Method for monitoring potential erosion holes of side slope of dispersive soil

Abstract

This invention relates to the field of soil slope monitoring technology, specifically to a method for monitoring cavitation erosion in dispersed soil slopes. The method involves acquiring soil moisture, soil temperature, air temperature, air humidity, current rainfall, slope runoff velocity, and images of the soil slope surface and surface water for a target dispersed soil slope. Based on these data, the method determines the rainfall erosion index and soil cavitation sensitivity index for the target dispersed soil slope. Finally, based on these indices, the method determines the risk level of cavitation erosion development for the target dispersed soil slope. This comprehensive and dynamic assessment of cavitation erosion in dispersed soil slopes improves the targeting and timeliness of slope disaster identification and diagnosis, thereby reducing slope disaster risks.

Description

Technical Field

[0001] This invention relates to the field of soil slope monitoring technology, specifically to a method for monitoring dispersed soil slope erosion cavities. Background Technology

[0002] Monitoring the erosion cavities in dispersible soil slopes is crucial for mitigating engineering disasters and ensuring the long-term safe operation of infrastructure such as water conservancy and roads. Because dispersible soils easily disperse and disintegrate in low-salinity water, they are prone to gully erosion, piping, and other damage in engineering practice, forming erosion cavities. Scientific monitoring can promptly detect early risk signals such as soil dispersion and abnormal cracking, allowing for proactive protective measures to safeguard life and property.

[0003] In existing technologies, the monitoring of cavitation erosion in dispersed soil slopes largely follows the monitoring model for ordinary soil slopes, failing to specifically adapt to the unique engineering characteristics of dispersed soils. For example, the degree of soil dispersion has a significant impact on the formation of cavitation erosion in dispersed soil slopes, but traditional monitoring methods are difficult to monitor the degree of soil dispersion. Therefore, they cannot fully capture slope risk signals and cannot comprehensively assess the degree of deterioration of the slope structure.

[0004] Therefore, developing a monitoring method for scattered soil erosion cavities on slopes has become a pressing technical challenge. Summary of the Invention

[0005] This invention provides a method for monitoring dispersed soil erosion cavities on slopes, in order to solve the problem of developing a monitoring method for dispersed soil erosion cavities on slopes.

[0006] In a first aspect, the present invention provides a method for monitoring cavitation erosion in dispersed soil slopes. The method includes: acquiring soil moisture data, soil temperature data, and ambient air temperature and humidity data for the target dispersed soil slope, as well as the current rainfall and surface runoff velocity during rainfall for the target dispersed soil slope; acquiring images of the soil slope surface and surface water at the interface between the target dispersed soil slope and a water body; determining the rainfall erosion index and soil cavitation sensitivity index for the target dispersed soil slope based on the soil moisture data, soil temperature data, air temperature data, air humidity data, current rainfall, surface runoff velocity, soil slope surface image, and surface water image; and determining the cavitation erosion erosion risk level for the target dispersed soil slope based on the rainfall erosion index and soil cavitation sensitivity index.

[0007] In one optional implementation, based on soil moisture data, soil temperature data, air temperature data, air humidity data, current rainfall, slope runoff velocity, soil slope surface images, and surface water images, the rainfall erosion index and soil erosion sensitivity index corresponding to the target dispersed soil slope are determined. This includes: identifying the soil slope surface images to determine the development characteristics of erosion cavities corresponding to the target dispersed soil slope; wherein, the development characteristics of erosion cavities include geometric morphology, distribution, and development degree; identifying the surface water images to determine the turbidity of the surface water at the interface between the target dispersed soil slope and the water body; and determining the rainfall erosion index and soil erosion sensitivity index corresponding to the target dispersed soil slope based on soil moisture data, soil temperature data, air temperature data, air humidity data, current rainfall, slope runoff velocity, erosion cavity development characteristics, and surface water turbidity.

[0008] In one optional implementation, the rainfall erosion index and soil erosion sensitivity index corresponding to the target dispersed soil slope are determined based on soil moisture data, soil temperature data, air temperature data, air humidity data, current rainfall, slope runoff velocity, cavitation development characteristics, and surface water turbidity. This includes: generating target features based on soil moisture data, soil temperature data, air temperature data, air humidity data, current rainfall, slope runoff velocity, cavitation development characteristics, and surface water turbidity; inputting the target features into a preset cavitation processing model; and outputting the rainfall erosion index and soil erosion sensitivity index corresponding to the target dispersed soil slope.

[0009] In one optional implementation, the preset erosion cavity treatment model includes a shared feature extraction layer, a rainfall erosion index prediction layer, and a soil erosion sensitivity index prediction layer. Inputting target features into the preset erosion cavity treatment model and outputting the rainfall erosion index and soil erosion sensitivity index corresponding to the target dispersed soil slope includes: inputting the target features into the shared feature extraction layer in the preset erosion cavity treatment model; the shared feature extraction layer extracting features from the target features and outputting a shared feature vector; inputting the shared feature vector into the rainfall erosion index prediction layer and outputting the rainfall erosion index; and inputting the shared feature vector into the soil erosion sensitivity index prediction layer and outputting the soil erosion sensitivity index.

[0010] In one optional implementation, the shared feature vector is input into the rainfall erosion index prediction layer to output the rainfall erosion index, including: obtaining the slope runoff velocity and surface water turbidity; determining the first target weights corresponding to the slope runoff velocity and surface water turbidity respectively based on the relationship between the slope runoff velocity and the rainfall erosion index, and the relationship between the surface water turbidity and the rainfall erosion index; generating rainfall erosion features based on the slope runoff velocity, surface water turbidity, and the first target weights; fusing the shared feature vector and the rainfall erosion features to generate a first fused feature; and inputting the first fused feature into the rainfall erosion index prediction layer to output the rainfall erosion index.

[0011] In one optional implementation, the shared feature vector is input into the soil erosion sensitivity index prediction layer, and the soil erosion sensitivity index is output. This includes: obtaining erosion cavity development characteristics; determining a second target weight corresponding to the erosion cavity development characteristics based on the relationship between the erosion cavity development characteristics and the soil erosion sensitivity index; generating crack development characteristics based on the erosion cavity development characteristics and the second target weight; fusing the shared feature vector and the crack development characteristics to generate a second fused feature; and inputting the second fused feature into the soil erosion sensitivity index prediction layer to output the soil erosion sensitivity index.

[0012] In one optional implementation, the risk level of cavitation development corresponding to the target dispersed soil slope is determined based on the rainfall erosion index and the soil cavitation sensitivity index, including: determining the current climate scenario type based on the current rainfall; determining the target index weights corresponding to the rainfall erosion index and the soil cavitation sensitivity index based on the current climate scenario type; calculating the cavitation development risk score corresponding to the target dispersed soil slope based on the target index weights; and determining the cavitation development risk level corresponding to the target dispersed soil slope based on the cavitation development risk score.

