured water depth data needs to be corrected based on the difference between the instantaneous tide level at the time of measurement and the instantaneous tide level at the time of remote sensing image capture. The water depth information obtained from the nautical chart needs to be corrected based on the difference between the depth datum of the nautical chart and the instantaneous tide level at the time of remote sensing image capture.

[0033] Furthermore, step S1 includes the following cases:

[0034] B1. Zoning based on water depth: The study area is divided into zones according to 5-meter intervals based on water depth information obtained from nautical charts and measured water depth data.

[0035] B2. Zoning based on sediment type and water quality: Based on the zoning based on water depth, the zoning is further subdivided by combining the sediment type data and specific water quality conditions of the study area;

[0036] B3. Comprehensive Zoning: The boundaries of the detailed zoning are expanded outward by 10 pixels in both the shallow and deep water areas. The boundary at the 0-meter water depth can be left unexpanded. Any land-to-land or other unreasonable situations after expansion are comprehensively processed to obtain the final comprehensive zoning result, which serves as the subdivided area for subsequent water depth inversion.

[0037] Furthermore, in step S1, the specific model form is as follows:

[0038] C1, Single-band model

[0039] M = A0 + A1X1

[0040] Where M is the retrieved water depth, A0 and A1 are model coefficients, and X1 is the spectral value of a certain band;

[0041] C2, Dual-band model

[0042] M = A0 + A1X1 + A2X2

[0043] Where M is the inverted water depth value, A0, A1 and A2 are model coefficients, and X1 and X2 are spectral values ​​of a certain band. Alternatively, two bands with strong penetrating power into the water body can be selected for ratio processing.

[0044] C3, Multiband Model

[0045] M = A0 + A1X1 + A2X2 + ... + A n X n

[0046] Where M is the inverted water depth value, and A0, A1 and A n For model coefficients, X1, X2, and X... n This refers to the spectral value of a certain wavelength band;

[0047] C4, Band Ratio Model

[0048]

[0049] Where M is the retrieved water depth, A0 and A1 are model coefficients, and X1 and X2 are the spectral values ​​of the two bands, respectively.

[0050] C5, Logarithmic Transformation Ratio Model

[0051]

[0052] Where M is the inverted water depth value, A0 and A1 are model coefficients, and R... ω (X1) and R ω (X2) represents the reflectance values ​​of the two bands, and a, b, m, and n are model adjustment factors.

[0053] C6. Neural Network Model

[0054] Including BP neural networks, LSTM, CNN, RNN, DBN, and RBM, a dataset spatially matched with remote sensing imagery and measured water depth data is first established. This dataset is then randomly divided into training and testing data for model training and testing. An optimal LSTM water depth retrieval model is then established. The operational formula for the LSTM neuron is:

[0055] i t =Sigmoid(W Xi ·X t +W Hi ·H t-1 +b i )

[0056] f t =Sigmoid(W Xf ·X t +W Hf ·H t-1 +b f )

[0057]

[0058]

[0059]

[0060] in, X represents the Hadamard product. t H represents the input to t neurons; t-1 W is the output of the (t-1)th neuron. Xi W Xf W XC W Xo For different operational bands X t Corresponding weights, W Hi W Hf W HC W Ho For different operational bands H t-1 Corresponding weight; b i b fb C b o For the corresponding operation bias value, i t f t o t Separately control input, forgetting, and output gating;

[0061] C t Let t be the cell state of the t-th neuron. It combines the cell state of the previous neuron with the new data input to the current neuron to form a new cell state.

[0062] Furthermore, step S1 includes the following steps:

[0063] D1. Selection of water depth points: Select water depth points according to the zoning range, and then randomly divide the water depth points into two parts according to the model building requirements. One part is used for model building and the other part is used for model verification.

[0064] D2. Sensitive band selection: Sensitive bands for regional water depth inversion are determined by analyzing the spectral characteristics of remote sensing images or by performing statistical correlation analysis between spectral data and water depth data.

[0065] D3. Model Construction: Based on the model, construct water depth points, and successively construct single-band model, dual-band model, multi-band model, band ratio model, logarithmic transformation ratio model, neural network model and machine learning model, and determine the specific parameters of each type of model;

[0066] D4. Model Validation and Optimization: Water depth inversion is performed on the regional remote sensing images using each constructed model. Water depth points are validated using the models, and the accuracy of the model inversion results is evaluated from both mean absolute error and mean relative error perspectives. The model with the highest accuracy is selected as the inversion model for this region.

