In-situ test method for bank soil resistance to impact based on adversarial neural network algorithm

Abstract

The application relates to the technical field of machine learning, in particular to a riverbank soil body in-situ resistance to scour test method based on an adversarial neural network algorithm. The method comprises the following steps: determining a to-be-tested soil body region, collecting initial three-dimensional terrain image data of the region to obtain pre-scouring terrain data, performing scouring through an in-situ scouring strategy, collecting final three-dimensional terrain image data of the region after the scouring is completed to obtain post-scouring terrain data, and collecting bed elevation time series data of a preset fixed point; inputting the pre-scouring terrain data, the post-scouring terrain data and the bed elevation time series data of the preset fixed point into an in-situ soil body resistance to scour prediction model to obtain a resistance to scour performance index of the region; and the in-situ soil body resistance to scour prediction model is obtained by training an adversarial neural network through a comprehensive training data set. The method solves the problem that the existing resistance to scour test lacks multi-source data fusion and process dynamic prediction capability of an erosion process, and greatly improves the efficiency and precision of riverbank resistance to scour evaluation.

Description

Technical Field

[0001] This application relates to the field of machine learning technology, and in particular to an in-situ erosion resistance testing method for riverbank soil based on an adversarial neural network algorithm. Background Technology

[0002] The erosion resistance of riverbank soil is a core parameter for assessing river stability and predicting the risk of bank erosion, and it is of great significance for water conservancy projects, ecological bank protection design, and flood control and disaster reduction.

[0003] In related technologies, soil erosion resistance is mainly assessed through indoor erosion tests or simple field testing methods, which can, to some extent, obtain soil erosion parameters such as critical shear stress and soil erosion rate coefficient. In recent years, the emergence of high-precision measurement methods such as 3D terrain scanning and ultrasonic bed elevation monitoring has laid the foundation for refined observation of the erosion process. Meanwhile, deep learning technology, especially GANs (Generative Adversarial Networks), has demonstrated remarkable nonlinear modeling capabilities in image processing and temporal prediction.

[0004] However, there is currently no mature solution that organically integrates three-dimensional topographic evolution data, dynamic bed elevation sequences, and deep learning algorithms for the automated assessment of in-situ riverbank soil erosion resistance. Existing testing methods are mostly based on simplified ideal experimental conditions, such as constant flow velocity and homogeneous soil, which are difficult to accurately reflect the complex spatial heterogeneity, natural soil structure, and erosion characteristics under non-constant hydrodynamic conditions of actual in-situ riverbank soil. Moreover, traditional testing methods often require a large amount of manpower, time, and equipment, resulting in low testing efficiency and limited data acquisition frequency and accuracy, making it impossible to capture the detailed process of bed evolution in real time and efficiently. In addition, existing erosion resistance test results usually rely on fitting a single physical parameter, lacking the ability to fuse multi-source data and dynamically predict the erosion process, making it difficult to meet the need for accurate simulation of erosion evolution under complex natural conditions. Therefore, there is an urgent need to develop a new method that can integrate multi-source spatial and temporal information, fully explore the laws of erosion evolution, and achieve rapid and accurate prediction of in-situ soil erosion resistance. Summary of the Invention

[0005] This application provides an in-situ erosion resistance testing method for riverbank soil based on an adversarial neural network algorithm, which solves the problem that existing erosion resistance testing lacks the ability to fuse multi-source data and dynamically predict the erosion process, thus greatly improving the efficiency and accuracy of riverbank erosion resistance evaluation.

[0006] To achieve the above objectives, the first aspect of this application proposes a method for in-situ erosion resistance testing of riverbank soil based on an adversarial neural network algorithm, comprising the following steps:

[0007] Determine the area of ​​soil to be tested;

[0008] Initial three-dimensional topographic image data of the soil area to be tested is collected to obtain the topographic data before scour. Based on the preset in-situ scour strategy, scour operation is carried out in the soil area to be tested. After the scour is completed, the final three-dimensional topographic image data of the soil area to be tested is collected to obtain the topographic data after scour. Time series data of bed elevation at preset fixed points are also collected.

[0009] The pre-scour topographic data, the post-scour topographic data, and the time series data of bed elevation at the preset fixed point are input into a pre-trained in-situ soil scour resistance prediction model to obtain the scour resistance performance index of the soil area to be tested. The pre-trained in-situ soil scour resistance prediction model is obtained by training an adversarial neural network using a comprehensive training dataset.

[0010] According to one embodiment of this application, before inputting the pre-scour topographic data, the post-scour topographic data, and the time series data of the bed elevation at the preset fixed point into the pre-trained in-situ soil scour resistance prediction model, the method further includes:

[0011] Construct the comprehensive training dataset,

[0012] Based on a preset partitioning ratio, the comprehensive training dataset is divided into a training set, a validation set, and a test set.

[0013] Construct the adversarial neural network by inputting the training set into the adversarial neural network for training to obtain the initial model parameters;

[0014] Based on the initial model parameters, the validation set is input into the adversarial neural network for performance evaluation, and the initial model parameters are adjusted according to the performance evaluation results until the joint loss function of the validation set converges to obtain the optimal model parameters.

[0015] Based on the optimal model parameters, the test set is input into the adversarial neural network for model testing, and when the test results meet the preset requirements, the pre-trained in-situ soil erosion resistance prediction model is obtained.

[0016] According to one embodiment of this application, constructing the comprehensive training dataset includes:

[0017] Construct a water flow shear stress-velocity model;

[0018] Acquire the initial topographic data and the final topographic data after scouring of the target soil sample, and simultaneously collect the time series data of the bed elevation of the target fixed point of the target soil sample;

[0019] Based on the water flow shear stress-velocity model, the erosion resistance index of the target soil sample is obtained according to the initial topographic data, the final topographic data, and the time series data of the bed elevation of the target fixed point.

[0020] The comprehensive training dataset is obtained based on the initial terrain data, the final terrain data, the time series data of the bed elevation of the target fixed point, and the erosion resistance index of the target soil sample.

[0021] According to one embodiment of this application, the construction of the water flow shear stress-velocity model includes:

[0022] Obtain water flow shear stress measurements at multiple different flow velocities;

[0023] Based on the measured values ​​of water flow shear stress at multiple different flow velocities, a water flow shear stress-flow velocity relationship curve was plotted.

[0024] Based on the water flow shear stress-velocity relationship curve, the water flow shear stress-velocity model is established.

[0025] According to one embodiment of this application, the step of acquiring initial topographic data and final topographic data after scour of the target soil sample, and simultaneously collecting time series data of bed elevation of the target fixed point of the target soil sample, includes:

[0026] The initial topographic data is obtained by using a 3D topographic scanner to perform an initial scan of the target soil sample;

[0027] Determine the initial flow velocity, and based on a preset duration, flush the target soil sample according to the initial flow velocity to obtain intermediate topographic data;

[0028] The initial flow velocity is adjusted based on a preset adjustment strategy, and the step of scouring the target soil sample according to the initial flow velocity for a preset duration to obtain intermediate terrain data is re-executed based on the adjusted flow velocity, thereby obtaining soil scouring response data under multiple flow velocity conditions.

[0029] The final terrain data is obtained by scanning the target soil sample after flushing it with the three-dimensional terrain scanner, and the time series data of the bed elevation of the target fixed point of the target soil sample are collected.

[0030] According to one embodiment of this application, the step of obtaining the erosion resistance index of the target soil sample based on the water flow shear stress-velocity model, according to the initial topographic data, the final topographic data, and the time series data of the bed elevation of the target fixed point, includes:

[0031] Based on the intermediate topographic data corresponding to each flow velocity, determine the change in soil volume corresponding to each flow velocity.

[0032] Based on the aforementioned water flow shear stress-velocity model, the water flow shear stress corresponding to each flow velocity is determined;

[0033] Based on the change in soil volume corresponding to each flow velocity, the soil erosion rate corresponding to each flow velocity is obtained according to the intermediate topographic data corresponding to each flow velocity and the water flow shear stress corresponding to each flow velocity.

[0034] Establish a mapping relationship between the soil erosion rate corresponding to each flow velocity and the water flow shear stress corresponding to each flow velocity, and calculate the erosion resistance index of the target soil sample based on the mapping relationship.

[0035] According to one embodiment of this application, obtaining the comprehensive training dataset based on the initial terrain data, the final terrain data, the time series data of the bed elevation of the target fixed point, and the erosion resistance index of the target soil sample includes:

[0036] The first training dataset is obtained based on the initial terrain data, the final terrain data, the time series data of the bed elevation of the target fixed point, and the erosion resistance index of the target soil sample.

[0037] Adjust the initial terrain conditions of the target soil sample, and based on the adjusted target soil sample, re-execute the steps of obtaining the initial terrain data and the final terrain data after scouring of the target soil sample, and simultaneously collect the time series data of the bed elevation of the target fixed point of the target soil sample to obtain multiple training datasets.

[0038] The comprehensive training dataset is obtained based on the first training dataset and the plurality of training datasets.