[0013] The method for monitoring pore formation on dispersed soil slopes provided in this application acquires soil moisture and temperature data of the surface soil corresponding to the target dispersed soil slope, as well as air temperature and humidity data of the surrounding air, current rainfall, and slope runoff velocity during rainfall. This comprehensively covers the core environmental and hydrological elements affecting the development of pore formation on dispersed soil slopes, providing complete basic data support for subsequent index calculations. It avoids analytical biases caused by missing data in a single dimension, while accurately capturing the interaction between soil, air, and water (such as the driving force of rainfall on slope runoff and the influence of air temperature and humidity on soil moisture evaporation), laying the foundation for the accuracy of subsequent index calculations. It acquires surface images of the soil slope corresponding to the target dispersed soil slope and surface water images at the interface between the target dispersed soil slope and water bodies. Visual data supplements the limitations of traditional sensor data, intuitively capturing the direct characteristics of slope structural damage (pore formation) and water erosion (turbidity changes). Image data can quantify geometric parameters such as the area and density of pores, as well as the degree of soil loss corresponding to surface water turbidity, providing crucial structural and erosion characterization for rainfall erosion index and soil pore erosion sensitivity index. Based on soil moisture data, soil temperature data, air temperature data, air humidity data, current rainfall, slope runoff velocity, soil slope surface images, and surface water images, the rainfall erosion index and soil pore erosion sensitivity index corresponding to the target dispersed soil slope are determined. Multi-dimensional raw data are transformed into quantified pore development risk assessment indicators, achieving precise characterization of slope erosion and structural damage. REI focuses on erosion risk in rainfall scenarios, while CDI focuses on fissure development risk in non-rainfall scenarios. These two indicators correspond to the core influencing factors of pore development under different working conditions, providing a scientific and quantifiable decision-making basis for subsequent pore development level determination, avoiding the subjectivity of traditional qualitative assessments. Based on the rainfall erosion index and soil pore erosion sensitivity index, the pore development risk level corresponding to the target dispersed soil slope is determined. By combining the synergistic evaluation of two types of core indices, a comprehensive and dynamic assessment of the risk of slope erosion cavity development can be achieved. Differentiated weights are applied to adapt to different climatic scenarios, ensuring that the risk level classification of erosion cavity development aligns with actual working conditions. Clearly defined classifications can directly guide engineering response measures, improving the targeting and timeliness of slope disaster prevention and control, and reducing slope disaster risks. Attached Figure Description

[0014] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0015] Figure 1 This is a schematic diagram of an application scenario according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the first process of a method for monitoring dispersed soil slope erosion cavities according to an embodiment of the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0018] 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 one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0019] According to an embodiment of the present invention, a method for monitoring dispersed soil slope erosion cavities is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0020] This embodiment provides a method for monitoring dispersed soil slope erosion cavities, which can be used in electronic devices. Figure 1 This is a flowchart of a method for monitoring dispersed soil slope erosion cavities according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps: Step S101: Obtain soil moisture data, soil temperature data, air temperature data, and air humidity data of the surrounding air corresponding to the target dispersed soil slope, as well as the current rainfall and slope runoff velocity during rainfall corresponding to the target dispersed soil slope.

[0021] Specifically, electronic devices can acquire soil moisture and soil temperature data of the surface soil corresponding to the target dispersed soil slope based on a communication connection with a high-frequency time domain reflectometer (TDR) soil sensor.

[0022] Electronic devices can acquire air temperature and humidity data of the surrounding air corresponding to the target dispersed soil slope based on communication connection with an integrated temperature and humidity sensor.

[0023] The electronic device can also obtain the current rainfall corresponding to the target dispersed soil slope based on the communication connection between the electronic device and the self-metering rain gauge placed at the corresponding location of the target dispersed soil slope.

[0024] In addition, a water storage tank containing a blue solution is placed at a predetermined location on the target dispersed soil slope. Electronic equipment can capture images of the flow of the blue solution in the tank and calculate the slope runoff velocity. Specifically, the electronic equipment can use image recognition to calculate the slope runoff velocity based on the time difference (Δt) between the tracer tip and two cross-sections, using the formula S=L / Δt (where L is the distance between the two cross-sections, set to 5m).

[0025] Step S102: Obtain the surface image of the soil slope corresponding to the target dispersed soil slope and the surface water image at the junction of the target dispersed soil slope and the water body.

[0026] Specifically, electronic devices can acquire images of the soil slope surface corresponding to the target dispersed soil slope and the surface water at the interface between the target dispersed soil slope and the water body based on the communication connection with the camera.

[0027] Step S103: Based on soil moisture data, soil temperature data, air temperature data, air humidity data, current rainfall, slope runoff velocity, soil slope surface image, and surface water image, determine the rainfall erosion index and soil erosion sensitivity index corresponding to the target dispersed soil slope.

[0028] Specifically, the electronic device can input soil moisture data, soil temperature data, air temperature data, air humidity data, current rainfall, slope runoff velocity, soil slope surface image, and surface water image into a preset erosion cavity processing model, and output the rainfall erosion index and soil erosion sensitivity index corresponding to the target dispersed soil slope.

[0029] This step will be explained in detail below.

[0030] Step S104: Based on the rainfall erosion index and the soil erosion sensitivity index, determine the risk level of erosion hole development corresponding to the target dispersed soil slope.

[0031] Specifically, electronic devices can determine the rainfall erosion level based on the rainfall erosion index and the fissure development level based on the soil erosion sensitivity index. Then, based on the rainfall erosion level and the fissure development level, the risk level of pore development corresponding to the target dispersed soil slope can be determined.

[0032] This step will be explained in detail below.

[0033] The method for monitoring pore formation on dispersed soil slopes provided in this application acquires soil moisture and temperature data of the surface soil corresponding to the target dispersed soil slope, as well as air temperature and humidity data of the surrounding air, current rainfall, and slope runoff velocity during rainfall. This comprehensively covers the core environmental and hydrological elements affecting the development of pore formation on dispersed soil slopes, providing complete basic data support for subsequent index calculations. It avoids analytical biases caused by missing data in a single dimension, while accurately capturing the interaction between soil, air, and water (such as the driving force of rainfall on slope runoff and the influence of air temperature and humidity on soil moisture evaporation), laying the foundation for the accuracy of subsequent index calculations. It acquires surface images of the soil slope corresponding to the target dispersed soil slope and surface water images at the interface between the target dispersed soil slope and water bodies. Visual data supplements the limitations of traditional sensor data, intuitively capturing the direct characteristics of slope structural damage (pore formation) and water erosion (turbidity changes). Image data can quantify geometric parameters such as the area and density of pores, as well as the degree of soil loss corresponding to surface water turbidity, providing crucial structural and erosion characterization for rainfall erosion index and soil pore erosion sensitivity index. Based on soil moisture data, soil temperature data, air temperature data, air humidity data, current rainfall, slope runoff velocity, soil slope surface images, and surface water images, the rainfall erosion index and soil pore erosion sensitivity index corresponding to the target dispersed soil slope are determined. Multi-dimensional raw data are transformed into quantitative evaluation indicators for pore development, achieving precise characterization of slope erosion and structural damage. REI focuses on erosion risk in rainfall scenarios, while CDI focuses on fissure development risk in non-rainfall scenarios. These two indicators correspond to the core influencing factors of pore development under different working conditions, providing a scientific and quantifiable decision-making basis for subsequent pore development risk level determination, avoiding the subjectivity of traditional qualitative evaluation. Based on the rainfall erosion index and soil pore erosion sensitivity index, the pore development risk level corresponding to the target dispersed soil slope is determined. By combining the synergistic evaluation of two types of core indices, a comprehensive and dynamic assessment of the development risk of burrowing cavities on slopes can be achieved. Differentiated weighting is applied to adapt to different climatic scenarios (emphasizing REI during rainfall and CDI during non-rainfall periods) to ensure that the classification of burrowing cavities aligns with actual working conditions. Clear classification can directly guide engineering response measures (such as routine monitoring and emergency reinforcement), improving the targeting and timeliness of slope disaster prevention and control, and reducing the risk of instability.