[0067]

[0068]

[0069] Among them, Z i Let ΔZ be the water depth value at the i-th verification point. i Let be the water depth error value at the i-th verification point.

[0070] Furthermore, step S1 includes the following steps:

[0071] E1. Regional Water Depth Inversion: Using the optimized regional water depth inversion model, regional water depth inversion is performed to obtain the regional water depth inversion results, which are D... Z1 D Z2 D Z3 ...;

[0072] E2. Regional Water Depth Fusion: The regional water depth inversion results are stitched together. First, results with different resolutions need to be unified by using the "remove-restore" method to form a constant resolution. Then, the water depth values ​​of the boundary overlapping areas obtained through outward expansion are linearly weighted. The weight is determined based on the distance from the boundary of the inversion area; the closer to the center of the area and the farther from the boundary, the greater the weight. The water depth values ​​in the overlapping area are determined by the following formula:

[0073] M = M l ×f l +M r ×(1-f l )

[0074]

[0075] Where M is the inverted water depth value at a certain location within the overlapping region, M l Mr and f are the water depth inversion values ​​at this location within the overlapping area of ​​two adjacent regions, respectively. l Let r be the weight of one of the regions, r be the distance of that location from the region boundary, and R be the total width of the overlapping region;

[0076] E3. Comprehensive Accuracy Analysis: The accuracy of the fused water depth inversion results is evaluated from both MAE and MRE perspectives using all models to verify the water depth points.

[0077] Furthermore, this solution discloses an electronic device, including a processor and a memory communicatively connected to the processor and used to store executable instructions of the processor, wherein the processor is used to execute a region-adaptive shallow water depth remote sensing inversion method.

[0078] Furthermore, this solution discloses a computer-readable storage medium storing a computer program, which, when executed by a processor, implements a region-adaptive shallow water depth remote sensing inversion method.

[0079] Compared with existing technologies, the shallow water depth remote sensing inversion method based on region adaptation described in this invention has the following advantages:

[0080] (1) The shallow water depth remote sensing inversion method based on regional adaptation described in this invention combines existing research foundations and comprehensively considers water depth range, bottom sediment type and water quality to establish a regional adaptive water depth remote sensing inversion method in order to maximize the accuracy of nearshore water depth remote sensing inversion.

[0081] (2) The shallow water depth remote sensing inversion method based on regional adaptation described in this invention is based on the idea of ​​selecting the best model and parameters according to the basic conditions of the depth, bottom sediment and water quality of the water depth inversion area, rather than using a single model to calculate the entire water depth inversion area, so as to minimize the inversion error caused by model differences and different parameters. Attached Figure Description

[0082] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0083] Figure 1 This is a schematic diagram of the shallow water depth remote sensing inversion method based on regional adaptation as described in an embodiment of the present invention. Detailed Implementation

[0084] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0085] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0086] like Figure 1 As shown, this scheme aims to combine existing research foundations and comprehensively consider water depth range, seabed type and water quality to establish a regionally adaptive water depth remote sensing inversion method, in order to maximize the accuracy of nearshore water depth remote sensing inversion.

[0087] The basic idea of ​​regional adaptation is to select the best model and parameters based on the basic conditions of the depth, bottom sediment and water quality of the water depth inversion area, rather than using a single model to calculate the entire water depth inversion area, so as to minimize the inversion error caused by model differences and parameter differences.

[0088] Specific steps:

[0089] Step 1: Data Collection and Processing

[0090] 1-1: Data Collection: Collect a wide range of data around the study area, including remote sensing imagery, measured water depth data, nautical charts, tidal data, and seabed type data.

[0091] 1-2: Remote sensing image processing: For remote sensing image data, the main processing methods include geometric correction, radiometric calibration, flare correction, atmospheric correction, and land-water separation.

[0092] (1) Radiometric calibration: The signals received by satellite sensors are affected by the absorption, scattering, reflection, refraction and transmission of aerosols, cloud particles and water vapor in the atmosphere, which causes distortion of remote sensing image information. Radiometric calibration is required to convert the image DN value into the apparent radiance value L.