[0039] The in-situ erosion resistance testing method for riverbank soil based on adversarial neural network algorithm proposed in this application involves collecting three-dimensional topographic data of the area before and after erosion using three-dimensional topographic scanning technology and collecting time series data of bed elevation using fixed-point elevation monitoring equipment. These data are then input into a comprehensive training dataset to train an adversarial neural network to obtain a pre-trained in-situ soil erosion resistance prediction model, thereby obtaining the erosion resistance performance index of the soil area to be tested. This method solves the problem that existing erosion resistance testing lacks the ability to fuse multi-source data and dynamically predict the erosion process, greatly improving the efficiency and accuracy of riverbank erosion resistance evaluation.

[0040] To achieve the above objectives, a second aspect of this application provides an in-situ erosion resistance testing device for riverbank soil based on an adversarial neural network algorithm, comprising:

[0041] The determination module is used to determine the area of ​​soil to be tested;

[0042] The acquisition module is used to acquire initial three-dimensional topographic image data of the soil area to be measured to obtain topographic data before scour, and to perform scour operation on the soil area to be measured based on a preset in-situ scour strategy. After the scour is completed, the module acquires final three-dimensional topographic image data of the soil area to be measured to obtain topographic data after scour, and also acquires time series data of bed elevation at preset fixed points.

[0043] The testing module is used to input the pre-scour topographic data, the post-scour topographic data, and the time series data of the bed elevation of the preset fixed point into a pre-trained in-situ soil scour resistance prediction model to obtain the scour resistance performance index of the soil area to be tested. The pre-trained in-situ soil scour resistance prediction model is obtained by training an adversarial neural network with a comprehensive training dataset.

[0044] According to one embodiment of this application, before inputting the pre-scour topographic data, the post-scour topographic data, and the time series data of bed elevation at the preset fixed point into the pre-trained in-situ soil erosion resistance prediction model, the testing module is further configured to:

[0045] Construct the comprehensive training dataset,

[0046] Based on a preset partitioning ratio, the comprehensive training dataset is divided into a training set, a validation set, and a test set.

[0047] Construct the adversarial neural network by inputting the training set into the adversarial neural network for training to obtain the initial model parameters;

[0048] Based on the initial model parameters, the validation set is input into the adversarial neural network for performance evaluation, and the initial model parameters are adjusted according to the performance evaluation results until the joint loss function of the validation set converges to obtain the optimal model parameters.

[0049] Based on the optimal model parameters, the test set is input into the adversarial neural network for model testing, and when the test results meet the preset requirements, the pre-trained in-situ soil erosion resistance prediction model is obtained.

[0050] According to one embodiment of this application, the testing module is further configured to:

[0051] Construct a water flow shear stress-velocity model;

[0052] Acquire the initial topographic data and the final topographic data after scouring of the target soil sample, and simultaneously collect the time series data of the bed elevation of the target fixed point of the target soil sample;

[0053] Based on the water flow shear stress-velocity model, the erosion resistance index of the target soil sample is obtained according to the initial topographic data, the final topographic data, and the time series data of the bed elevation of the target fixed point.

[0054] The comprehensive training dataset is obtained based on the initial terrain data, the final terrain data, the time series data of the bed elevation of the target fixed point, and the erosion resistance index of the target soil sample.

[0055] According to one embodiment of this application, the testing module is further configured to:

[0056] Obtain water flow shear stress measurements at multiple different flow velocities;

[0057] Based on the measured values ​​of water flow shear stress at multiple different flow velocities, a water flow shear stress-flow velocity relationship curve was plotted.

[0058] Based on the water flow shear stress-velocity relationship curve, the water flow shear stress-velocity model is established.

[0059] According to one embodiment of this application, the testing module is further configured to:

[0060] The initial topographic data is obtained by using a 3D topographic scanner to perform an initial scan of the target soil sample;

[0061] Determine the initial flow velocity, and based on a preset duration, flush the target soil sample according to the initial flow velocity to obtain intermediate topographic data;

[0062] The initial flow velocity is adjusted based on a preset adjustment strategy, and the step of scouring the target soil sample according to the initial flow velocity for a preset duration to obtain intermediate terrain data is re-executed based on the adjusted flow velocity, thereby obtaining soil scouring response data under multiple flow velocity conditions.

[0063] The final terrain data is obtained by scanning the target soil sample after flushing it with the three-dimensional terrain scanner, and the time series data of the bed elevation of the target fixed point of the target soil sample are collected.

[0064] According to one embodiment of this application, the testing module is further configured to:

[0065] Based on the intermediate topographic data corresponding to each flow velocity, determine the change in soil volume corresponding to each flow velocity.

[0066] Based on the aforementioned water flow shear stress-velocity model, the water flow shear stress corresponding to each flow velocity is determined;

[0067] Based on the change in soil volume corresponding to each flow velocity, the soil erosion rate corresponding to each flow velocity is obtained according to the intermediate topographic data corresponding to each flow velocity and the water flow shear stress corresponding to each flow velocity.

[0068] Establish a mapping relationship between the soil erosion rate corresponding to each flow velocity and the water flow shear stress corresponding to each flow velocity, and calculate the erosion resistance index of the target soil sample based on the mapping relationship.

[0069] According to one embodiment of this application, the testing module is further configured to:

[0070] The first training dataset is obtained based on the initial terrain data, the final terrain data, the time series data of the bed elevation of the target fixed point, and the erosion resistance index of the target soil sample.

[0071] Adjust the initial terrain conditions of the target soil sample, and based on the adjusted target soil sample, re-execute the steps of obtaining the initial terrain data and the final terrain data after scouring of the target soil sample, and simultaneously collect the time series data of the bed elevation of the target fixed point of the target soil sample to obtain multiple training datasets.

[0072] The comprehensive training dataset is obtained based on the first training dataset and the plurality of training datasets.

[0073] The in-situ erosion resistance testing device for riverbank soil based on the adversarial neural network algorithm proposed in this application uses three-dimensional terrain scanning technology to collect three-dimensional terrain data of the area before and after erosion and fixed-point elevation monitoring equipment to collect time series data of bed elevation. These data are then input into a comprehensive training dataset to train the adversarial neural network to obtain a pre-trained in-situ soil erosion resistance prediction model, thereby obtaining the erosion resistance performance index of the soil area to be tested. This method solves the problem that existing erosion resistance testing lacks the ability to fuse multi-source data and dynamically predict the erosion process, greatly improving the efficiency and accuracy of riverbank erosion resistance evaluation.

[0074] To achieve the above objectives, a third aspect of this application provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the in-situ erosion resistance testing method for riverbank soil based on an adversarial neural network algorithm as described in the above embodiments.

[0075] To achieve the above objectives, a fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement the in-situ erosion resistance testing method for riverbank soil based on an adversarial neural network algorithm as described in the above embodiments.

[0076] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0077] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0078] Figure 1 This is a flowchart of an in-situ erosion resistance testing method for riverbank soil based on an adversarial neural network algorithm, according to an embodiment of this application.

[0079] Figure 2 This is a schematic diagram of a soil in-situ erosion resistance testing device based on a neural network algorithm according to an embodiment of this application;

[0080] Figure 3 This is a schematic diagram of the soil scour channel structure of an in-situ soil erosion resistance testing device based on a neural network algorithm according to an embodiment of this application;

[0081] Figure 4 This is a schematic diagram showing the detailed structure of an in-situ soil erosion resistance testing device based on a neural network algorithm, according to one embodiment of this application.

[0082] Figure 5 This is a schematic diagram of the control center of a soil in-situ erosion resistance testing device based on a neural network algorithm, according to one embodiment of this application;

[0083] Figure 6 This is a schematic diagram of an adversarial neural network algorithm structure according to an embodiment of this application;

[0084] Figure 7 This is a schematic diagram of a soil sample provided according to an embodiment of this application;

[0085] Figure 8 This is a schematic diagram of a soil erosion process according to an embodiment of this application;

[0086] Figure 9 This is a schematic diagram of the relationship between water flow shear stress and soil erosion rate according to an embodiment of this application;

[0087] Figure 10 This is a flowchart of an in-situ soil erosion resistance testing method based on an adversarial neural network algorithm according to an embodiment of this application;

[0088] Figure 11 This is a block diagram of an in-situ erosion resistance testing device for riverbank soil based on an adversarial neural network algorithm, provided according to an embodiment of this application.

[0089] Figure 12 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0090] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0091] The following description, with reference to the accompanying drawings, describes an in-situ erosion resistance testing method for riverbank soil based on an adversarial neural network algorithm, according to an embodiment of this application. Addressing the issues mentioned in the background art regarding existing erosion resistance testing methods that rely on fitting a single physical parameter, lack the ability to fuse multi-source data and dynamically predict the erosion process, and are insufficient to meet the need for accurate simulation of erosion evolution under complex natural conditions, this application proposes an in-situ erosion resistance testing method for riverbank soil based on an adversarial neural network algorithm. This method involves acquiring three-dimensional topographic data of the area before and after erosion using three-dimensional topographic scanning technology and collecting time-series data of bed elevation using fixed-point elevation monitoring equipment. These data are input into a comprehensive training dataset to train a pre-trained in-situ soil erosion resistance prediction model obtained from the adversarial neural network, thus obtaining the erosion resistance performance index of the soil area to be tested. This method solves the problems of existing erosion resistance testing methods that rely on fitting a single physical parameter, lack the ability to fuse multi-source data and dynamically predict the erosion process, and are insufficient to meet the need for accurate simulation of erosion evolution under complex natural conditions. It significantly improves the efficiency and accuracy of riverbank erosion resistance evaluation and reduces measurement costs.