[0034] This embodiment provides a method for monitoring dispersed soil slope erosion cavities, which can be used in electronic devices. Figure 2 This is a flowchart of a method for monitoring dispersed soil slope erosion cavities according to an embodiment of the present invention, such as... Figure 2 As shown, the process includes the following steps: Step S201: Obtain soil moisture data, soil temperature data, air temperature data and air humidity data of the surrounding air corresponding to the target dispersed soil slope, as well as the current rainfall and slope runoff velocity during rainfall corresponding to the target dispersed soil slope.

[0035] Please refer to the above description of step S101 for details on this step, which will not be repeated here.

[0036] Step S202: Obtain the surface image of the soil slope corresponding to the target dispersed soil slope and the surface water image at the interface between the target dispersed soil slope and the water body.

[0037] Please refer to the above description of step S102 for details on this step, which will not be repeated here.

[0038] Step S203: Based on soil moisture data, soil temperature data, air temperature data, air humidity data, current rainfall, slope runoff velocity, soil slope surface image, and surface water image, determine the rainfall erosion index and soil erosion sensitivity index corresponding to the target dispersed soil slope.

[0039] Specifically, step S203 above may include the following steps: Step S2031: Identify the surface image of the soil slope and determine the development characteristics of the cavities corresponding to the target dispersed soil slope.

[0040] Among them, the development characteristics of burrowing cavities include geometric morphological characteristics, distribution characteristics, and development degree characteristics.

[0041] Specifically, the electronic device can use the SIFT (Scale Invariant Feature Transform) algorithm to match feature points and stitch images from overlapping soil slope surface images of the same monitoring unit, generating a complete 1m×1m unit image. Then, it divides the image into 50cm×50cm sub-regions and crops them into sub-images, reducing the computational complexity of a single image. Next, the electronic device uses a bilateral filtering algorithm to process each sub-image. This algorithm can remove Gaussian noise and salt-and-pepper noise while preserving detailed information about the hole edges. The filtering parameters are set as follows: spatial domain standard deviation σs=5, gray-level domain standard deviation σr=25. Finally, the electronic device can use the CLAHE (Contrast Limiting Adaptive Histogram Equalization) algorithm to enhance image contrast, maximizing the gray-level difference between the holes (dark areas) and the slope soil (bright areas). The parameters are set as follows: block size 8×8, contrast limit threshold 4.0, improving the identifiability of the holes.

[0042] Then, the electronic device can convert the color image into a grayscale image, and then use the Otsu's algorithm to automatically calculate the binarization threshold, dividing the image into "foreground (cavities)" and "background (soil)" to generate a black and white binary image, thus achieving the initial separation of cavities from the background.

[0043] Electronic devices can perform morphological opening operations (erosion followed by dilation) on binary images to remove small noise points (such as pixel clusters of weeds or pebbles), with a 3×3 rectangular kernel selected as the structuring element. Then, morphological closing operations (dilation followed by erosion) are performed to fill the tiny gaps inside the holes and repair broken parts at the edges of the holes, resulting in complete candidate regions for holes.

[0044] Electronic devices can perform target detection on candidate hole regions based on a preset target detection model. This preset target detection model can be the YOLOv8-nano model, or other models.

[0045] The training process of the preset target detection model is as follows: Slope images under different working conditions are collected, and the areas of erosion holes are manually labeled. The images are then divided into training, validation, and test sets in a 7:2:1 ratio. Labeling categories include "small holes (diameter < 5cm)," "medium holes (5-15cm)," and "large holes (> 15cm)." The input image size is set to 640×640, the batch size to 16, the learning rate to 0.001, and the number of training epochs to 100. The CIoU loss function is used to optimize the bounding box regression accuracy and improve the accuracy of hole localization. The preprocessed images are input into the trained model, which outputs the bounding box coordinates (x1, y1, x2, y2), category labels, and confidence scores for all erosion holes in each image. A confidence threshold of 0.7 is set to eliminate candidate regions with confidence scores below the threshold, reducing the false detection rate.

[0046] For each detected undercut hole, the electronic device extracts and quantifies three core indicators: geometric morphology, distribution, and development level, to comprehensively characterize its development status.

[0047] Among them, geometric morphological characteristics reflect the size and shape of a single hole and are a basic indicator for judging the degree of hazard caused by a hole. The calculation method is as follows: Hole area (S): Count the number of pixels within the bounding box of each hole, and convert it to the actual area (unit: cm²) based on the image pixel resolution (e.g., 1 pixel corresponds to 0.1 cm); Hole equivalent diameter (D): Approximate the hole as a circle, and calculate it according to the formula... Calculate the equivalent diameter, reflecting the size of the hole; Hole shape factor (F): according to the formula The shape factor (L is the perimeter of the hole) is calculated as follows: the closer the shape factor is to 1, the closer the hole is to a circle; the further it deviates from 1, the more irregular the shape of the hole (such as long strips or branching shapes). Irregular holes have stronger connectivity and higher risks. Hole depth (H): The depth is estimated using binocular stereo vision: two cameras are used to take stereo images of the same hole, the parallax is calculated, and combined with the camera calibration parameters, the actual depth of the hole is calculated according to the principle of triangulation. The formula is H=B×f / d (B is the camera baseline distance, f is the focal length, and d is the parallax).

[0048] Among them, the distribution characteristics reflect the spatial distribution pattern of holes on the slope surface and are a key indicator for judging the overall damage level of the slope: Pore ​​density (ρ): The number of pores per square meter of slope surface, calculated using the formula ρ = N / S. total (N is the number of holes, S) total =1m 2 The higher the density, the more severe the soil erosion on the slope; Cavity aggregation degree (C): The aggregation degree is calculated using spatial autocorrelation analysis (Moran's I index). A Moran's I index > 0 indicates that the pores are clustered, and the clustered areas are prone to forming a connected pore network, which is a high-risk area for slope instability; an index < 0 indicates that the pores are discretely distributed, and the risk is relatively low; Cavity depth gradient: The maximum, minimum, and average values ​​of the pore depth within the same monitoring unit are calculated to reflect the longitudinal development differences of the pores. The larger the depth gradient, the more uneven the erosion inside the soil.