[0093] L = Gain × DN + Offset

[0094] Gain is the gain coefficient, and Offset is the offset, which are obtained from the image header file.

[0095] (2) Solar flare correction. Solar flares on water surfaces are caused by the specular reflection of solar radiation entering the water, appearing as bright white patches that obscure the true radiation characteristics of the target water body. The existing Hedley method can be used for flare removal.

[0096]

[0097] Among them, L′ i Let L be the radiance of the i-th band after flare removal. i Let L be the radiance of the i-th band before flare removal, θ be the inclination angle of the regression line, and L be the radiance of the i-th band before flare removal. NIR The radiance is in the near-infrared band. This represents the minimum radiance value in the near-infrared band.

[0098] (3) Atmospheric Correction: Atmospheric correction is mainly to eliminate the influence of atmospheric and illumination factors on the reflectance of ground objects and obtain the true reflectance data of ground objects. This includes absolute atmospheric correction methods such as Mortran, Lowtran, Flaash, 6S, and dark pixel correction, as well as relative correction methods such as histogram matching and statistically invariant target methods. If the study area has only one image, absolute atmospheric correction methods should be chosen whenever possible. If the study area contains multiple images from different periods, relative atmospheric correction methods are recommended, and normalization processing should be performed.

[0099] (4) Land-water separation. To prevent land pixels from participating in water depth inversion calculations and to improve computational efficiency, land-water separation is required. The NDWI index is used for land-water separation.

[0100]

[0101] Where, p Green and p NIR These are the reflectance values ​​for the green light band and the near-infrared band.

[0102] 1-3: Water depth data processing: including chart correction and information extraction, as well as tidal correction of water depth information, etc.

[0103] (1) Chart correction: If it is a paper chart, first scan it, and then perform geometric correction based on the information at kilometer distance or the reference base map.

[0104] (2) Extraction of nautical chart information: Extract thematic information such as depth points and contour lines based on the corrected nautical chart.

[0105] (3) Tidal Correction: Sea level is constantly changing due to the influence of tides, so all water depth data from all sources need to be consistent with the tide level at the same time before they can be used. Measured water depth data needs to be corrected based on the difference between the instantaneous tide level at the time of measurement and the instantaneous tide level at the time of remote sensing image capture. Water depth information obtained from nautical charts needs to be corrected based on the difference between the depth datum of the nautical chart and the instantaneous tide level at the time of remote sensing image capture.

[0106] Step 2: Comprehensive Area Delineation

[0107] 2-1: Zoning Based on Water Depth: Based on water depth information obtained from nautical charts and measured water depth data, the study area is divided into zones at 5-meter intervals. The zoning boundaries are defined as follows: Z d1 ∈[0,5), Z d2 ∈[5,10), Z d3 ∈[10,15), Z d4 ∈[15,20), Z d5 ∈[20,25), Z d6 ∈[25,30)……

[0108] 2-2: Zoning Based on Sediment Type and Water Quality: Building upon the zoning based on water depth, the zoning is further subdivided by considering sediment type data and specific water quality conditions within the study area. For example, in Z... d1 Within a zone, if there are significantly different substrate types or some areas exhibit marked differences in water quality, then zone Z should be categorized according to the substrate type or water quality. d1 The zones are then further subdivided. Finally, detailed zones are obtained based on the overall substrate type and water quality, and are named Z1, Z2, Z3, and so on.

[0109] 2-3: Comprehensive Zoning: The boundaries of the detailed zoning are extended outward by 10 pixels (remote sensing image) in both the shallow and deep water areas. The boundary at the 0-meter water depth can be left unexpanded. Any land-to-land or other unreasonable situations after the expansion are comprehensively processed to obtain the final comprehensive zoning result, which serves as the subdivided area for subsequent water depth inversion.

[0110] Step 3: Model Library Construction

[0111] Based on an analysis of the current state of research both domestically and internationally, water depth remote sensing inversion models are categorized into single-band models, dual-band models, multi-band models, band ratio models, logarithmic transformation ratio models, neural network models, and machine learning models. The specific model forms are as follows:

[0112] (1) Single-band model

[0113] M = A0 + A1X1

[0114] M is the inverted water depth value, A0 and A1 are model coefficients, and X1 is the spectral value of a certain band (which can be the value after differential or logarithmic processing).

