Remote sensing inversion early warning method and system for suspended sediment concentration in dredging area

Abstract

The present application provides a dredging area suspended sediment concentration remote sensing inversion early warning method and system, relates to the field of ocean remote sensing and environmental engineering monitoring, and the method comprises the following steps: obtaining a training sample set comprising characteristic band reflectivity, construction parameters and hydrological parameters; constructing and training an inversion model; the attention mechanism layer of the inversion model performs weighted fusion on the characteristic band reflectivity, construction parameters and hydrological parameters to obtain a weighted feature vector; the inversion layer obtains a suspended sediment concentration value according to the weighted feature vector; the current characteristic spectral band, construction parameters and hydrological parameters are input into the trained inversion model to predict the suspended sediment concentration; different levels of dynamic early warning thresholds are set, corresponding level early warning is carried out according to the predicted suspended sediment concentration and the early warning threshold, and the early warning is diffused to the entire ecological sensitive area. The present application realizes accurate and efficient monitoring of suspended sediment diffusion.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of ocean remote sensing and environmental engineering monitoring, in particular to a remote sensing inversion early warning method and system for suspended sediment concentration in dredging areas. BACKGROUND

[0002] The dredging operation in the construction of large-scale cross-sea bridges and other marine engineering projects will disturb the seabed sediments intensively, resulting in a sharp increase in the suspended sediment concentration in the surrounding water area, and the diffusion and migration under the action of hydrodynamic force. Such high-concentration suspended sediment will reduce water transparency, affect the photosynthesis of aquatic plants, and interfere with the feeding, respiration and reproduction processes of aquatic organisms, posing a serious threat to adjacent ecologically sensitive areas.

[0003] In the prior art, remote sensing monitoring of suspended sediment concentration relies on a single satellite data source and a static machine learning model, and only a statistical relationship between reflectivity and concentration is established. Such method has obvious limitations: first, the spatial scale of dredging engineering is relatively small, but the temporal and spatial dynamic changes of the suspended sediment plume are extremely fast, and traditional large-area, low-time-resolution methods are difficult to capture its instantaneous changes; second, the model only relies on spectral information, without incorporating key physical mechanisms leading to sediment diffusion, such as water flow, waves and construction intensity, resulting in unclear physical meaning of the inversion results and easy lag; third, the existing methods generally lack direct linkage with ecological protection targets, and cannot provide immediate and intuitive decision support for engineering environmental management. SUMMARY

[0004] The main purpose of the present application is to provide a remote sensing inversion early warning method and system for suspended sediment concentration in dredging areas, to realize accurate and efficient monitoring of suspended sediment diffusion.

[0005] The present application is implemented by the following technical solutions:

[0006] The remote sensing inversion early warning method for suspended sediment concentration in dredging areas comprises the following steps:

[0007] Step S1, obtaining a training sample set, the training samples in the training sample set including feature band reflectivity sensitive to suspended sediment concentration obtained according to historical images of the dredging area, construction parameters and hydrological parameters of the dredging area on-site synchronized with the historical images;

[0008] Step S2, constructing and training an inversion model, the inversion model including an attention mechanism layer and an inversion layer, the attention mechanism layer performing weighted fusion on the feature band reflectivity, the construction parameters and the hydrological parameters to obtain a weighted feature vector, the inversion layer obtaining a suspended sediment concentration value according to the weighted feature vector, the training sample set being used to train the inversion model, and a loss function L being used in the training, wherein total =L MSE +λL eco , wherein L MSETo predict the mean square error of the concentration and the measured concentration, L eco is an ecological penalty term related to the predicted concentration, and λ is a penalty coefficient.

[0009] Step S3, obtaining a remote sensing image of the current dredging area, processing the image to obtain the current characteristic band reflectivity, and obtaining the current construction parameters and hydrological parameters, inputting the current characteristic spectral band, construction parameters and hydrological parameters into the trained inversion model to predict the suspended sediment concentration;

[0010] Step S4, setting different levels of dynamic early warning thresholds, the early warning thresholds are related to the dredging power, according to the predicted suspended sediment concentration and the early warning thresholds, corresponding level early warning is carried out, and the early warning is diffused to the whole ecological sensitive area, the radius of the ecological sensitive area is related to the hydrological parameters and the dredging time.