[0049] Among them, the developmental degree characteristics reflect the dynamic evolution trend of the pores and need to be calculated in conjunction with historical monitoring data: Pore ​​expansion rate (v): Comparing the pore area at the same location at different times, according to the formula... Calculation (S) t For the current area, S t-1 (where Δt is the historical area and Δt is the monitoring time interval). Hole connectivity (K): The skeleton of the hole is extracted by the skeleton extraction algorithm. The length of the skeleton is calculated as a percentage of the total perimeter of the hole. The higher the percentage, the stronger the connectivity between holes, and the easier it is for water to form seepage channels inside.

[0050] The electronic device splices together the geometric morphological features, distribution features, and development degree features of the target dispersed soil slope to generate the development features of the cavitation holes corresponding to the target dispersed soil slope.

[0051] Step S2032: Identify the surface water image and determine the turbidity of the surface water at the junction of the target dispersed soil slope and the water body.

[0052] Specifically, electronic devices can perform Canny edge detection (threshold 100-200) on surface water images (captured by a shore-side camera) to extract edge contours with abrupt changes in grayscale. Combining this with the geometric features of the slope toe (slope angle 30°-45°), a straight line is fitted based on the edge contours with abrupt changes in grayscale to serve as the land-water boundary (e.g., in an image, the boundary line is 5m horizontally and 0.8m vertically from the shore-side camera). The area below the land-water boundary line, with a water depth of 0-30cm (i.e., the surface water area), is retained, while the slope soil above the boundary line and the deep water below the boundary line are removed.

[0053] Next, there may be floating objects (such as fallen leaves and foam) in the surface water, which the electronic device can filter out using "color and shape".

[0054] Specifically, the electronic device can mark and remove floating objects (such as the yellowish-brown color of fallen leaves) as impurities based on the light blue hue of the surface water (RGB range: R80-120, G150-190, B200-240). The device can also use an "area threshold" (impurity area < 100 mm²) to remove small floating objects, ensuring that the purity of the extracted water area is ≥ 95%.

[0055] Then, the electronic device can calculate the average gray value of the surface water area after removing impurities. The higher the turbidity, the stronger the scattering of light by the water, and the lower the average gray value (e.g., average gray value of clear water = 180, average gray value of slightly turbid water = 120, average gray value of heavily turbid water = 60).

[0056] In addition, electronic devices can also calculate the dispersion of the average gray value. In turbid water, due to the uneven distribution of sediment particles, the standard deviation of gray value is larger (e.g., the standard deviation of gray value in clear water = 10, and the standard deviation of gray value in heavily turbid water = 35).

[0057] Next, the electronic device can use a gray-level co-occurrence matrix (GLCM) to extract the texture features of the water body, reflecting the distribution density of sediment particles. These texture features include two key texture values: contrast and correlation. Higher turbidity indicates more sediment particles, resulting in greater contrast (significant gray-level difference between particles and water) and lower correlation (more disordered particle distribution). For example, clear water has a Contrast of 20 and a Correlation of 0.8, while heavily turbid water has a Contrast of 80 and a Correlation of 0.3.

[0058] Then, the electronic device can construct a multiple linear regression model based on historical data (500 sets of "image features - measured turbidity" data), and integrate the average gray value, gray standard deviation, and extracted contrast and correlation to form the surface water turbidity D (unit: NTU): D = -2.5 × average gray value + 1.8 × gray standard deviation + 0.5 × Contrast - 30 × Correlation + 50.

[0059] Step S2033: Based on soil moisture data, soil temperature data, air temperature data, air humidity data, current rainfall, slope runoff velocity, characteristics of cavitation development, and surface water turbidity, determine the rainfall erosion index and soil cavitation sensitivity index corresponding to the target dispersed soil slope.

[0060] Specifically, step S2033 above may include the following steps: Step a1: Based on soil moisture data, soil temperature data, air temperature data, air humidity data, current rainfall, slope runoff velocity, development characteristics of cavitation holes, and turbidity of surface water, generate target features.

[0061] Specifically, step a1 above may include the following steps: Step a11 involves classifying soil moisture data, soil temperature data, air temperature data, air humidity data, current rainfall, slope runoff velocity, characteristics of cavitation development, and surface water turbidity to obtain basic environmental characteristics, hydrological erosion characteristics, and structural damage characteristics.

[0062] Among them, structural damage characteristics include soil moisture data, soil temperature data, air temperature data, and air humidity data; hydrological erosion characteristics include current rainfall, slope runoff velocity, and surface water turbidity; and structural damage characteristics include the development characteristics of burrowing cavities.

[0063] Specifically, based on the mechanism of dispersed soil slope erosion, electronic devices can classify eight raw data points into three categories: environmental basic characteristics, hydrological erosion characteristics, and structural damage characteristics. The physical meaning of each category is clarified, laying the foundation for subsequent weighted fusion. Among these, structural damage characteristics include soil moisture data, soil temperature data, air temperature data, and air humidity data; hydrological erosion characteristics include current rainfall, slope runoff velocity, and surface water turbidity; and structural damage characteristics include the development characteristics of erosion cavities.

[0064] Step a12: For each sub-feature among the three types of features, calculate the weight value corresponding to each sub-feature.

[0065] Specifically, step a12 above may include the following steps: Step a121: Calculate the feature percentage corresponding to each sub-feature.

[0066] Specifically, all sub-features are standardized using Min-Max, mapping the values ​​to the [0,1] interval to eliminate dimensional differences. The formula is as follows: Among them, X ij X is the original value of the j-th sub-feature of the i-th sample; min,j X max,j The j-th sub-feature is the historical extreme value (based on statistics of ≥1000 sets of full-condition data).

[0067] For electronic devices, the proportion of standardized sub-features in all samples is calculated to obtain the feature proportion corresponding to the sub-feature. The formula is as follows: Where n is the total number of samples; if p ij =0, then define ln(p) ij = 0 (to avoid logarithms being meaningless).

[0068] Step a122: Calculate the feature entropy value corresponding to each sub-feature based on the feature proportion.

[0069] Specifically, the feature entropy value reflects the degree of information disorder of a sub-feature. The smaller the entropy value, the more effective information the feature contains, and the stronger its ability to characterize latent risk. The formula is: Among them, e j Let be the entropy value of the j-th sub-feature, with a value range of [0,1].

[0070] Step a123: Calculate the weight value corresponding to each sub-feature based on the feature entropy value.

[0071] Specifically, electronic devices can calculate entropy weights based on feature entropy values. Entropy weights are a quantitative indicator of the importance of sub-features, i.e., the weight values ​​corresponding to the sub-features, and the formula is as follows: , where m is the number of sub-features of a certain type of feature; (The sum of the weights is 1, which satisfies the normalization requirement).

[0072] Step a13: Based on the weight values ​​corresponding to each sub-feature, perform inter-group fusion processing on each sub-feature in each category of features to generate comprehensive environmental features, comprehensive hydrological erosion features, and comprehensive structural damage features, respectively.

[0073] Specifically, the electronic device can perform weighted summation of sub-features within each feature category based on the sub-feature weights calculated in step a12, generating comprehensive environmental features, comprehensive hydrological erosion features, and comprehensive structural damage features, thereby achieving information aggregation of similar features and reducing dimensional complexity. The values ​​of the three comprehensive features—environmental, hydrological erosion, and structural damage—are all within the range of [0,1]. A larger value indicates a higher contribution to the latent erosion risk corresponding to that feature.