[0115] (2) Dual-band model

[0116] M = A0 + A1X1 + A2X2

[0117] M is the inverted water depth value, A0, A1 and A2 are model coefficients, and X1 and X2 are spectral values ​​of a certain band (which can be obtained after differential or logarithmic processing). Alternatively, two bands with strong penetrating power into the water body can be selected and their values ​​can be compared.

[0118] (3) Multiband model

[0119] M = A0 + A1X1 + A2X2 + ... + A n X n

[0120] M represents the inverted water depth value, and A0, A1, and A n For model coefficients, X1, X2, and X... n This represents the spectral value for a certain wavelength band (which can be the value after differential or logarithmic processing).

[0121] (4) Band ratio model

[0122]

[0123] M represents the retrieved water depth, A0 and A1 are model coefficients, and X1 and X2 are the spectral values ​​for the two bands, respectively.

[0124] (5) Logarithmic transformation ratio model

[0125]

[0126] M is the inverted water depth value, A0 and A1 are model coefficients, and R... ω (X1) and R ω (X2) represents the reflectance values ​​of the two bands, and a, b, m, and n are model adjustment factors.

[0127] (6) Neural Network Model

[0128] There are various neural networks, including BP neural networks, LSTM, CNN, RNN, DBN, and RBM, among which LSTM is the most commonly used. LSTM models can effectively solve the gradient explosion problem. First, a dataset spatially matched with remote sensing imagery and measured water depth data is established. Then, this dataset is randomly divided into training and testing data for model training and testing, establishing the optimal LSTM water depth inversion model. The computational formula for LSTM neurons is:

[0129] i t =Sigmoid(W Xi ·X t +W Hi ·H t-1 +b i )

[0130] f t =Sigmoid(W Xf ·X t +W Hf ·H t-1 +b f )

[0131]

[0132]

[0133]

[0134] in, X represents the Hadamard product. t H represents the input to t neurons; t-1 W is the output of the (t-1)th neuron. Xi W Xf W XC W Xo For different operational bands X t Corresponding weights, W Hi W Hf W HC W Ho For different operational bands H t-1 Corresponding weight; b i b f b C b o For the corresponding operation bias value, i t f t o t Separately control input, forget, and output gating; C t Let t be the cell state of the t-th neuron. It combines the cell state of the previous neuron with the new data input to the current neuron to form a new cell state.

[0135] Step 4: Optimization of Region Model

[0136] 4-1: Selection of water depth points: Select water depth points according to the zoning range, and then randomly divide the water depth points into two parts according to the model building requirements. One part is used for model building, and the other part is used for model validation.

[0137] 4-2: Sensitive band selection: Sensitive bands for regional water depth inversion are determined by analyzing the spectral characteristics of remote sensing images or by performing statistical correlation analysis between spectral data and water depth data.

[0138] 4-3: Model Construction: Based on the model, water depth points are constructed in sequence, including single-band model, dual-band model, multi-band model, band ratio model, logarithmic transformation ratio model, neural network model, and machine learning model, and the specific parameters of each type of model are determined.

[0139] 4-4: Model Validation and Optimization: Water depth inversion was performed on the regional remote sensing images one by one using each of the constructed models. Water depth points were validated using the models. The accuracy of the model inversion results was evaluated from two aspects: Mean Absolutely Error (MAE) and Mean Relative Error (MRE). The model with the best accuracy was selected as the inversion model for this region.

[0140]

[0141]

[0142] Among them, Z i Let ΔZ be the water depth value at the i-th verification point. i Let be the water depth error value at the i-th verification point.

[0143] Step 5: Water Depth Inversion

[0144] 5-1: Regional Water Depth Inversion: The optimized regional water depth inversion model is used to perform regional water depth inversion, yielding regional water depth inversion results. These are D... Z1 D Z2 D Z3 ...

[0145] 5-2: Regional Water Depth Fusion: To stitch together the regional water depth inversion results, it is first necessary to unify the results at different resolutions using the "remove-restore" method to achieve a constant resolution. Then, the water depth values ​​in the overlapping boundary regions obtained through outward expansion are linearly weighted. The weights are determined based on the distance from the boundary of the inversion region; the closer to the region center and the farther from the boundary, the greater the weight. The water depth values ​​within the overlapping region are determined using the following formula:

[0146] M = M l ×f l +M r ×(1-f l )

[0147]

[0148] Where M is the inverted water depth value at a certain location within the overlapping region, M l Mr and f are the water depth inversion values ​​at this location within the overlapping area of ​​two adjacent regions, respectively. l Let r be the weight of one of the regions, r be the distance of that location from the region boundary, and R be the total width of the overlapping region.