[0092] First, with reference to the accompanying drawings, a method for testing the in-situ erosion resistance of riverbank soil based on an adversarial neural network algorithm, according to an embodiment of this application, will be described.

[0093] Figure 1 This is a flowchart of an embodiment of the present application of a method for in-situ erosion resistance testing of riverbank soil based on an adversarial neural network algorithm.

[0094] like Figure 1 As shown, the in-situ erosion resistance testing method for riverbank soil based on the adversarial neural network algorithm includes the following steps:

[0095] In step S101, the area of ​​soil to be tested is determined.

[0096] The soil area to be tested refers to the specific soil area for which the erosion resistance needs to be evaluated, and can be defined according to actual needs.

[0097] Specifically, the area to be measured can be any soil area that may be affected by water erosion, such as riverbank slopes, dam foundations, highway slopes, and farmland soil. When determining the area, representativeness, clear boundaries, and on-site accessibility should be considered. In this embodiment, the land area to be measured is a riverbank in the wild.

[0098] In step S102, initial three-dimensional topographic image data of the soil area to be tested is collected to obtain the topographic data before scour. Based on the preset in-situ scour strategy, scour operation is carried out in the soil area to be tested. After the scour is completed, the final three-dimensional topographic image data of the soil area to be tested is collected to obtain the topographic data after scour. Time series data of bed elevation at preset fixed points are also collected.

[0099] The initial 3D topographic image data refers to the topographic data of the area to be measured, collected before the scouring operation using technologies such as 3D scanning, UAV mapping, and LiDAR, including spatial information such as the elevation, slope, and undulation of the soil surface; the preset in-situ scouring strategy can be a user-preset in-situ scouring strategy, an in-situ scouring strategy obtained through a limited number of experiments, or an in-situ scouring strategy obtained through a limited number of computer simulations, without specific limitations; the final 3D topographic image data is the 3D topographic data collected again for the same area to be measured after the scouring operation; the preset fixed point can be a user-preset fixed point, a fixed point obtained through a limited number of experiments, or a fixed point obtained through a limited number of computer simulations, without specific limitations; the bed elevation refers to the altitude or relative height of the soil surface at the fixed point; the time series data is the continuous recording of the bed elevation changes at time intervals, reflecting the real-time erosion or deposition at that point during the scouring process.

[0100] Specifically, the acquisition of pre- and post-scour topographic data employs 3D topographic scanning technology to perform a comprehensive scan of the determined soil area to be measured. Before scour, the complete 3D morphological data of the soil before scour is obtained through scanning, generating an initial 3D topographic image, i.e., pre-scour topographic data, which serves as a benchmark for subsequent erosion changes. After scour, the same area is scanned again to capture the final morphology of the soil after scour, obtaining post-scour topographic data. The two scans must ensure consistent scanning range and accuracy to allow for comparative analysis of soil erosion and topographic change amplitude. The acquisition of time-series data on bed elevation at pre-selected fixed points within the soil area to be measured involves installing elevation monitoring equipment at each point. Throughout the scour operation based on the pre-set in-situ scour strategy, these devices continuously record the real-time elevation data of each fixed point at a set frequency, forming a continuous data sequence that changes over time, i.e., bed elevation time-series data, thereby reflecting the dynamic changes in soil elevation during the scour process.

[0101] In step S103, the topographic data before scour, the topographic data after scour, and the time series data of bed elevation at preset fixed points are input into the pre-trained in-situ soil scour resistance prediction model to obtain the scour resistance performance index of the soil area to be tested. The pre-trained in-situ soil scour resistance prediction model is obtained by training an adversarial neural network with a comprehensive training dataset.

[0102] Among them, the pre-scour topographic data refers to the initial three-dimensional image data obtained by scanning the soil topography of the test area using equipment such as a three-dimensional topographic scanner before the soil scour resistance experiment begins; the post-scour topographic data refers to the final three-dimensional topographic image data obtained by scanning the soil topography of the test area again after the scour experiment is completed; the in-situ soil scour resistance prediction model is a computational model trained with data, which can predict the performance indicators of in-situ soil in the field against water scour based on the input topographic data and elevation time series data; the scour resistance performance index of the soil area to be tested is a quantitative parameter used to measure the ability of the soil to resist water scour, which is the result of the model output and can reflect the stability of the soil under the action of water flow; the comprehensive training dataset is a data set used to train the model, which contains a variety of data related to soil scour resistance, providing rich samples for model learning; the adversarial neural network is a deep learning model composed of a generative network and a discriminative network.

[0103] Specifically, the topographic baseline data before scour, the topographic change data after scour, and the dynamic change data of the bed elevation at fixed points over time are input into a pre-trained in-situ soil scour resistance prediction model. This model then outputs the final scour resistance performance index of the soil area to be tested. It should be noted that the pre-trained in-situ soil scour resistance prediction model is built upon an adversarial neural network, and this model is trained using a comprehensive training dataset. The specific training process of the in-situ soil scour resistance prediction model will be described in detail later.

[0104] Therefore, through the complete chain of data input, model calculation, and result output, the measured topographic and time series data are transformed into quantitative indicators of soil erosion resistance by using a pre-trained adversarial neural network model.

[0105] The following section will elaborate on the specific training process of the in-situ soil erosion resistance prediction model.

[0106] As one possible implementation, in some embodiments, before inputting the pre-scour topographic data, post-scour topographic data, and time series data of bed elevation at preset fixed points into the pre-trained in-situ soil scour resistance prediction model, the method further includes: constructing a comprehensive training dataset, dividing the comprehensive training dataset into a training set, a validation set, and a test set based on a preset partitioning ratio; constructing an adversarial neural network, inputting the training set into the adversarial neural network for training to obtain initial model parameters; based on the initial model parameters, inputting the validation set into the adversarial neural network for performance evaluation, and adjusting the initial model parameters according to the performance evaluation results until the joint loss function of the validation set converges to obtain the optimal model parameters; based on the optimal model parameters, inputting the test set into the adversarial neural network for model testing, and obtaining the pre-trained in-situ soil scour resistance prediction model when the test results meet preset requirements.

[0107] The training dataset consists of several parts: a comprehensive training dataset, a set of original data for training the model, and a validation dataset. The training set is used to evaluate the model's generalization ability during training, preventing it from excessively memorizing training data and adjusting parameters based on the evaluation results. The test set is used to ultimately verify the model's true performance, determining whether it can stably and accurately predict results on new data. Initial model parameters are the initial settings for the adversarial neural network at the start of training; these parameters are continuously optimized and adjusted during training. Performance evaluation uses validation data to verify the model's prediction performance under the current parameters. The joint loss function measures the difference between the model's predictions and the actual results. Optimal model parameters are the parameter combination that achieves the best model performance after repeated adjustments on the validation set; this is the final core configuration of the model. The pre-trained in-situ soil erosion resistance prediction model is a model trained using the above process that can stably output soil erosion resistance performance indicators.

[0108] Before training and optimizing the in-situ soil erosion resistance prediction model, this application provides an in-situ soil erosion resistance testing device, such as... Figure 2 As shown, the in-situ soil erosion resistance testing device 20 consists of a centrifugal pump 201, a control valve 202, an electromagnetic flowmeter 203, a soil erosion channel 30, an ultrasonic acoustic sensor, a three-dimensional terrain scanner, a frequency converter, a power supply module, and a central processing unit. The soil erosion channel structure is as follows: Figure 3 As shown, the soil scour channel 30 includes: a transition pipe 301, a top sensor opening 302, and a sample tray 303. The detailed structure of the in-situ soil scour resistance testing device based on a neural network algorithm is shown below. Figure 4As shown, it includes: inlet section 401, middle transition section 402, and soil scour channel section 403. The control center of the soil in-situ scour resistance testing device based on neural network algorithm is shown below. Figure 5 As shown, it includes: main switch 501 and frequency converter 502.

[0109] The frequency converter is electrically connected to the centrifugal pump and integrated into the power supply module. It precisely controls the pump speed by adjusting the output frequency, thus achieving stepless adjustment of the water flow rate. The central processing unit is responsible for collecting, storing, and displaying various sensor signals in real time, realizing automated control and data management of the experimental process. The centrifugal pump 201, control valve 202, electromagnetic flowmeter 203, and frequency converter constitute the water flow control unit. The centrifugal pump 201 serves as the power source, drawing water from the river and providing a stable flow. The control valve 202 is installed on the pipeline between the centrifugal pump and the electromagnetic flowmeter for manual flow adjustment, forming dual flow control in conjunction with the frequency converter. The electromagnetic flowmeter 203 monitors the instantaneous and cumulative flow in the pipe and feeds the signal back to the central processing unit for flow control and data recording. The frequency converter is connected to the centrifugal pump and integrated into the power supply module, controlling the water flow rate by adjusting the output power. Real-time and precise flow control is achieved through the integrated water flow control unit. The soil scour channel 30 is made of transparent acrylic sheet, closed on the top and front and back sides, and open on the left, right and bottom sides. Symmetrical grooves are opened on the inner walls of the front and back side panels. The sample tray is fixed by the grooves and sealing strips, ensuring convenient loading and unloading and good water tightness. The left side is connected to the outlet of the electromagnetic flowmeter via a transition pipe, and the right side is connected to the river via an inlet pipe. The top of the soil scour channel 30 has 6 sensor mounting holes in 3 rows and 2 columns, with a row spacing of 15cm and a column spacing of 8cm. The hole diameter matches the diameter of the ultrasonic acoustic sensor. Before the test, the sensor is inserted vertically into the hole, with the bottom of the probe 3cm away from the soil surface of the sample tray, for continuous monitoring of the bed elevation at the fixed point and generation of complete time series data.