[0074] Step a14 involves identifying soil moisture data, soil temperature data, air temperature data, air humidity data, current rainfall, slope runoff velocity, characteristics of cavitation erosion development, and surface water turbidity to determine the original features whose cavitation risk sensitivity is greater than a preset threshold.

[0075] Specifically, the electronic device can calculate the Pearson correlation coefficient *r* between each original feature and the actual erosion risk label (0 = low risk, 1 = high risk). The absolute value of the correlation coefficient is the risk sensitivity of that feature. Based on engineering experience, the electronic device can set a sensitivity threshold of |r| ≥ 0.7 (i.e., the feature is highly correlated with erosion risk). Typically, the highly sensitive original features selected are: slope runoff velocity (S) and erosion cavity connectivity (C). Among them, slope runoff velocity (S) directly reflects the intensity of water erosion, with a sensitivity of |r| ≈ 0.85; erosion cavity connectivity (C) directly reflects the integrity of the soil structure, with a sensitivity of |r| ≈ 0.9.

[0076] The number of sensitive features selected by electronic devices is statistically analyzed to ensure that the number is ≤2 (to avoid feature redundancy), and the sensitivity of each feature meets the threshold requirement.

[0077] Step a15: Identify the current rainfall and determine the current scene label.

[0078] Specifically, the electronic device can compare the current rainfall with a preset rainfall threshold. If the current rainfall is greater than the preset rainfall threshold, the current scene is labeled as a rainy scene; if the current rainfall is less than or equal to the preset rainfall threshold, the current scene is labeled as a non-rainy scene.

[0079] Step a15 involves splicing and fusing the comprehensive environmental features, comprehensive hydrological erosion features, comprehensive structural damage features, each original feature, and the current scene label to generate the target feature.

[0080] Specifically, the electronic device splices and merges comprehensive environmental features, comprehensive hydrological erosion features, comprehensive structural damage features, various original features, and current scene labels to generate target features.

[0081] Step a2: Input the target features into the preset erosion cavity treatment model, and output the rainfall erosion index and soil erosion sensitivity index corresponding to the target dispersed soil slope.

[0082] Specifically, the pre-defined erosion cavity treatment model includes a shared feature extraction layer, a rainfall erosion index prediction layer, and a soil erosion sensitivity index prediction layer; step a2 above may include the following steps: Step a21: Input the target features into the shared feature extraction layer in the preset hole-penetration processing model; Specifically, the electronic device can input the target features into the shared feature extraction layer in the preset hole processing model.

[0083] Step a22: The shared feature extraction layer extracts features from the target features and outputs a shared feature vector.

[0084] Specifically, the shared feature extraction layer adopts a 3-layer FC architecture of "gradual dimensionality increase + scene adaptation", with each layer undertaking different feature extraction functions. The parameter configuration is shown in Table 1 below: Table 1. Parameter Configuration Table for Shared Feature Extraction Layer

[0085] Specifically, the first layer (FC) extracts basic features. A 6-dimensional combined input (5-dimensional features + 1-dimensional scene label) is passed to the first layer FC and calculated using the formula H1=ReLU(W1×X+b1+Wscene×L) (W1 is a 16×6 weight matrix, b1 is a 16-dimensional bias, Wscene is the weight vector of the scene label, and L is the scene label). In rainfall scenarios, the weights for "development of cavities" and "runoff" in Wscene are 0.3, while in non-rainfall scenarios, the weights for "crack development" and "hydrothermal" are also 0.3, ensuring that the initial extraction prioritizes capturing key scene features. Output: 16-dimensional basic feature vector, where the dimension value of "embedded hole development association feature" in the rainfall scenario (e.g., 0.7) is significantly higher than that in the non-rainfall scenario (e.g., 0.3).

[0086] Layer 2 FC: Cross-feature association fusion.

[0087] Input 16-dimensional basic features, calculate using H2=ReLU(W2×H1+b2) (W2 is a 32×16 weight matrix, b2 is a 32-dimensional bias), and randomly discard 20% of neurons using Dropout to avoid overfitting.

[0088] Output: 32-dimensional fused feature vector, of which cross-feature association dimension accounts for ≥40%, strengthening the common association required by REI / CDI.

[0089] Layer 3 FC: Common feature enhancement output.

[0090] Input 32-dimensional fused features and calculate them using H3=Tanh(W3×H2+b3) (W3 is a 64×32 weight matrix, b3 is a 64-dimensional bias). The Tanh function maps the output to [-1,1], enhancing the stability of the features.

[0091] Output: 64-dimensional shared feature vector. For example, in the rainfall scenario, the "penetrating pores-erosion correlation" dimension is 0.8 and the "soil integrity" dimension is 0.7; in the non-rainfall scenario, the "cracks-hydrothermal correlation" dimension is 0.7 and the "soil integrity" dimension is 0.6, realizing the effective extraction of common features across scenarios.

[0092] Step a23: Input the shared feature vector into the rainfall erosion index prediction layer and output the rainfall erosion index.

[0093] Specifically, step a23 above may include the following steps: Step a231: Obtain the slope runoff velocity and surface water turbidity.

[0094] Specifically, electronic devices can filter out slope runoff velocity and surface water turbidity from raw characteristics.

[0095] Step a232: Based on the relationship between slope runoff velocity and rainfall erosion index, and the relationship between surface water turbidity and rainfall erosion index, determine the first target weights corresponding to slope runoff velocity and surface water turbidity, respectively.

[0096] Specifically, the electronic device can retrieve corresponding samples (at least 500 sets of scenario data) from historical monitoring data for "historical rainfall erosion index - historical slope runoff velocity - historical surface water turbidity", and construct univariate linear regression models for historical slope runoff velocity and historical rainfall erosion index, and historical surface water turbidity and historical rainfall erosion index, respectively.

[0097] For example, a univariate linear regression model of historical slope runoff velocity and historical rainfall erosion index (REI): using historical slope runoff velocity S as the independent variable and historical rainfall erosion index REI as the dependent variable, the regression equation REI=a1×S is obtained by fitting the data using the least squares method. i +b1. Then, calculate the first coefficient of determination based on the following formula. (This reflects S's explanatory power for REI; for example, R12 = 0.82 indicates that S can explain 82% of the REI variation):

[0098] in, Historical slope runoff velocity ( ) and historical rainfall erosion index ( The first coefficient of determination of ) takes a value range of [0,1]. The closer the value is to 1, the stronger the linear correlation between the two; n is the number of historical samples involved in the calculation (≥50 groups are required to cover different rainfall conditions). The historical rainfall erosion index for the i-th sample; Let a1 be the mean of the REI for all samples, b1 be the first regression coefficient, and b1 be the first intercept.