[0149] 5-3: Comprehensive Accuracy Analysis: Using all models, the accuracy of the fused water depth inversion results is evaluated from both MAE and MRE perspectives.

[0150] Those skilled in the art will recognize that the units and method steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0151] In the several embodiments provided in this application, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the division of units described above is merely a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. The aforementioned units may or may not be physically separated. The components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of the present invention according to actual needs.

[0152] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

[0153] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A shallow water depth remote sensing inversion method based on regional adaptation, characterized in that, Includes the following steps: S1. Data collection and processing; S2. Based on the data collected and processed in step S1, the comprehensive area is delineated. S3. Model library construction: The water depth remote sensing inversion models are categorized into single-band models, dual-band models, multi-band models, band ratio models, logarithmic transformation ratio models, neural network models, and machine learning models. S4. Optimize regional models based on the model library constructed in step S3; S5. Perform water depth inversion based on the regional model selected in step S4; Step S2 specifically includes the following steps: B1. Zoning based on water depth: The study area is divided into zones according to 5-meter intervals based on water depth information obtained from nautical charts and measured water depth data. B2. Zoning based on sediment type and water quality: Based on the zoning based on water depth, the zoning is further subdivided by combining the sediment type data and specific water quality conditions of the study area; B3. Comprehensive Zoning: The boundaries of the detailed zoning are expanded outward by 10 pixels in both the shallow and deep water areas. The boundary at the 0-meter water depth can be left unexpanded. The situation where there is land after the expansion is comprehensively processed to obtain the comprehensive zoning result, which serves as the subdivision area for subsequent water depth inversion. Step S4 specifically includes the following steps: D1. Selection of water depth points: Select water depth points according to the zoning range, and then randomly divide the water depth points into two parts according to the model building requirements. One part is used for model building and the other part is used for model verification. D2. Sensitive band selection: Sensitive bands for regional water depth inversion are determined by analyzing the spectral characteristics of remote sensing images or by performing statistical correlation analysis between spectral data and water depth data. D3. Model Construction: Based on the model, construct water depth points, and successively construct single-band model, dual-band model, multi-band model, band ratio model, logarithmic transformation ratio model, neural network model and machine learning model, and determine the specific parameters of each type of model; D4. Model Validation and Optimization: Water depth inversion is performed on the regional remote sensing images using each constructed model. Water depth points are validated using the models, and the accuracy of the model inversion results is evaluated from both mean absolute error and mean relative error perspectives. The model with the highest accuracy is selected as the inversion model for this region. in, For the first Water depth values ​​at each verification point For the first Water depth error values ​​at each verification point; Step S5 specifically includes the following steps: E1. Regional Water Depth Inversion: Using the optimized regional water depth inversion model, regional water depth inversion is performed to obtain the regional water depth inversion results, which are as follows: , , ...; E2. Regional Water Depth Fusion: The regional water depth inversion results are stitched together. First, results with different resolutions need to be unified by using the "remove-restore" method to form a constant resolution. Then, the water depth values ​​of the boundary overlapping areas obtained through outward expansion are linearly weighted. The weight is determined based on the distance from the boundary of the inversion area; the closer to the center of the area and the farther from the boundary, the greater the weight. The water depth values ​​in the overlapping area are determined by the following formula: M=M l ×f l +M r ×(1-f l ) Where M is the inverted water depth value at a certain location within the overlapping region, M l Mr and Mr are the water depth inversion values ​​at this location within the overlapping area of ​​two adjacent regions, respectively. Let r be the weight of one of the regions, r be the distance of that location from the region boundary, and R be the total width of the overlapping region; E3. Comprehensive Accuracy Analysis: The accuracy of the fused water depth inversion results is evaluated from both MAE and MRE perspectives using all models to verify the water depth points.