[0011] Further, in steps S1 and S3, the image of the dredging area is a hyperspectral remote sensing image obtained by remote sensing, the image is radiometrically calibrated and atmospherically corrected to obtain surface reflectance data, and based on Pearson correlation analysis, the characteristic band reflectivity sensitive to suspended sediment concentration is extracted from the surface reflectance data.

[0012] Further, the construction parameters include dredging ship power, number of dredging ships, dredging depth and sediment particle roundness, and the hydrological parameters include water flow rate and water wave height.

[0013] Further, in step S2, in the inversion model, the weighted feature vector obtained by the attention mechanism layer is represented as wherein, is the attention weight, P is the dredging power, which is determined by the dredging ship power and the number of dredging ships, is the dredging depth, is the sediment particle roundness, is the characteristic band reflectivity sequence, is the water flow rate, is the water wave height, is the Sigmoid function, and are learnable parameters.

[0014] Further, in step S2, the inversion layer obtains the suspended sediment concentration value according to the formula wherein, W2, b2 are learnable parameters, C max is the upper limit of the suspended sediment concentration.

[0015] Further, in step S2, in the loss function, the mean square error L MSE is represented as ​, ecological penalty term L eco is expressed as , wherein n represents the number of training samples participating in the calculation of the mean square error, is the predicted concentration corresponding to the i-th training sample, is the measured concentration corresponding to the i-th training sample, S represents a set of sampling points in the ecological sensitive area, the i-th sampling point corresponds to the i-th training sample, is the concentration threshold of the ecological sensitive area.

[0016] Further, in the step S4, the blue warning threshold , the yellow warning threshold and the red warning threshold are set in turn, wherein P is the dredging power.

[0017] Further, in the step S4, the radius of the ecological sensitive area is expressed as , wherein T is the dredging duration, and g is the acceleration of gravity.

[0018] The present application also realizes the following technical solutions:

[0019] The remote sensing inversion early warning system for suspended sediment concentration in the dredging area is used to implement the remote sensing inversion early warning method for suspended sediment concentration in the dredging area as described in any one of the above, and comprises a data acquisition module, a prediction module and an alarm module. The acquisition module is used to obtain hyperspectral remote sensing images of the dredging area by using a drone, to obtain the current characteristic band reflectivity by processing the images, and to obtain the current construction parameters and hydrological parameters. The characteristic band reflectivity, the construction parameters and the hydrological parameters are jointly input into the prediction module. The prediction module is a trained inversion model. The training data used to train the inversion model include the characteristic band reflectivity sensitive to the suspended sediment concentration obtained according to the historical images of the dredging area, the construction parameters and the hydrological parameters of the dredging area on site synchronized with the historical images. The inversion model comprises an attention mechanism layer and an inversion layer. The attention mechanism layer performs weighted fusion on the characteristic band reflectivity, the construction parameters and the hydrological parameters to obtain a weighted feature vector. The inversion layer obtains the suspended sediment concentration value according to the weighted feature vector. The loss function used during training is L total =L MSE +λL eco , wherein L MSE is the mean square error of the predicted concentration and the measured concentration, L eco is the ecological penalty term related to the predicted concentration, and λ is a penalty coefficient. The alarm module sets different levels of warning thresholds, the warning thresholds are related to the dredging ship power, the corresponding level of warning is performed according to the predicted suspended sediment concentration of the prediction module and the warning threshold, and the warning is diffused to the entire ecological sensitive area. The radius of the ecological sensitive area is related to the hydrological parameters and the dredging duration.