[0099] Historical surface water turbidity ( ) and historical rainfall erosion index ( The univariate linear regression model: similar fitting =a2× +b2. Then, calculate the second coefficient of determination based on the following formula. (like ,show It can explain 75% of the REI variation:

[0100] in, Historical surface water turbidity ( ) and historical rainfall erosion index ( The second coefficient of determination, with a value range of [0,1], indicates a stronger linear correlation between the two as the value is closer to 1; n is the number of historical samples involved in the calculation (≥50 groups, covering different rainfall conditions). The actual rainfall erosion index for the i-th sample; α is the mean of the REI for all samples, a2 is the second regression coefficient, and b2 is the second intercept.

[0101] Then, the electronic device determines the first target weights corresponding to the slope runoff velocity and the surface water turbidity based on the first and second determination coefficients.

[0102] Step a233: Based on the slope runoff velocity, surface water turbidity, and the first target weight, rainfall erosion characteristics are generated.

[0103] Specifically, the electronic device multiplies the slope runoff velocity and surface water turbidity by the first target weight and then adds them together to generate rainfall erosion characteristics.

[0104] Step a234: The shared feature vector and the rainfall erosion feature are fused to generate the first fused feature.

[0105] Specifically, the electronic device will combine the shared feature vector and the rainfall erosion feature to generate the first fused feature.

[0106] Step a235: Input the first fused feature into the rainfall erosion index prediction layer and output the rainfall erosion index.

[0107] Specifically, the rainfall erosion index prediction layer is a two-layer fully connected neural network, which focuses on capturing the nonlinear correlation between fused features and REI. The architecture parameters are shown in Table 2 below: Table 2 Parameters of Rainfall Erosion Index Prediction Layer Architecture

[0108] This architecture reduces the risk of overfitting by using a small number of neurons, while the Sigmoid activation function ensures that the output range of REI is consistent with the original technical solution (0-1).

[0109] Specifically, the first fusion feature Vfusion1 is input into the rainfall erosion index prediction layer. After the first layer of FC calculation, the 32-dimensional intermediate feature HREI is obtained. Then, after the second layer of FC and the Sigmoid activation function, the initial REI prediction value REI is output. raw (such as REI) raw =0.65); In addition, electronic devices can set a correction factor k based on the rainfall intensity level of the current rainfall scenario (e.g., light rain: rainfall intensity <10mm / h, moderate rain: 10-25mm / h, heavy rain: >25mm / h). rain (Light rain: k) rain =0.9, moderate rain: k rain =1.0, Heavy rain: k rain =1.1), calculate the final rainfall erosion index REI final REI final =REI raw ×k rain Example: REI raw =0.65, currently heavy rain (k rain =1.1), then REI final =0.65 × 1.1 = 0.715; Corrected REI final It must be in the range of 0-1. If it exceeds this range (e.g., REIfinal=1.05), it will be truncated to 1.0.

[0110] Step a24: Input the shared feature vector into the soil erosion sensitivity index prediction layer and output the soil erosion sensitivity index.

[0111] Specifically, step a24 above may include the following steps: Step a241: Obtain the development characteristics of the burrowing cavities.

[0112] Specifically, electronic devices can screen out the development characteristics of undercut holes from the original features.

[0113] Step a242: Based on the relationship between the development characteristics of burrowing cavities and the soil burrowing sensitivity index, determine the second target weight corresponding to the development characteristics of burrowing cavities.

[0114] Specifically, the electronic device can retrieve corresponding samples (at least 500 sets of scene data) of "historical soil erosion sensitivity index - historical erosion hole development characteristics" from historical monitoring data, and construct a univariate linear regression model of historical erosion hole development characteristics and historical soil erosion sensitivity index.

[0115] An exemplary univariate linear regression model of historical pore development characteristics and soil erosion sensitivity index (CDI): Using historical pore development characteristics (G) as the independent variable and the historical soil erosion sensitivity index (CDI) as the dependent variable, the regression equation CDI is obtained by fitting the data using the least squares method. i =a3×G i +b3. Then, calculate the third coefficient of determination based on the following formula. (Reflecting G's explanatory power for CDI, such as...) =0.82, indicating that G i It can explain 82% of CDI i change):

[0116] in, Characteristics of historical erosion hole development ( ) and historical soil erosion sensitivity index ( The third coefficient of determination of ), with a value range of [0,1]. The closer the value is to 1, the stronger the linear correlation between the two; n is the number of historical samples involved in the calculation (≥50 groups, covering different rainfall conditions). is the historical soil erosion sensitivity index for the i-th sample; denoted as the mean of CDI for all samples, a3 is the third regression coefficient, and b3 is the third intercept.

[0117] Then, the electronic device determines the second target weights corresponding to the development characteristics of the target erosion holes and the development characteristics of the target fractures, respectively, based on the third determination coefficient.

[0118] Step a243: Based on the development characteristics of the erosion cavity and the weight of the second target, generate the crack development characteristics.

[0119] Specifically, the electronic device can multiply the development characteristics of the target erosion hole by the corresponding weight of the second target, and then add them together to generate the crack development characteristics.

[0120] Step a244: The shared feature vector is fused with the fracture development feature to generate a second fused feature.

[0121] Specifically, electronic devices can concatenate shared feature vectors with crack development features to generate a second fused feature.

[0122] Step a245: Input the second fusion feature into the soil erosion sensitivity index prediction layer and output the soil erosion sensitivity index.

[0123] Specifically, the soil erosion sensitivity index prediction layer is a two-layer fully connected neural network, which focuses on capturing the nonlinear correlation between fused features and the soil erosion sensitivity index (CDI). The architecture parameters are shown in Table 3 below: Table 3. Prediction Layer Structure Parameters for Soil Pitted Erosion Sensitivity Index

[0124] This architecture uses the Sigmoid activation function to limit the CDI output range to 0-0.9 (consistent with the reasonable range of CDI in the documentation), avoiding exceeding the predicted value in actual engineering.

[0125] Specifically, the second fusion feature Vfusion2 is input into the soil erosion sensitivity index prediction layer. After the first layer FC calculation, the 32-dimensional intermediate feature HCDI is obtained. Then, after the second layer FC and the Sigmoid activation function, the initial CDI prediction value CDI is output. raw (such as CDI) raw =0.72).

[0126] Then, the electronic device can combine the soil hydrothermal state of the current non-rainfall scenario (e.g., SMC < 15% is dry, 15%-25% is wet, > 25% is damp) to set a correction coefficient k. hydro (Drying: k) hydro =0.9, moist: k hydro =1.0, humid: k hydro =1.1), calculate the final soil erosion sensitivity index CDI final CDI final =CDI raw ×k hydro For example, CDI raw =0.72, currently in a wet state (k hydro =1.0), then CDI final =0.72 × 1.0 = 0.72; if it is a humid state (k hydro =1.1), then CDI final =0.72 × 1.1 = 0.792; Corrected CDI final It must be in the range of 0-0.9. If it exceeds this range (e.g., CDIfinal=0.95), it will be truncated to 0.9.

[0127] Step S204: Based on the rainfall erosion index and the soil erosion sensitivity index, determine the risk level of erosion hole development corresponding to the target dispersed soil slope.

[0128] Specifically, step S204 above may include the following steps: Step S2041: Determine the current climate scenario type based on the current rainfall.