[0110] Specifically, a comprehensive training dataset is constructed by conducting in-situ soil erosion resistance experiments at the target riverbank; initial topographic data of the in-situ soil, final topographic data after erosion, and time series data of bed elevation at preset fixed points are collected. The specific steps are as follows:

[0111] (1) Select the soil area to be tested on the target riverbank, sprinkle water evenly on the area to pre-wet the soil and simulate the natural saturated or semi-saturated state during the scouring process.

[0112] (2) A portable high-precision three-dimensional terrain scanner was used to perform an initial scan of the area to be measured to obtain the initial three-dimensional image data of the original terrain;

[0113] (3) Set up an in-situ flushing test device in the selected area, control the water flow and flushing direction, and carry out the flushing test in stages with reference to the flow rate and flushing time set in the laboratory test.

[0114] (4) Set up acoustic sensors in the test area and collect time series data of bed surface elevation at preset fixed points;

[0115] (5) After the scouring experiment is completed, a three-dimensional scanner is used to perform a final scan of the test area to obtain the final three-dimensional terrain image data after scouring.

[0116] Among them, deep learning models based on adversarial neural network algorithms include generator design, discriminator design, and loss function design:

[0117] (1) Generator Design: The initial topography, final topography, and bed elevation time series obtained from laboratory soil erosion resistance tests are used as input conditions. A topography encoder is used to extract the spatial features of the initial and final topography, and a temporal encoder is used to extract the dynamic features of the bed elevation changes at fixed points. Subsequently, a feature fusion module is constructed to fuse the spatial and temporal features to generate a unified feature representation. The fused feature vector is mapped to the predicted soil erosion resistance value through a regression decoder, which serves as the output of the generator.

[0118] (2) Discriminator Design: Topographic data, time-series bed elevation data, and real / predicted data are used as input conditions. A joint embedding discriminator is used to extract topographic features, extract time-series bed elevation features, and expand soil erosion resistance features. Then, the three types of features are concatenated, and the consistency between the topographic / time-series features and the erosion resistance features is judged by a matching evaluation module. Finally, a multi-scale discriminant grid is used to output the corresponding authenticity probability.

[0119] (3) Design of the loss function. The generator's loss function includes three parts: Regression loss, combating loss and physical loss constraints Multi-objective optimization is achieved through weighted partitioning. The total loss expression is:

[0120]

[0121] In the formula: It is the total loss of the generator. , , They are , and The loss weight parameters.

[0122] Regression loss: Measures the absolute error between the predicted erosion rate and the actual rate, expressed as follows:

[0123]

[0124] in, It is the predicted erosion rate of the i-th sample. It is the first The true erosion rate of each sample, where N is the total number of samples.

[0125] Physical loss constraints: Based on the fundamental equation of sediment transport (Exner equation), physical constraints are introduced to improve the physical rationality of the model prediction. The erosion rate predicted by the model is substituted into the Exner equation to calculate the residual, and the physical loss constraints are obtained.

[0126]

[0127]

[0128] In the formula: z is the bed surface elevation, It is the sediment transport rate. It is the density of the sediment. It is the porosity of sediment. The predicted bed elevation at time t. It is the predicted erosion rate at time t. It is a time length window.

[0129] Specifically, the structure of the adversarial neural network algorithm is as follows: Figure 6 As shown, Figure 6 This is a schematic diagram of the structure of an adversarial neural network algorithm according to an embodiment of this application.

[0130] The adversarial neural network algorithm structure includes a generator and a discriminator:

[0131] The generator primarily generates predicted erosion rates based on input data, including initial and final topographic features, as well as time-series bed elevation data. These data reflect the spatial morphology and dynamic elevation changes of the soil mass, respectively. Next, a CNN (Convolutional Neural Network, Topographic Encoder) is used to extract spatial features of the initial and final topographic features, such as slope and undulation; an LSTM (Long Short-Term Memory, Temporal Encoder) is used to extract dynamic features of the bed elevation time-series, such as the rate and trend of elevation change. Feature fusion is then performed to obtain a joint feature vector that simultaneously contains both static and dynamic topographic patterns. Finally, a regression decoder outputs the predicted erosion rate.

[0132] The discriminator is primarily used to evaluate the realism of the generator's predictions. The input includes terrain and time-series data, as well as either the actual or predicted velocity. The actual velocity is the measured velocity, and the predicted velocity is the generator's output. First, a multimodal encoder extracts features from the terrain and time-series data, expanding the velocity features of the actual or predicted velocity to enrich its feature dimensions. Then, the terrain and time-series features are concatenated with the velocity features and input into the discriminator network. The output is a probability of realism, ranging from 0 to 1; the closer to 1, the more closely the generator's prediction resembles the real data.

[0133] The generator aims to make the predicted erosion rate closer to the actual rate, while the discriminator strives to more accurately distinguish between the actual rate and the generator's prediction. Through repeated competition and iteration, the generator can accurately predict the erosion rate, and the discriminator can reliably assess the authenticity of the data, thus outputting the final result.

[0134] Thus, by having the generator learn to generate prediction results, the discriminator evaluates the authenticity of the results, and the loss function drives the process of both optimizing together, the adversarial neural network can accurately learn the relationship between terrain, time-series data and soil erosion rate, ultimately achieving accurate prediction of soil erosion resistance.

[0135] The training experiment for the in-situ soil erosion resistance prediction model includes the following steps:

[0136] Step 1: Select a suitable location near the riverbank in the wild to conduct in-situ field tests and clear away excess debris from the test area;

[0137] Step 2: Attach Figure 2 The test apparatus shown is removed from the base and placed on the ground. The area to be measured is selected, the soil sample to be tested is determined, and the initial topography of the soil sample is recorded using a 3D scanner.

[0138] Step 3: Insert the lower end of the erosion channel directly into the soil of the area to be measured, to a depth of 3-5 cm;

[0139] Step 4: Fix the ultrasonic sensor at the designated position on the device, open the control valve, start the centrifugal pump, and pump water from the river to conduct an in-situ soil erosion resistance test.

[0140] Step 5: Adjust the frequency converter and conduct a flushing experiment using the power of a centrifugal pump;

[0141] Step 6: Every 5 minutes, adjust the frequency converter to change the centrifugal pump power, ensuring that the centrifugal pump power and flushing time remain consistent after each frequency converter adjustment with the power and flushing time changed during the indoor experiment. After each flow gradient flushing experiment is completed, do not stop the experiment to perform intermediate terrain measurements, but directly change the centrifugal pump power to start the next flow gradient flushing experiment;

[0142] Step 7: After the last flow gradient scour experiment is completed, close the control valve and centrifugal pump, stop the experiment, dismantle the device, and use a 3D scanner to record the final soil scour topographic image data.

[0143] Step 8: Input the collected initial terrain image data, final terrain image data, and fixed location elevation data as a dataset into the trained calculation model for calculation, and finally obtain the erosion resistance result of the field sample soil.

[0144] Therefore, by clearly defining the soil area to be tested, collecting three-dimensional topographic image data of the area before and after scour, and simultaneously operating according to a preset strategy during the scour process, collecting time-series data of the bed elevation at preset fixed points, the necessary basic data for evaluation is obtained. Finally, these data are input into a pre-trained in-situ soil scour resistance prediction model obtained by training an adversarial neural network using a comprehensive training dataset, thereby obtaining the scour resistance performance index of the soil area to be tested. This achieves an effective evaluation of the soil scour resistance performance.

[0145] Optionally, in some embodiments, constructing a comprehensive training dataset includes: constructing a water flow shear stress-velocity model; acquiring initial topographic data and final topographic data after scouring of the target soil sample, and simultaneously collecting time series data of bed elevation at target fixed points of the target soil sample; obtaining scour resistance indices of the target soil sample based on the water flow shear stress-velocity model, according to the initial topographic data, final topographic data, and time series data of bed elevation at target fixed points; and obtaining a comprehensive training dataset based on the initial topographic data, final topographic data, time series data of bed elevation at target fixed points, and scour resistance indices of the target soil sample.

[0146] The water flow shear stress-velocity model describes the relationship between water flow shear stress and water flow velocity. Water flow shear stress is the driving force of water on the soil surface, and is the core driving force causing erosion. The target soil sample is a specific soil sample used in the experiment, and is the object of data collection and index calculation. Specifically, as shown... Figure 7 As shown, Figure 7 This is a schematic diagram of a soil sample provided according to an embodiment of this application; the scour resistance index is a parameter that quantitatively characterizes the soil's ability to resist scour, and is calculated by analyzing topographic data, dynamic monitoring data, and combining the water flow shear stress-velocity model.