[0020] From the above description of the present application, compared with the prior art, the present application has the following beneficial effects:

[0021] The present application constructs an inversion model including an attention mechanism layer and an inversion layer, trains the inversion model using training data including feature band reflectivity, construction parameters and hydrological parameters, uses a loss function including an ecological penalty term during training, sets different levels of warning thresholds in combination with dredging power, performs corresponding level warning according to the suspended sediment concentration predicted by the inversion model and the warning threshold, and diffuses the warning to the entire ecological sensitive area, solves the warning lag problem caused by strong spatio-temporal dynamics in dredging engineering through dynamic suspended sediment concentration inversion and warning of multi-source data fusion, realizes accurate and efficient monitoring of suspended sediment diffusion, and forward-looking warning of ecological risk, which is crucial for green construction and ecological protection of marine engineering. BRIEF DESCRIPTION OF DRAWINGS

[0022] The present application will be further described below in combination with the drawings and specific embodiments.

[0023] Figure 1 The flowchart of the present application.

[0024] Figure 2 The spatial distribution diagram of the suspended sediment concentration of the present application.

[0025] Figure 3 The ecological sensitive area schematic diagram of the present application. DETAILED DESCRIPTION

[0026] The present application will be further described below in combination with the drawings and specific embodiments.

[0027] As shown in the drawings, Figure 1 The remote sensing inversion and warning method of the suspended sediment concentration in the dredging area includes the following steps:

[0028] Step S1, obtaining a training sample set, the training samples in the training sample set including feature band reflectivity sensitive to the suspended sediment concentration obtained according to historical images of the dredging area, construction parameters and hydrological parameters of the dredging area on-site synchronous with the historical images;

[0029] The hyperspectral remote sensing image is a satellite image taken by a multispectral camera carried on a UAV, the hyperspectral remote sensing image is radiometrically calibrated and atmospherically corrected to obtain the ground reflectivity data, more specifically, the radiometric calibration converts the original digital value of the image into apparent radiance, and the atmospheric correction is used to eliminate the atmospheric absorption and scattering effect, thereby obtaining the ground reflectivity data. Among them, the radiometric calibration and atmospheric correction are both prior art.

[0030] Based on the Pearson correlation analysis, several feature band reflectances most sensitive to the suspended sediment concentration are extracted from the surface reflectance data. Specifically, the spectral reflectance corresponding to different suspended sediment concentrations at a certain wavelength is obtained, and the corresponding spectral reflectance and suspended sediment concentration are taken as a data pair. The Pearson correlation coefficient at this fixed wavelength is calculated according to all data pairs. Repeat the process to calculate the Pearson correlation coefficient at all wavelengths. The wavelengths with a Pearson correlation coefficient greater than a certain threshold are taken as the feature band, and the corresponding reflectance is taken as the feature band reflectance.

[0031] For example, for the wavelength λ = 550 nm, there are 5 sets of data pairs: suspended sediment concentration Y i 0.1, 0.3, 0.5, 0.7, 0.9 kg / m3, and the corresponding reflectance X i 0.25, 0.30, 0.38, 0.45, 0.52, according to the formula The Pearson correlation coefficient r ≈ 0.998 at this wavelength is obtained, which is close to 1, indicating that the spectral reflectance at 550 nm wavelength has a strong positive linear correlation with the suspended sediment concentration, and it is also greater than the set correlation coefficient threshold, so it is taken as the sensitive band for inverting the suspended sediment concentration.

[0032] The construction parameters include the dredging ship power, the number of dredging ships, the dredging depth, and the sediment particle roundness. The hydrological parameters include the water flow rate and the water wave height.

[0033] Step S2, constructing and training the inversion model, the inversion model includes an attention mechanism layer and an inversion layer, the attention mechanism layer weights and fuses the feature band reflectance, the construction parameters and the hydrological parameters to obtain a weighted feature vector, the inversion layer obtains the suspended sediment concentration value according to the weighted feature vector, and the inversion model is trained using the training sample set. The loss function used in training is L total =L MSE +λL eco , wherein L MSE is the mean square error of the predicted concentration and the measured concentration, L eco is an ecological penalty term related to the predicted concentration, and λ is a penalty coefficient.