[0129] Specifically, electronic devices can determine the current climate scenario type corresponding to a preset missing value based on the current rainfall amount corresponding to the missing value.

[0130] For example, a rainy scenario: the current rainfall R > 0.5 mm / h or the cumulative rainfall in the past 30 minutes > 2 mm (meeting either condition is sufficient for judgment, applicable to short-term showers, continuous rainfall, etc.); Non-raining scenario: Current rainfall R = 0 mm / h and cumulative rainfall in the past 30 minutes = 0 mm (double conditions ensure no rainfall interference).

[0131] Step S2042: Determine the target index weights corresponding to the rainfall erosion index and the soil erosion sensitivity index, respectively, based on the current climate scenario type.

[0132] Specifically, electronic devices can determine the target index weights corresponding to the rainfall erosion index and the soil erosion sensitivity index, respectively, based on the current climate scenario type.

[0133] For example, in the case of rainfall, the target index weight corresponding to the rainfall erosion index is 0.6, and the target index weight corresponding to the soil erosion sensitivity index is 0.4; in the case of non-rainfall scenarios, the target index weight corresponding to the rainfall erosion index is 0.3, and the target index weight corresponding to the soil erosion sensitivity index is 0.7.

[0134] Step S2043: Calculate the risk score of the development of cavities corresponding to the target dispersed soil slope based on the target index weights.

[0135] Specifically, electronic devices can multiply the rainfall erosion index and the soil erosion sensitivity index by the corresponding target index weights to calculate the risk score of erosion hole development corresponding to the target dispersed soil slope.

[0136] Step S2044: Determine the risk level of the development of the burrowing cavity corresponding to the target dispersed soil slope based on the risk score of the burrowing cavity development.

[0137] Specifically, electronic devices can determine the risk level of percolation cavity development corresponding to a target dispersed soil slope based on the percolation cavity development score. For example, Table 3 below shows the correspondence between percolation cavity development scores and percolation cavity development risk levels.

[0138] Table 3. Correspondence between the development score of burrowing cavities and the risk level of burrowing cavities.

[0139] The method for monitoring cavitation erosion in dispersed soil slopes provided in this application identifies surface images of soil slopes to determine the development characteristics of cavitation erosion ... The weights of sub-features are calculated using the entropy weighting method to objectively quantify the contribution of each sub-feature to the risk of latent erosion, avoiding the bias of subjective assignment. Sub-features within each category are weighted and summed to generate three types of comprehensive features, achieving information aggregation and dimensional compression of similar features and reducing redundant information. These comprehensive features reflect the overall impact of the "environment-hydrology-structure" dimensions on latent erosion risk, providing global information support for subsequent target feature generation. Sensitive original features are selected, and the key features with the strongest indicative power for latent erosion risk are accurately extracted. Scene labels are determined based on current rainfall, assigning working condition adaptation attributes to the target features, enabling subsequent models to specifically capture the patterns of latent erosion risk under different scenarios. The three types of comprehensive features, highly sensitive original features, and scene labels are concatenated to generate the final target features, taking into account both global and local key information, while incorporating working condition attributes to achieve comprehensive integration of multi-dimensional information. The target features are input into the shared feature extraction layer of the pre-defined latent erosion cavity development processing model. This provides unified common feature support for subsequent dual-index prediction, enabling cross-task feature reuse, avoiding resource waste caused by repeated extraction, and improving the overall computational efficiency of the model. A shared feature extraction layer extracts features from the target features and outputs a shared feature vector. It extracts core common information from the target features (such as soil foundation status and environmental correlation patterns) to provide a unified feature basis for predicting the rainfall erosion index and the soil erosion sensitivity index, ensuring the synergy of the two index predictions. It acquires slope runoff velocity and surface water turbidity to accurately locate key indicator data related to rainfall erosion, providing targeted input for subsequent specialized calculations of the rainfall erosion index and ensuring that the index calculation focuses on core influencing factors. Based on the relationship between slope runoff velocity and the rainfall erosion index, and the relationship between surface water turbidity and the rainfall erosion index, it determines the first target weights corresponding to slope runoff velocity and surface water turbidity, respectively.The influence of two key erosion indicators on the rainfall erosion index is quantified to make subsequent feature fusion more consistent with actual correlation patterns, thereby improving the representativeness and accuracy of rainfall erosion features. Rainfall erosion features are generated based on slope runoff velocity, surface water turbidity, and the first objective weight. Weighted fusion integrates the two dispersed erosion indicators into a single feature, condensing the core information of rainfall erosion and providing efficient and accurate input to the rainfall erosion index prediction layer, simplifying the model's calculation logic. Shared feature vectors and rainfall erosion features are fused to generate the first fused feature. Combining common features with rainfall-specific erosion features enriches the dimensions of input information, preserving basic environmental correlation information while strengthening rainfall-specific signals, thus improving the prediction accuracy of the rainfall erosion index. The first fused feature is input into the rainfall erosion index prediction layer, outputting the rainfall erosion index. The fused feature is transformed into a quantified erosion risk index, accurately characterizing the degree of slope erosion under rainfall scenarios and providing a scientific quantitative basis for assessing the development of pore formation. Based on the relationship between pore formation development features and the soil pore erosion sensitivity index, the second objective weight corresponding to the pore formation development features is determined. The impact of burrowing cavities on the soil burrowing sensitivity index is quantified to adapt to the actual impact patterns of structural damage, providing reasonable weighting support for the generation of crack development characteristics. Based on the burrowing cavity development characteristics and the second objective weight, crack development characteristics are generated. Structural damage-related indicators are weighted and integrated to form a unique feature that comprehensively reflects the degree of slope structural damage, providing accurate input for predicting the soil burrowing sensitivity index and highlighting the core impact of structural damage on burrowing cavity development. Shared feature vectors are fused with crack development characteristics to generate a second fused feature, integrating common features with structural damage-specific features, taking into account both basic environmental information and structural damage signals, improving the comprehensiveness and accuracy of the soil burrowing sensitivity index prediction. The second fused feature is input into the soil burrowing sensitivity index prediction layer, outputting the soil burrowing sensitivity index. Based on the current rainfall, the current climate scenario type is determined. The core working condition of the slope (rainfall / non-rainfall) is clarified, providing a working condition basis for subsequent indicator weight adjustments, ensuring that the burrowing cavity development evaluation adapts to the dominant influencing factors of different scenarios. Based on the current climate scenario, the weights of the target indicators corresponding to the rainfall erosion index and the soil erosion sensitivity index are determined. The evaluation weights of the two indices are dynamically adjusted according to the scenario type (rainfall erosion index is emphasized, non-rainfall erosion index is emphasized), making the assessment of pore development more closely aligned with actual working conditions and improving the relevance of the assessment. Based on the target indicator weights, the pore development score corresponding to the target dispersed soil slope is calculated. The two indices are integrated into a single score according to dynamic weights, achieving a quantitative characterization of pore development on slopes, making the pore development assessment results more intuitive and comparable, and providing accurate numerical basis for grading. Based on the pore development score, the pore development risk level corresponding to the target dispersed soil slope is determined.The quantitative scores are converted into clear risk levels for the development of cavities, and corresponding engineering countermeasures are matched to provide clear decision-making guidance for on-site maintenance, improve the timeliness and pertinence of slope disaster prevention and control, and reduce the risk of instability.