[0147] Specifically, the relationship between water flow shear stress and flow velocity is first established. A scour experiment is then conducted on the target soil sample, simultaneously acquiring: initial topographic data before scour and final topographic data after scour.

[0148] The time series data of bed elevation at fixed points can be obtained using the aforementioned in-situ soil erosion resistance testing device. Erosion resistance indices are calculated, and combined with the established shear stress-flow velocity model, the three types of data are analyzed to derive the erosion resistance indices for the soil sample. The initial topography, final topography, and bed elevation time series are correlated with the calculated erosion resistance indices to form a complete set of samples. A large number of such samples are then collected to constitute the comprehensive training dataset used to train the model.

[0149] Therefore, by combining theoretical models, experimental data, and index calculations, the dataset is ensured to include both topographic and dynamic change information of the soil and corresponding erosion resistance quantification results, providing indicators for subsequent model learning of the relationship between data features and erosion resistance.

[0150] Optionally, in some embodiments, constructing a water flow shear stress-velocity model includes: obtaining water flow shear stress measurements at multiple different flow velocities; fitting and plotting a water flow shear stress-velocity relationship curve based on the water flow shear stress measurements at multiple different flow velocities; and establishing a water flow shear stress-velocity model based on the water flow shear stress-velocity relationship curve.

[0151] Among them, the water flow shear stress measurement value is the specific value of water flow shear stress obtained through experiments under specific flow velocity conditions; the water flow shear stress-flow velocity relationship curve is a curve obtained by plotting multiple sets of flow velocity-shear stress measurements on a coordinate system with flow velocity as the horizontal axis and shear stress as the vertical axis, and then by mathematical fitting, which intuitively reflects the relationship between the two.

[0152] Specifically, the relationship between water flow shear stress and velocity was established through a water tank experiment, applying different water flow velocities. And measure the corresponding water flow shear stress. Perform parameter calibration and establish flow velocity. With shear stress Based on experience, the specific steps are as follows:

[0153] Step 1: Before the experiment begins, stress plates are attached to the bottom of the water tank to measure the shear stress generated by the water flow in real time.

[0154] Step 2: Fill the water tank with water until it completely submerges the shear stress sensing device;

[0155] Step 3: Start the water tank flushing experiment and use a flow meter to measure the water flow velocity. Once the flow velocity stabilizes, record the velocity at that time. and the corresponding water flow shear stress ;

[0156] Step 4: Gradually adjust the water flow velocity and repeat Step 3 to obtain multiple water flow shear stress measurements at different flow velocities.

[0157] Step 5, based on the flow rate collected in the above experiment... Corresponding water flow shear stress Data, fitting and plotting Relationship curves were obtained, empirical parameters were calibrated, and a quantitative relationship model between flow velocity and water flow shear stress was established.

[0158] The determination of the relationship between indoor calibrated water flow velocity and water flow shear stress includes the following steps:

[0159] Step 1: After attaching the stress-retaining sheet to the surface of the sample tray, fix it in place as shown in the image. Figure 2 The test apparatus shown was used to conduct a water flow velocity and water flow shear stress determination experiment below the scouring channel 30.

[0160] Step 2: Open the control valve 202, place the water pipe at the inlet end of the flushing device into the water source, and start the centrifugal pump 201 to fill the entire flushing device with water.

[0161] Step 3: Adjust the output frequency of the frequency converter to change the pump speed, observe the reading of the electromagnetic flowmeter 203, and record the water flow shear stress data corresponding to the current reading after the reading reaches the predetermined value and stabilizes.

[0162] Step 4: Repeat step 3 to record multiple sets of electromagnetic flowmeter data and water flow shear stress data;

[0163] Step 5: Based on the specifications of the electromagnetic flowmeter, convert the flow rate into water velocity and establish the water velocity. With water flow shear stress The relationship between indoor water flow velocity and water flow shear stress was calibrated.

[0164] Therefore, by simultaneously measuring multiple sets of different flow velocities, the comprehensiveness of the data is ensured. Combined with mathematical fitting and parameter calibration, the model has a clear quantitative relationship, providing a reliable basis for calculating the shear stress of water flow through flow velocity. Moreover, the indoor experimental environment is controllable, and the data accuracy and consistency are high.

[0165] Optionally, in some embodiments, acquiring initial topographic data and final topographic data after scouring of the target soil sample, and simultaneously collecting time series data of bed elevation of the target fixed point of the target soil sample, includes: using a 3D topographic scanner to perform an initial scan of the target soil sample to obtain initial topographic data; determining an initial flow velocity, and based on a preset duration, scouring the target soil sample according to the initial flow velocity to obtain intermediate topographic data; adjusting the initial flow velocity based on a preset adjustment strategy, and based on the adjusted flow velocity, re-executing the step of scouring the target soil sample according to the initial flow velocity based on a preset duration to obtain intermediate topographic data, thereby obtaining soil scouring response data under multiple flow velocity conditions; using a 3D topographic scanner to scan the target soil sample after scouring to obtain final topographic data, and collecting time series data of bed elevation of the target fixed point of the target soil sample.

[0166] The target fixed point is a pre-selected, fixed observation point on the surface of the soil sample, used to track the elevation change of that point during scouring and reflect the dynamic characteristics of local scouring. A 3D terrain scanner is a device used to acquire the 3D morphology of an object's surface. It scans soil samples using laser, optical, and other technologies to generate 3D coordinate data containing a large number of points, thereby constructing a terrain model for accurately acquiring initial, intermediate, and final terrain data. The preset duration can be a user-defined duration, a duration obtained through a limited number of experiments, or a duration obtained through a limited number of computer simulations; no specific limitation is made here. Intermediate terrain data is soil terrain data acquired after each flow velocity scouring stage during the phased scouring process. It is used to record the terrain state at the end of different flow velocity stages, reflecting the phased scouring results. The preset adjustment strategy can be a user-defined adjustment strategy, an adjustment strategy obtained through a limited number of experiments, or an adjustment strategy obtained through a limited number of computer simulations; no specific limitation is made here. Soil scouring response data under multiple flow velocity conditions refers to the collection of intermediate terrain data and other information recorded after scouring at different flow velocities.

[0167] Specifically, indoor water tank experiments were conducted to test the erosion resistance of soil samples; initial topographic data before erosion and final topographic data after erosion were obtained using a 3D topographic scanner; and time-series data of bed elevation at preset fixed points on the soil surface were simultaneously collected using an ultrasonic ranging system. The specific steps are as follows:

[0168] Step 1: Before the experiment begins, place the soil sample in the sample tray and manually shape the preset initial terrain.

[0169] Step 2: Use a high-precision 3D terrain scanner to perform an initial scan of the soil sample to obtain complete initial 3D terrain data;

[0170] Step 3: Secure the sample tray to the water tank and fill it with water until the soil sample is completely submerged.

[0171] Step 4: Start the flushing experiment and control the water flow rate to flush. After the preset flushing time (5 minutes), pause the experiment, slowly remove the sample tray, and immediately use a 3D scanner to scan the current terrain to obtain intermediate terrain data. Unlike field experiments, there is no need to stop the experiment to collect intermediate terrain data between each stage of the field experiment. Only the flow rate parameters are adjusted to continuously implement the next stage of flushing.

[0172] Step 5: After scanning is complete, put the sample tray back into the water tank, adjust the flow rate to the next set value, and continue the rinsing experiment. Repeat step 4.

[0173] Step 6: Repeat steps 4 and 5 multiple times to obtain soil scour response data under multiple flow velocity conditions. The scour results are as follows: Figure 8 As shown, Figure 8 The results of soil scour response under different flow velocities are presented.

[0174] Step 7: After all the tests are completed, remove the sample tray and scan the sample again using a 3D scanner to collect the final terrain data;

[0175] Step 8: Throughout the process, real-time time series data of bed elevation at preset fixed points on the soil surface are collected using an acoustic ranging device (such as an acoustic sensor).

[0176] The indoor soil erosion resistance test includes the following steps:

[0177] Step 1: Before the experiment begins, fill the sample tray with soil to create the initial terrain. Then, use a 3D scanner to scan the soil and record the initial terrain of the sample.

[0178] Step 2: Fix the sample tray in the groove at the bottom of the soil flushing channel 30 using the rubber sealing strip, and at the same time use the flange to connect the soil flushing channel to the transition pipes on the left and right sides.

[0179] Step 3: Place the ultrasonic acoustic sensor into the six sensor holes at the top of the soil scour channel 30 and fix it 3cm above the soil surface.

[0180] Step 4: Place the inlet pipe at one end of the soil flushing channel into the water source, open the control valve 202, start the centrifugal pump power supply, and fill the device with water until the entire device is completely filled with water.

[0181] Step 5: Adjust the frequency converter to change the power of the centrifugal pump. First, conduct the experiment with low power and observe the reading of the electromagnetic flowmeter. When the reading stabilizes, start timing. After timing for 5 minutes, turn off the control valve and the power supply of the centrifugal pump to drain the water from the device.

[0182] Step 6: After all the water in the device has been drained, remove the ultrasonic acoustic sensor, remove the acrylic plate at the top of the soil flushing channel, and use a 3D scanner to scan the soil topography after flushing.