[0034] The inversion model also includes an input layer at the front end of the attention mechanism layer. During training, the input layer receives the feature band reflectance, the construction parameters and the hydrological parameters and quantizes them into an input vector The attention mechanism layer is connected to the output end of the input layer, and the weighted feature vector obtained by the attention mechanism layer is represented as , wherein is the attention weight, P is the dredging power (unit: kW) determined by the dredging ship power and the number of dredging ships, is the dredging depth (unit: m), is the roundness of the sediment particle, is the sequence of characteristic band reflectivity, is the flow velocity of the water body (unit: m / s), is the wave height of the water body (unit: m), is the Sigmoid function, and are learnable parameters, A is the projected area of the sediment particle, and L is the projected perimeter of the sediment particle, 1 represents a perfect circle.

[0035] The inversion layer includes a hidden layer that performs linear transformation, ReLU activation, and Dropout regularization, which is obtained according to the formula obtains the suspended sediment concentration value where W2 and b2 are learnable parameters, C max = 2000 mg / L.

[0036] An ecological sensitive area penalty term is added to the loss function to directly link the inversion result to the ecological sensitive area, so as to realize quantitative and spatial ecological risk real-time early warning. Specifically, the mean square error L MSE is expressed as The ecological penalty term L eco is expressed as where n represents the number of training samples participating in the mean square error calculation, is the predicted concentration corresponding to the i-th training sample, is the measured concentration corresponding to the i-th training sample, and S represents a set of sampling points in the ecological sensitive area, the i-th sampling point corresponding to the i-th training sample, is the concentration threshold value of the ecological sensitive area.

[0037] Step S3, obtain the current dredging area remote sensing image, process the image to obtain the current characteristic band reflectivity, and obtain the current construction parameters and hydrological parameters. The current characteristic spectral band, construction parameters and hydrological parameters are input into the trained inversion model to predict the suspended sediment concentration;

[0038] After the inversion model predicts the suspended sediment concentration, the suspended sediment concentration is rendered into an intuitive spatial distribution map through a visualization module, such as Figure 2The specific working process of the visualization module is as follows: saving the trained inversion model and the corresponding wavelength list as an ssc_rf_model.pkl file; reading the hyperspectral remote sensing image (such as the construction area.dat) and its supporting hdr header file, extracting the wavelengths of each band from the hdr header file, generating a water mask and a cloud mask, taking the intersection of the two to obtain an effective water mask, then extracting the blue, yellow and red bands from the hyperspectral remote sensing image to generate a true color RGB image, traversing the ssc_rf_model.pkl file, selecting the wavelengths with a wavelength difference of <1 nm from the wavelengths in the hyperspectral remote sensing image as matching bands, remodeling the matching band data as pixel-level feature input to the inversion model to predict the suspended sediment concentration value, then using a custom color system color scale to superimpose the suspended sediment concentration prediction result on the RGB image in the form of an RGBA layer (only the effective water mask area displays the suspended sediment concentration color), and finally generating a high-resolution suspended sediment concentration distribution map.

[0039] Step S4, setting different levels of dynamic early warning thresholds, the early warning thresholds are related to the dredging power, and corresponding level early warning is performed according to the predicted suspended sediment concentration and the early warning threshold, and the early warning is diffused to the entire ecological sensitive area, and the ecological sensitive area radius is related to the hydrological parameters and the dredging time.

[0040] The schematic diagram of the ecological sensitive area is shown in Figure 3 , and the ecological sensitive area radius is represented as , through the formula, the early warning area can be delineated in advance to ensure that the sediment caused by the construction does not spread to the ecological protection area, wherein T is the dredging time, and g is the acceleration of gravity. When the concentration exceeds the threshold, an alarm is immediately sent to the management personnel.

[0041] In this embodiment, the blue early warning threshold , the yellow early warning threshold and the red early warning threshold are set to be raised in turn, wherein P is the dredging power. When the predicted suspended sediment concentration of the inversion module is less than the blue early warning threshold, it indicates safety and no early warning information is sent. If the suspended sediment concentration is greater than the blue early warning threshold and less than the yellow threshold, a blue early warning information is sent. The early warning thresholds of each level are dynamically adjusted with the dredging power, which can make the early warning information more reliable.