2. The shallow water depth remote sensing inversion method based on regional adaptation according to claim 1, characterized in that, Step S1 includes the following steps: A1. Data Collection: Collect all kinds of data around the study area, including remote sensing image data, measured water depth data, nautical chart data, tidal data and seabed type data; A2. Remote sensing image processing: Perform geometric correction, radiometric calibration, flare correction, atmospheric correction, and land-water separation processing on remote sensing image data; A3. Water depth data processing: including chart correction and information extraction, as well as tidal correction of water depth information.

3. The shallow water depth remote sensing inversion method based on regional adaptation according to claim 2, characterized in that, Step A2 includes the following steps: (1) Radiometric calibration: Through radiometric calibration processing, the image is... Value converted to apparent radiance value : in, This is the gain coefficient. The offset is obtained from the image header file; (2) Solar flare correction: Solar flares on water surfaces are caused by the mirror reflection of solar radiation entering the water, appearing as bright white patches that obscure the true radiation characteristics of the target water body. Therefore, the following correction method is used: Methods for removing solar flares: in, For the first Radiance after flare removal in each band For the first Remove the radiance of each band before the flare. The inclination angle of the tropics. The radiance is in the near-infrared band. This represents the minimum radiance value in the near-infrared band. (3) Atmospheric correction: used to eliminate the influence of atmospheric and light factors on the reflection of ground objects and obtain the true reflectance data of ground objects; (4) Land-water segmentation: To prevent land pixels from participating in water depth inversion calculations and improve computational efficiency, land-water segmentation is required. The index is used to separate land and water: in, and These are the reflectance values ​​for the green light band and the near-infrared band.

4. The shallow water depth remote sensing inversion method based on regional adaptation according to claim 2, characterized in that, Step A3 includes the following steps: (1) Chart calibration: If it is a paper chart, it is first scanned, and then geometric calibration is performed based on information from kilometers away or a reference base map; (2) Chart information extraction: Thematic information on depth points and contour lines is extracted based on the corrected charts; (3) Tidal correction: All water depth data from all sources must be unified to the same tide level at the same time before they can be used. The measured water depth data is corrected based on the difference between the instantaneous tide level at the time of measurement and the instantaneous tide level at the time of remote sensing image capture. The water depth information obtained from the nautical chart is corrected based on the difference between the depth datum of the nautical chart and the instantaneous tide level at the time of remote sensing image capture.

5. The shallow water depth remote sensing inversion method based on regional adaptation according to claim 1, characterized in that, In step S3, the specific model form is as follows: C1, Single-band model in, To retrieve the water depth value, and These are the model coefficients. This refers to the spectral value of a certain wavelength band; C2, Dual-band model in, To retrieve the water depth value, and These are the model coefficients. and It is the spectral value of a certain band, or you can choose two bands with strong penetrating power to water and make a ratio. C3, Multiband Model in, To retrieve the water depth value, and These are the model coefficients. , and This refers to the spectral value of a certain wavelength band; C4, Band Ratio Model in, To retrieve the water depth value, and These are the model coefficients. and These are the spectral values ​​for the two wavebands, respectively. C5, Logarithmic Transformation Ratio Model in, To retrieve the water depth value, and These are the model coefficients. and These are the reflectance values ​​for the two wavebands, respectively. This is a model adjustment factor; C6. Neural Network Model Including BP neural networks, LSTM, CNN, RNN, DBN, and RBM, a dataset spatially matched with remote sensing imagery and measured water depth data is first established. This dataset is then randomly divided into training and testing data for model training and testing. An optimal LSTM water depth retrieval model is then established. The operational formula for the LSTM neuron is: Where "°" represents the Hadamard product, X t H represents the input to t neurons; t-1 W is the output of the (t-1)th neuron. Xi W Xf W XC W Xo For different operational bands X t Corresponding weights, W Hi W Hf W HC W Ho For different operational bands H t-1 Corresponding weight; b i b f b C b o For the corresponding operation bias value, i t f t o t Separately control input, forgetting, and output gating; C t Let t be the cell state of the t-th neuron. It combines the cell state of the previous neuron with the new data input to the current neuron to form a new cell state.

6. An electronic device, comprising a processor and a memory communicatively connected to the processor and used for storing processor-executable instructions, characterized in that: The processor is used to execute the region-adaptive shallow water depth remote sensing inversion method according to any one of claims 1-5.

7. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by the processor, it implements the shallow water depth remote sensing inversion method based on regional adaptation as described in any one of claims 1-5.

Images