[0183] Step 7: After measuring the terrain, fix the top cover of the soil flushing channel back to its original position using rubber seals and screws, and re-fix the ultrasonic sensor back to its original position.

[0184] Step 8: Repeat steps 4-7. Each time the frequency converter is adjusted, ensure that the working power of the centrifugal pump is slightly higher than the power of the previous experiment (power increment ΔP ≥ 10%). Record 7-10 sets of electromagnetic flowmeter readings and topographic data obtained after soil scouring.

[0185] Step 9: After the experiment, remove the sample tray and use a 3D scanner to record the final topography of the soil sample.

[0186] Step 10: Replace the soil sample and repeat steps 1-9 to obtain a sufficient dataset of soil sample erosion resistance performance.

[0187] Therefore, by shaping the soil sample into an initial terrain and scanning and recording it before the experiment, then fixing the sample tray, installing sensors, injecting water and controlling the flow rate by adjusting the power of the centrifugal pump, rinsing for a preset time and then pausing the experiment, scanning the intermediate terrain, and repeating this process multiple times to obtain multiple sets of data, and synchronously collecting the time series of elevations at fixed points throughout the process, the final terrain is obtained and the sample is replaced to accumulate the dataset. This method can precisely adjust the flow rate and obtain the intermediate terrain through multiple interruptions in scanning, resulting in detailed and comprehensive data collection. Through multiple sets of experiments with different flow rates and sample replacements, a rich dataset can be accumulated, providing standardized and high-quality basic data for subsequent model training, and it can be compared with field experiments to facilitate the verification of data correlation.

[0188] Optionally, in some embodiments, based on the water flow shear stress-velocity model, the erosion resistance index of the target soil sample is obtained according to the initial topographic data, the final topographic data, and the time series data of the bed elevation of the target fixed point. This includes: determining the volume change of soil at each flow velocity based on the intermediate topographic data corresponding to each flow velocity; determining the water flow shear stress at each flow velocity based on the water flow shear stress-velocity model; obtaining the soil erosion rate at each flow velocity based on the volume change of soil at each flow velocity, according to the intermediate topographic data and the water flow shear stress at each flow velocity; establishing a mapping relationship between the soil erosion rate at each flow velocity and the water flow shear stress at each flow velocity; and calculating the erosion resistance index of the target soil sample according to the mapping relationship.

[0189] Among them, the erosion resistance index of the target soil sample is a parameter that quantitatively characterizes the soil's ability to resist erosion, and is the final result obtained through data analysis; the change in soil volume is the difference between the volume of the soil sample and the initial volume after erosion at a certain flow rate; the soil erosion rate is the volume or depth of soil eroded per unit time, which is obtained by dividing the change in volume at a certain flow rate by the erosion duration, reflecting how fast the soil is eroded at that flow rate.

[0190] Specifically, the calculation of erosion resistance index and the construction of the dataset are as follows: Based on the initial topographic data, final topographic data, and bed elevation time series data obtained in the above steps, the erosion resistance index of soil samples under each experimental condition is calculated, and a dataset containing the initial topographic data, final topographic data, bed elevation time series data, and their corresponding erosion resistance indexes is constructed. The specific steps are as follows:

[0191] Step 1: Based on the intermediate topographic data collected in the previous steps and the previously collected topographic data, calculate the change in soil volume at the current flow velocity. ;

[0192] Step 2, establish the water flow velocity based on the above steps. With water flow shear stress Based on empirical relationships, determine the water flow shear stress corresponding to the current flow velocity. ;

[0193] Step 3: Based on the intermediate topographic data collected in the two steps above, calculate the soil erosion rate E after adjusting the flow velocity.

[0194] Step 4: Repeat step 3 to obtain multiple sets of different flow rates. and its corresponding soil erosion rate E;

[0195] Step 5: Establish the relationship between soil erosion rate E and water flow shear stress The mapping relationship is used to calculate the erosion resistance index of soil samples by fitting the following erosion resistance model:

[0196]

[0197] In the formula, E is the soil erosion rate. It is the shear stress of the water flow (Pa). It represents the soil shear strength (Pa); k and m are coefficients.

[0198] Specifically, water flow shear force The quantitative relationship between E and soil erosion rate is as follows: Figure 9 As shown, Figure 9 This is a schematic diagram of the relationship between water flow shear stress and erosion rate according to an embodiment of this application.

[0199] Where the horizontal axis represents the shear force of the water flow. The unit is Pa, representing the shear strength of the water flow on the soil surface; the vertical axis is the soil erosion rate E, in meters per second. 3 / s reflects the volume of soil eroded per unit time. Figure 9 The scatter points represent experimentally measured data points of water flow shear force versus erosion rate. The dashed line is the power function fitting curve for these scatter points, and the fitting formula is: This indicates that the soil erosion rate E is related to the water flow shear force. The relationship is power-law, and the soil erosion rate increases at a faster rate as the water flow shear force increases. Meanwhile, the goodness of fit... The value is close to 1, indicating that the power function can very accurately describe the relationship between water flow shear force and soil erosion rate, and the model has strong reliability.

[0200] The training of the in-situ soil erosion resistance calculation model based on the adversarial neural network algorithm includes the following steps:

[0201] Step 1: For each set of sample data, calculate the water flow velocity corresponding to each reading based on the electromagnetic flowmeter reading and the electromagnetic flowmeter specifications;

[0202] Step 2: Analyze the soil topography obtained after each different flow rate scouring of the soil sample and the soil topography obtained at the previous moment, and calculate the average soil erosion rate corresponding to the current flow rate.

[0203] Step 3: Based on the data processed in Step 1 and Step 2, obtain 7-10 sets of data on different water flow velocities and average erosion rates of different soil types;

[0204] Step 4: Based on the relationship between water flow velocity and water flow shear stress obtained from the indoor calibration water flow velocity and water flow shear stress test, and the relationship between different water flow velocities and average erosion rates of different soils obtained in Step 3, calculate the relationship between water flow shear stress and soil erosion rate.

[0205] Step 5: Based on the data of different water flow shear stresses and soil erosion rates obtained in Step 4, plot the relationship between soil erosion rate E and water flow shear stress. The relationship curve, calibration parameters, and calculation formulas are used to determine the erosion resistance of soil samples.

[0206]

[0207] In the formula, E is the soil erosion rate. It is the shear stress of the water flow (Pa). It represents the soil shear strength (Pa); k and m are coefficients.

[0208] Step 6: Process the data recorded by the ultrasonic acoustic sensor to obtain the data on the change of bed elevation at different locations of the soil sample;

[0209] Step 7: The initial topographic image, the final topographic image, the data on the elevation changes of the bed surface at different points, and the erosion resistance value of the soil sample are used as a set of data to process all experimental samples, forming a large number of original datasets.

[0210] Step 8: Divide the original dataset into a training set and a validation set in a 7:3 ratio. Input the training set into the in-situ soil erosion resistance calculation model based on the adversarial neural network algorithm for model training. Use the sample validation set to validate the trained model. Continuously adjust and optimize the model until it meets the accuracy requirements, thus completing the training of the in-situ soil erosion resistance calculation model based on the adversarial neural network algorithm.

[0211] Thus, through the logic of index calculation, data integration, and model training, the transformation from experimental data to predictive models was realized, providing quantitative tools and data support for soil erosion resistance assessment.

[0212] To make the comprehensive training dataset more complete and representative, and to improve the model's adaptability to different scenarios, it is necessary to consider soil erosion resistance data under various initial terrain conditions. Specifically, the comprehensive training dataset is constructed in the following way.

[0213] Optionally, in some embodiments, a comprehensive training dataset is obtained based on initial terrain data, final terrain data, time series data of bed elevation of target fixed points, and erosion resistance index of target soil samples. This includes: obtaining a first training dataset based on initial terrain data, final terrain data, time series data of bed elevation of target fixed points, and erosion resistance index of target soil samples; adjusting the initial terrain conditions of the target soil samples, and re-executing the steps of acquiring the initial terrain data and the final terrain data after erosion of the target soil samples, and synchronously collecting the time series data of bed elevation of target fixed points of the target soil samples, to obtain multiple training datasets; and obtaining a comprehensive training dataset based on the first training dataset and multiple training datasets.

[0214] The multiple training datasets are obtained by repeatedly experimenting with soil samples under different initial conditions by adjusting the initial terrain conditions. Each dataset corresponds to a different initial terrain.

[0215] Specifically, based on the target soil sample with initial topographic conditions, the initial topographic data, final topographic data, and fixed-point bed elevation time series data obtained from the experiment are correlated with the calculated erosion resistance index to form the first complete set of training data. By adjusting the initial topographic conditions of the target soil sample, the erosion experiment is carried out again for each new adjusted initial topographic condition. The steps of collecting initial topographic, final topographic, and fixed-point elevation time series data are repeated, and the corresponding erosion resistance index is calculated, thereby obtaining multiple sets of training datasets under different initial topographic conditions. The first training dataset is then combined with multiple training datasets to form a comprehensive training dataset covering multiple initial topographic conditions.

[0216] Therefore, by controlling variables to ensure that the dataset contains soil erosion resistance data under different terrain features, the trained model can better adapt to the diverse initial soil conditions in reality, thereby improving the model's generalization ability and prediction accuracy.