[0042] The system for realizing the remote sensing inversion early warning method of the suspended sediment concentration of the dredging area comprises a data acquisition module, a prediction module and an alarm module; the acquisition module is used for acquiring the hyperspectral remote sensing image of the dredging area by using a UAV, obtaining the current characteristic band reflectivity by processing the image, and acquiring the current construction parameters and hydrological parameters; the characteristic band reflectivity, the construction parameters and the hydrological parameters are jointly input into the prediction module; the prediction module is a trained inversion model; the training data used for training the inversion model comprises the characteristic band reflectivity sensitive to the suspended sediment concentration obtained according to the historical image of the dredging area, the construction parameters and the hydrological parameters of the dredging area on site synchronous with the historical image; the inversion model comprises an attention mechanism layer and an inversion layer; the attention mechanism layer performs weighted fusion on the characteristic band reflectivity, the construction parameters and the hydrological parameters to obtain a weighted feature vector; the inversion layer obtains the suspended sediment concentration value according to the weighted feature vector; the loss function used during training is L total =L MSE +λL eco , wherein L MSE is the mean square error of the predicted concentration and the measured concentration, L eco is an ecological penalty term related to the predicted concentration, and lambda is a penalty coefficient; the alarm module sets different levels of early warning thresholds, the early warning thresholds are related to the dredging ship power, the corresponding level of early warning is performed according to the predicted suspended sediment concentration of the prediction module and the early warning threshold, and the early warning is diffused to the entire ecological sensitive area; the radius of the ecological sensitive area is related to the hydrological parameters and the dredging time.

[0043] In the present application, the terms "first", "second", "third" and the like are only used to distinguish similar objects, and do not have to be used to describe a specific order or sequence, and cannot be understood as indicating or implying relative importance. In the description, the directions or positions indicated by "up", "down", "left", "right", "front" and "back" are based on the directions or positions shown in the drawings, and are only for the convenience of describing the present application, and do not indicate or imply that the devices must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the scope of protection of the present application. For ordinary skilled persons in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0044] In addition, in the description of the present application, "a plurality of" means two or more, unless otherwise specified. The association relationship between the associated objects is described as "and / or", which means that there can be three relationships, for example, A and / or B can mean that A exists alone, A and B exist together, and B exists alone. The character " / " generally represents an "or" relationship between the associated objects before and after it.

[0045] The above merely illustrates the specific embodiments of the present application, but the design concept of the present application is not limited thereto, and any non-essential modification of the present application by using the concept shall be deemed as the infringement of the protection scope of the present application.

Claims

1. A remote sensing inversion early warning method for suspended sediment concentration in dredged areas, characterized by: Includes the following steps: Step S1: Obtain the training sample set. The training samples in the training sample set include the characteristic band reflectance sensitive to suspended sediment concentration obtained from historical images of the dredging area, as well as the construction parameters and hydrological parameters of the dredging area synchronized with the historical images. Step S2: Construct and train the inversion model. The inversion model includes an attention mechanism layer and an inversion layer. The attention mechanism layer performs weighted fusion of characteristic band reflectance, construction parameters, and hydrological parameters to obtain a weighted feature vector. The inversion layer obtains the suspended sediment concentration value based on the weighted feature vector. The inversion model is trained using a training sample set, and the loss function used during training is L. total =L MSE +λL eco , where L MSE To determine the mean square error between the predicted and measured concentrations, L eco The ecological penalty term is related to the predicted concentration, and λ is the penalty coefficient; Step S3: Obtain the current remote sensing image of the dredging area, process the image to obtain the current characteristic band reflectance, and obtain the current construction parameters and hydrological parameters. Input the current characteristic spectral band, construction parameters and hydrological parameters into the trained inversion model to predict the suspended sediment concentration. Step S4: Set dynamic early warning thresholds for different levels. The early warning thresholds are related to the dredging power. Based on the predicted suspended sediment concentration and the early warning threshold, the corresponding level of early warning is issued and the early warning is disseminated to the entire ecologically sensitive area. The radius of the ecologically sensitive area is related to the hydrological parameters and the dredging duration. In step S2, the weighted feature vector obtained by the attention mechanism layer in the inversion model is represented as follows: ,in, The attention weight is P, which represents dredging power, determined by the power of the dredging vessels and the number of dredging vessels. For dredging depth, For the roundness of silt particles, The reflectance sequence is a characteristic band. For water flow velocity, For water wave height, For the Sigmoid function, and These are learnable parameters; In step S2, the inversion layer is based on the formula Obtain suspended sediment concentration values Where W2 and b2 are learnable parameters, and C max This represents the upper limit of suspended sediment concentration.