[0140] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for monitoring dispersed soil slope erosion cavities, characterized in that, The method includes: Acquire soil moisture data, soil temperature data, air temperature data, and air humidity data of the surrounding air corresponding to the target dispersed soil slope, as well as the current rainfall and slope runoff velocity during rainfall corresponding to the target dispersed soil slope. Acquire images of the soil slope surface corresponding to the target dispersed soil slope and the surface water at the interface between the target dispersed soil slope and the water body; Based on the soil moisture data, soil temperature data, air temperature data, air humidity data, current rainfall, slope runoff velocity, soil slope surface image, and surface water image, the rainfall erosion index and soil erosion sensitivity index corresponding to the target dispersed soil slope are determined. Based on the rainfall erosion index and the soil erosion sensitivity index, the risk level of erosion cavity development corresponding to the target dispersed soil slope is determined.

2. The method according to claim 1, characterized in that, The determination of the rainfall erosion index and soil erosion sensitivity index corresponding to the target dispersed soil slope based on the soil moisture data, soil temperature data, air temperature data, air humidity data, current rainfall, slope runoff velocity, soil slope surface image, and surface water image includes: The surface image of the soil slope is identified to determine the development characteristics of the cavities corresponding to the target dispersed soil slope; wherein, the development characteristics of the cavities include geometric morphological characteristics, distribution characteristics, and development degree characteristics; The surface water image is identified to determine the turbidity of the surface water at the interface between the target dispersed soil slope and the water body. Based on the soil moisture data, soil temperature data, air temperature data, air humidity data, current rainfall, slope runoff velocity, development characteristics of cavitation erosion, and surface water turbidity, the rainfall erosion index and soil erosion sensitivity index corresponding to the target dispersed soil slope are determined.

3. The method according to claim 2, characterized in that, The determination of the rainfall erosion index and soil erosion sensitivity index corresponding to the target dispersed soil slope based on the soil moisture data, soil temperature data, air temperature data, air humidity data, current rainfall, slope runoff velocity, development characteristics of erosion cavities, and surface water turbidity includes: Based on the soil moisture data, soil temperature data, air temperature data, air humidity data, current rainfall, slope runoff velocity, development characteristics of cavitation holes, and surface water turbidity, target features are generated. The target features are input into a preset erosion cavity treatment model, and the rainfall erosion index and the soil erosion sensitivity index corresponding to the target dispersed soil slope are output.

4. The method according to claim 3, characterized in that, The pre-defined erosion cavity treatment model includes a shared feature extraction layer, a rainfall erosion index prediction layer, and a soil erosion sensitivity index prediction layer; the step of inputting the target features into the pre-defined erosion cavity treatment model and outputting the rainfall erosion index and soil erosion sensitivity index corresponding to the target dispersed soil slope includes: The target features are input into the shared feature extraction layer in the preset hole-penetration processing model; The shared feature extraction layer extracts features from the target features and outputs a shared feature vector; The shared feature vector is input into the rainfall erosion index prediction layer, and the rainfall erosion index is output. The shared feature vector is input into the soil erosion sensitivity index prediction layer, and the soil erosion sensitivity index is output.

5. The method according to claim 4, characterized in that, The step of inputting the shared feature vector into the rainfall erosion index prediction layer and outputting the rainfall erosion index includes: The slope runoff velocity and the surface water turbidity are obtained; Based on the relationship between the slope runoff velocity and the rainfall erosion index, and the relationship between the surface water turbidity and the rainfall erosion index, the first target weights corresponding to the slope runoff velocity and the surface water turbidity are determined respectively. Rainfall erosion characteristics are generated based on the slope runoff velocity, the surface water turbidity, and the first target weight. The shared feature vector and the rainfall erosion feature are fused to generate a first fused feature; The first fusion feature is input into the rainfall erosion index prediction layer, and the rainfall erosion index is output.

6. The method according to claim 4, characterized in that, The step of inputting the shared feature vector into the soil erosion sensitivity index prediction layer and outputting the soil erosion sensitivity index includes: Obtain the development characteristics of the eroded holes; Based on the relationship between the development characteristics of the cavitation holes and the soil cavitation sensitivity index, a second target weight corresponding to the development characteristics of the cavitation holes is determined. Based on the development characteristics of the erosion cavity and the second target weight, a fracture development characteristic is generated; The shared feature vector is fused with the fracture development feature to generate a second fused feature; The second fusion feature is input into the soil erosion sensitivity index prediction layer, and the soil erosion sensitivity index is output.

7. The method according to claim 1, characterized in that, The generation of target features based on the soil moisture data, soil temperature data, air temperature data, air humidity data, current rainfall, slope runoff velocity, development characteristics of burrowing cavities, and surface water turbidity includes: The soil moisture data, soil temperature data, air temperature data, air humidity data, current rainfall, slope runoff velocity, surface erosion cavity development characteristics, and surface water turbidity are classified to obtain basic environmental characteristics, hydrological erosion characteristics, and structural damage characteristics. The structural damage characteristics include soil moisture data, soil temperature data, air temperature data, and air humidity data; the hydrological erosion characteristics include current rainfall, slope runoff velocity, and surface water turbidity; and the structural damage characteristics include surface erosion cavity development characteristics. For each sub-feature among the three types of features, calculate the weight value corresponding to each sub-feature; Based on the weight values ​​corresponding to each sub-feature, inter-group fusion processing is performed on each sub-feature in each category of features to generate comprehensive environmental features, comprehensive hydrological erosion features, and comprehensive structural damage features, respectively. The soil moisture data, soil temperature data, air temperature data, air humidity data, current rainfall, slope runoff velocity, development characteristics of cavitation holes, and turbidity of surface water are identified to determine the original features whose cavitation risk sensitivity is greater than a preset threshold. The current rainfall is identified to determine the current scene label; The target feature is generated by splicing and fusing the comprehensive environmental features, the comprehensive hydrological erosion features, the comprehensive structural damage features, each of the original features, and the current scene label.

8. The method according to claim 7, characterized in that, The calculation of the weight value corresponding to each of the sub-features includes: Calculate the feature proportion corresponding to each of the sub-features; Based on the aforementioned feature proportions, calculate the feature entropy value corresponding to each of the sub-features; Based on the feature entropy value, calculate the weight value corresponding to each of the sub-features.

9. The method according to claim 1, characterized in that, The determination of the risk level of cavitation development corresponding to the target dispersed soil slope based on the rainfall erosion index and the soil erosion sensitivity index includes: Based on the current rainfall, determine the current climate scenario type; Based on the current climate scenario type, determine the target index weights corresponding to the rainfall erosion index and the soil erosion sensitivity index, respectively; Based on the target index weights, calculate the risk score for the development of cavitation holes corresponding to the target dispersed soil slope; Based on the risk score of the development of the cavities, the risk level of the development of the cavities corresponding to the target dispersed soil slope is determined.

Images