[0217] In summary, the specific method for in-situ soil erosion resistance testing based on adversarial neural network algorithms is as follows: Figure 10 As shown, Figure 10 This is a flowchart of an in-situ soil erosion resistance testing method based on an adversarial neural network algorithm according to an embodiment of this application.

[0218] S1001, Conduct indoor experiments to determine the relationship between water flow shear stress and flow velocity;

[0219] S1002, Conduct indoor water tank soil erosion resistance test;

[0220] S1003, collect experimental data, including initial terrain, intermediate terrain, final terrain, and ground elevation data at different time series;

[0221] S1004, based on the topographic / temporal data of S1003, calculates the soil erosion rate under different conditions;

[0222] S1005, integrate the terrain data, subsurface elevation data at different time series, and soil erosion rate to obtain the model training dataset, and then proceed to step S1009.

[0223] S1006, Conduct in-situ soil erosion resistance tests in the field;

[0224] S1007 collects field experimental data, including initial terrain, final terrain, and ground surface elevation data at different time series.

[0225] S1008, Integrate the field data to obtain the model test dataset, and proceed to step S1009;

[0226] S1009: Train the adversarial neural network model using the training dataset of S1005, and input the model test dataset into the trained model.

[0227] S1010, the output of the adversarial neural network model predicts the soil's erosion resistance.

[0228] Therefore, by combining the process of training the model through indoor controlled experiments with verifying the model through real field experiments, and leveraging the learning capabilities of adversarial neural networks, accurate prediction of soil erosion resistance can be achieved, providing technical support for soil stability assessment.

[0229] The in-situ erosion resistance testing method for riverbank soil based on adversarial neural network algorithm proposed in this application involves collecting three-dimensional topographic data of the area before and after erosion using three-dimensional topographic scanning technology and collecting time series data of bed elevation using fixed-point elevation monitoring equipment. These data are then input into a comprehensive training dataset to train an adversarial neural network to obtain a pre-trained in-situ soil erosion resistance prediction model, thereby obtaining the erosion resistance performance index of the soil area to be tested. This method solves the problem that existing erosion resistance testing lacks the ability to fuse multi-source data and dynamically predict the erosion process, greatly improving the efficiency and accuracy of riverbank erosion resistance evaluation.

[0230] Next, referring to the accompanying drawings, a test device for in-situ erosion resistance of riverbank soil based on an adversarial neural network algorithm is described according to an embodiment of this application.

[0231] Figure 11 This is a block diagram of an in-situ erosion resistance testing device for riverbank soil based on an adversarial neural network algorithm according to an embodiment of this application.

[0232] like Figure 11As shown, the in-situ erosion resistance testing device 10 for riverbank soil based on an adversarial neural network algorithm includes: a determination module 100, a data acquisition module 200, and a testing module 300, wherein...

[0233] Module 100 is used to determine the area of ​​soil to be measured.

[0234] The acquisition module 200 is used to acquire initial three-dimensional topographic image data of the soil area to be measured to obtain topographic data before scour, and to carry out scour operation in the soil area to be measured based on the preset in-situ scour strategy. After the scour is completed, the module acquires the final three-dimensional topographic image data of the soil area to be measured to obtain topographic data after scour, and also acquires time series data of bed elevation at preset fixed points.

[0235] The test module 300 is used to input the topographic data before scour, the topographic data after scour, and the time series data of the bed elevation of the preset fixed points into the pre-trained in-situ soil scour resistance prediction model to obtain the scour resistance performance index of the soil area to be tested. The pre-trained in-situ soil scour resistance prediction model is obtained by training an adversarial neural network with a comprehensive training dataset.

[0236] According to one embodiment of this application, before inputting the pre-scour topographic data, post-scour topographic data, and time series data of bed elevation at preset fixed points into the pre-trained in-situ soil scour resistance prediction model, the testing module 300 is further configured to: construct a comprehensive training dataset, dividing the comprehensive training dataset into a training set, a validation set, and a test set based on a preset partitioning ratio; construct an adversarial neural network, inputting the training set into the adversarial neural network for training to obtain initial model parameters; based on the initial model parameters, inputting the validation set into the adversarial neural network for performance evaluation, and adjusting the initial model parameters according to the performance evaluation results until the joint loss function of the validation set converges to obtain the optimal model parameters; based on the optimal model parameters, inputting the test set into the adversarial neural network for model testing, and obtaining the pre-trained in-situ soil scour resistance prediction model when the test results meet preset requirements.

[0237] According to one embodiment of this application, the test module 300 is further configured to: construct a water flow shear stress-velocity model; acquire initial topographic data and final topographic data after scouring of the target soil sample, and simultaneously collect time series data of bed elevation at the target fixed point of the target soil sample; based on the water flow shear stress-velocity model, obtain the scour resistance index of the target soil sample according to the initial topographic data, final topographic data, and time series data of bed elevation at the target fixed point; and obtain a comprehensive training dataset according to the initial topographic data, final topographic data, time series data of bed elevation at the target fixed point, and the scour resistance index of the target soil sample.

[0238] According to one embodiment of this application, the test module 300 is further configured to: acquire water flow shear stress measurements at multiple different flow velocities; fit and plot a water flow shear stress-velocity relationship curve based on the water flow shear stress measurements at multiple different flow velocities; and establish a water flow shear stress-velocity model based on the water flow shear stress-velocity relationship curve. According to one embodiment of this application, the test module is further configured to: perform an initial scan of the target soil sample using a 3D terrain scanner to obtain initial terrain data; determine the initial flow velocity, and based on a preset duration, scour the target soil sample according to the initial flow velocity to obtain intermediate terrain data; adjust the initial flow velocity based on a preset adjustment strategy, and based on the adjusted flow velocity, re-execute the step of scour the target soil sample according to the initial flow velocity for a preset duration to obtain intermediate terrain data, thereby obtaining soil scour response data under multiple flow velocity conditions; and obtain final terrain data by scanning the target soil sample after scourning it using a 3D terrain scanner, and collect time series data of the bed elevation of the target fixed point of the target soil sample.

[0239] According to one embodiment of this application, the test module 300 is further configured to: determine the volume change of soil at each flow velocity based on intermediate topographic data corresponding to each flow velocity; determine the shear stress of water flow at each flow velocity based on the water flow shear stress-velocity model; obtain the soil erosion rate at each flow velocity based on the volume change of soil at each flow velocity, according to the intermediate topographic data and the shear stress of water flow at each flow velocity; establish a mapping relationship between the soil erosion rate at each flow velocity and the shear stress of water flow at each flow velocity; and calculate the erosion resistance index of the target soil sample according to the mapping relationship.

[0240] According to one embodiment of this application, the testing module 300 is further configured to: obtain a first training dataset based on initial terrain data, final terrain data, time series data of bed elevation of target fixed points, and erosion resistance index of target soil samples; adjust the initial terrain conditions of target soil samples, and based on the adjusted target soil samples, re-execute the steps of obtaining initial terrain data and final terrain data after erosion of target soil samples, and synchronously collect time series data of bed elevation of target fixed points of target soil samples, to obtain multiple training datasets; and obtain a comprehensive training dataset based on the first training dataset and multiple training datasets.

[0241] It should be noted that the foregoing explanation of the embodiment of the in-situ erosion resistance testing method for riverbank soil based on the adversarial neural network algorithm also applies to the in-situ erosion resistance testing device for riverbank soil based on the adversarial neural network algorithm in this embodiment, and will not be repeated here.

[0242] The in-situ erosion resistance testing device for riverbank soil based on the adversarial neural network algorithm proposed in this application uses three-dimensional terrain scanning technology to collect three-dimensional terrain data of the area before and after erosion and fixed-point elevation monitoring equipment to collect time series data of bed elevation. These data are then input into a comprehensive training dataset to train the adversarial neural network to obtain a pre-trained in-situ soil erosion resistance prediction model, thereby obtaining the erosion resistance performance index of the soil area to be tested. This method solves the problem that existing erosion resistance testing lacks the ability to fuse multi-source data and dynamically predict the erosion process, greatly improving the efficiency and accuracy of riverbank erosion resistance evaluation.

[0243] Figure 12 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. The electronic device may include:

[0244] The memory 1201, the processor 1202, and the computer program stored on the memory 1201 and executable on the processor 1202.

[0245] When the processor 1202 executes the program, it implements the in-situ erosion resistance test method for riverbank soil based on the adversarial neural network algorithm provided in the above embodiments.

[0246] Furthermore, electronic devices also include:

[0247] Communication interface 1203 is used for communication between memory 1201 and processor 1202.

[0248] The memory 1201 is used to store computer programs that can run on the processor 1202.

[0249] The memory 1201 may include high-speed RAM (Random Access Memory) memory, and may also include non-volatile memory, such as at least one disk storage.

[0250] If the memory 1201, processor 1202, and communication interface 1203 are implemented independently, then the communication interface 1203, memory 1201, and processor 1202 can be interconnected via a bus to complete communication between them. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 12The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0251] Optionally, in a specific implementation, if the memory 1201, processor 1202, and communication interface 1203 are integrated on a single chip, then the memory 1201, processor 1202, and communication interface 1203 can communicate with each other through an internal interface.