2. The remote sensing inversion early warning method for suspended sediment concentration in dredged areas according to claim 1, characterized in that: In steps S1 and S3, the image of the dredged area is a hyperspectral remote sensing image acquired by remote sensing. Radiometric calibration and atmospheric correction are performed on the image to obtain surface reflectance data. Based on Pearson correlation analysis, characteristic band reflectance sensitive to suspended sediment concentration is extracted from the surface reflectance data.

3. The remote sensing inversion early warning method for suspended sediment concentration in dredged areas according to claim 2, characterized in that: The construction parameters include dredging vessel power, number of dredging vessels, dredging depth, and sphericity of sediment particles. The hydrological parameters include water flow velocity and wave height.

4. The remote sensing inversion early warning method for suspended sediment concentration in dredged areas according to any one of claims 1 to 3, characterized in that: In step S2, the loss function includes the mean square error L. MSE Represented as Ecological penalty item L eco Represented as Where n represents the number of training samples used in the mean squared error calculation. Let i be the predicted concentration corresponding to the i-th training sample. Let S represent the measured concentration corresponding to the i-th training sample, and let S represent the set of sampling points in the ecologically sensitive area, where the i-th sampling point corresponds to the i-th training sample. This refers to the concentration threshold for ecologically sensitive areas.

5. The remote sensing inversion early warning method for suspended sediment concentration in dredged areas according to any one of claims 1 to 3, characterized in that: In step S4, a blue warning threshold with progressively increasing levels is set. Yellow warning threshold and red alert threshold Where P is the dredging power.

6. The remote sensing inversion early warning method for suspended sediment concentration in dredged areas according to any one of claims 1 to 3, characterized in that: In step S4, the radius of the ecologically sensitive area is expressed as: Where T is the dredging time and g is the gravitational acceleration.

7. A remote sensing inversion early warning system for suspended sediment concentration in dredged areas, used to implement the remote sensing inversion early warning method for suspended sediment concentration in dredged areas as described in any one of claims 1 to 6, characterized in that: The system includes a data acquisition module, a prediction module, and an alarm module. The acquisition module uses a drone to acquire hyperspectral remote sensing images of the dredged area, processes these images to obtain the current characteristic band reflectance, and acquires the current construction parameters and hydrological parameters. These parameters are then input into the prediction module. The prediction module is a pre-trained inversion model. The training data for the inversion model includes characteristic band reflectance sensitive to suspended sediment concentration obtained from historical images of the dredged area, and construction parameters and hydrological parameters of the dredged area synchronized with the historical images. The inversion model includes an attention mechanism layer and an inversion layer. The attention mechanism layer performs weighted fusion of the characteristic band reflectance, construction parameters, and hydrological parameters to obtain a weighted feature vector. The inversion layer obtains the suspended sediment concentration value based on the weighted feature vector. The loss function used during training is L. total =L MSE +λL eco , where L MSE To determine the mean square error between the predicted and measured concentrations, L eco The ecological penalty term is related to the predicted concentration, and λ is the penalty coefficient. The alarm module sets different levels of warning thresholds, which are related to the power of the dredging vessel. Based on the suspended sediment concentration predicted by the prediction module and the warning threshold, the corresponding level of warning is issued and the warning is spread to the entire ecologically sensitive area. The radius of the ecologically sensitive area is related to the hydrological parameters and the dredging time.

Images