[0252] Processor 1202 may be a CPU (Central Processing Unit), an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement embodiments of the present invention.

[0253] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for in-situ erosion resistance testing of riverbank soil based on an adversarial neural network algorithm.

[0254] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0255] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0256] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for in-situ erosion resistance testing of riverbank soil based on an adversarial neural network algorithm, characterized in that, Includes the following steps: Determine the area of ​​soil to be tested; Initial three-dimensional topographic image data of the soil area to be tested is collected to obtain the topographic data before scour. Based on the preset in-situ scour strategy, scour operation is carried out in the soil area to be tested. After the scour is completed, the final three-dimensional topographic image data of the soil area to be tested is collected to obtain the topographic data after scour. Time series data of bed elevation at preset fixed points are also collected. The preset in-situ scour strategy includes scour time and scour velocity. The topographic data before scour, the topographic data after scour, and the time series data of bed elevation at the preset fixed point are input into a pre-trained in-situ soil scour resistance prediction model to obtain the scour resistance performance index of the soil area to be tested. The pre-trained in-situ soil scour resistance prediction model is obtained by training an adversarial neural network with a comprehensive training dataset. The method further includes, before inputting the pre-scour topographic data, the post-scour topographic data, and the time series data of bed elevation at the preset fixed points into the pre-trained in-situ soil scour resistance prediction model, the following steps: constructing the comprehensive training dataset by dividing it into a training set, a validation set, and a test set based on a preset partitioning ratio; constructing the adversarial neural network by inputting the training set into the adversarial neural network for training to obtain initial model parameters; inputting the validation set into the adversarial neural network for performance evaluation based on the initial model parameters, and adjusting the initial model parameters according to the performance evaluation results until the joint loss function of the validation set converges to obtain the optimal model parameters; and inputting the test set into the adversarial neural network for model testing based on the optimal model parameters, and obtaining the pre-trained in-situ soil scour resistance prediction model when the test results meet preset requirements. The construction of the comprehensive training dataset includes: constructing a water flow shear stress-velocity model; acquiring initial topographic data and final topographic data after scouring of the target soil sample, and simultaneously collecting time series data of bed elevation at target fixed points of the target soil sample; based on the water flow shear stress-velocity model, obtaining the scour resistance index of the target soil sample according to the initial topographic data, the final topographic data, and the time series data of bed elevation at the target fixed points; and obtaining the comprehensive training dataset according to the initial topographic data, the final topographic data, the time series data of bed elevation at the target fixed points, and the scour resistance index of the target soil sample. The process of acquiring initial topographic data and final topographic data after scouring of a target soil sample, and simultaneously collecting time-series data on the bed elevation of the target fixed point of the target soil sample, includes: using a 3D topographic scanner to perform an initial scan of the target soil sample to obtain the initial topographic data; determining an initial flow velocity, and based on a preset duration, scouring the target soil sample according to the initial flow velocity to obtain intermediate topographic data; adjusting the initial flow velocity based on a preset adjustment strategy, and based on the adjusted flow velocity, re-executing the step of scouring the target soil sample according to the initial flow velocity for a preset duration to obtain intermediate topographic data, thereby obtaining soil scouring response data under multiple flow velocity conditions; and using the 3D topographic scanner to scan the target soil sample after scouring to obtain the final topographic data, and collecting time-series data on the bed elevation of the target fixed point of the target soil sample. The method for obtaining the erosion resistance index of the target soil sample based on the water flow shear stress-velocity model, according to the initial topographic data, the final topographic data, and the time series data of the bed elevation of the target fixed point, includes: determining the soil volume change corresponding to each flow velocity based on the intermediate topographic data corresponding to each flow velocity; determining the water flow shear stress corresponding to each flow velocity based on the water flow shear stress-velocity model; obtaining the soil erosion rate corresponding to each flow velocity based on the soil volume change corresponding to each flow velocity, according to the intermediate topographic data and the water flow shear stress corresponding to each flow velocity; establishing a mapping relationship between the soil erosion rate and the water flow shear stress corresponding to each flow velocity; and calculating the erosion resistance index of the target soil sample according to the mapping relationship.

2. The method for in-situ erosion resistance testing of riverbank soil based on adversarial neural network algorithm according to claim 1, characterized in that, The construction of the water flow shear stress-velocity model includes: Obtain water flow shear stress measurements at multiple different flow velocities; Based on the measured values ​​of water flow shear stress at multiple different flow velocities, a water flow shear stress-flow velocity relationship curve was plotted. Based on the water flow shear stress-velocity relationship curve, the water flow shear stress-velocity model is established.

3. The method for in-situ erosion resistance testing of riverbank soil based on adversarial neural network algorithm according to claim 1, characterized in that, The comprehensive training dataset, obtained based on the initial terrain data, the final terrain data, the time series data of the bed elevation of the target fixed point, and the erosion resistance index of the target soil sample, includes: The first training dataset is obtained based on the initial terrain data, the final terrain data, the time series data of the bed elevation of the target fixed point, and the erosion resistance index of the target soil sample. Adjust the initial terrain conditions of the target soil sample, and based on the adjusted target soil sample, re-execute the steps of obtaining the initial terrain data and the final terrain data after scouring of the target soil sample, and simultaneously collect the time series data of the bed elevation of the target fixed point of the target soil sample to obtain multiple training datasets. The comprehensive training dataset is obtained based on the first training dataset and the plurality of training datasets.

4. A device for testing the in-situ erosion resistance of riverbank soil based on an adversarial neural network algorithm, characterized in that, include: The determination module is used to determine the area of ​​soil to be tested; The acquisition module is used to acquire initial three-dimensional topographic image data of the soil area to be measured to obtain topographic data before scour, and to perform scour operation on the soil area to be measured based on a preset in-situ scour strategy. After scour is completed, the module acquires final three-dimensional topographic image data of the soil area to be measured to obtain topographic data after scour, and also acquires time series data of bed elevation at preset fixed points. The preset in-situ scour strategy includes scour time and scour velocity. The testing module is used to input the pre-scour topographic data, the post-scour topographic data, and the time series data of the bed elevation of the preset fixed point into a pre-trained in-situ soil scour resistance prediction model to obtain the scour resistance performance index of the soil area to be tested. The pre-trained in-situ soil scour resistance prediction model is obtained by training an adversarial neural network with a comprehensive training dataset. Before inputting the pre-scour topographic data, the post-scour topographic data, and the time series data of bed elevation at the preset fixed points into the pre-trained in-situ soil scour resistance prediction model, the testing module is further configured to: construct the comprehensive training dataset, dividing the comprehensive training dataset into a training set, a validation set, and a test set based on a preset partitioning ratio; construct the adversarial neural network, inputting the training set into the adversarial neural network for training to obtain initial model parameters; based on the initial model parameters, inputting the validation set into the adversarial neural network for performance evaluation, and adjusting the initial model parameters according to the performance evaluation results until the joint loss function of the validation set converges to obtain the optimal model parameters; based on the optimal model parameters, inputting the test set into the adversarial neural network for model testing, and obtaining the pre-trained in-situ soil scour resistance prediction model when the test results meet preset requirements; The testing module is further configured to: construct a water flow shear stress-velocity model; acquire initial topographic data and final topographic data after scouring of the target soil sample, and simultaneously collect time-series data of bed elevation at target fixed points of the target soil sample; based on the water flow shear stress-velocity model, obtain the scour resistance index of the target soil sample according to the initial topographic data, the final topographic data, and the time-series data of bed elevation at the target fixed points; and obtain the comprehensive training dataset according to the initial topographic data, the final topographic data, the time-series data of bed elevation at the target fixed points, and the scour resistance index of the target soil sample. The testing module is further configured to: perform an initial scan of the target soil sample using a 3D terrain scanner to obtain the initial terrain data; determine the initial flow velocity, and based on a preset duration, scour the target soil sample according to the initial flow velocity to obtain intermediate terrain data; adjust the initial flow velocity based on a preset adjustment strategy, and based on the adjusted flow velocity, re-execute the step of scourting the target soil sample according to the initial flow velocity for a preset duration to obtain intermediate terrain data, thereby obtaining soil scour response data under multiple flow velocity conditions; and use the 3D terrain scanner to scour and scan the target soil sample to obtain the final terrain data, and collect time series data of the bed elevation of the target fixed point of the target soil sample. The testing module is further configured to: determine the change in soil volume for each flow velocity based on intermediate topographic data corresponding to each flow velocity; determine the shear stress for each flow velocity based on the water flow shear stress-velocity model; obtain the soil erosion rate for each flow velocity based on the change in soil volume for each flow velocity, according to the intermediate topographic data and the shear stress for each flow velocity; establish a mapping relationship between the soil erosion rate and the shear stress for each flow velocity; and calculate the erosion resistance index of the target soil sample based on the mapping relationship.

5. An electronic device, characterized in that, include: The memory, the processor, and the computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the in-situ erosion resistance testing method for riverbank soil based on an adversarial neural network algorithm as described in any one of claims 1-3.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by the processor to implement the in-situ erosion resistance testing method for riverbank soil based on the adversarial neural network algorithm as described in any one of claims 1-3.

Images