Soil analysis test equipment

Abstract

The present invention relates to the technical field of soil analysis, and particularly to soil analysis test equipment, including: a water tank, soil disintegration test equipment, electronic equipment, and imaging equipment. The water tank contains water; the soil disintegration test equipment simulates the disintegration process of a soil sample under the relative movement state of the soil body and water, and records the mass change data of the soil sample during the disintegration process; the imaging equipment takes water body images corresponding to the water in the soil disintegration test equipment during the disintegration process of the soil sample; the electronic equipment identifies the mass change data to determine the soil water sensitivity corresponding to the soil sample; and identifies the water body images to determine the soil dispersion degree corresponding to the soil sample. The interference of human factors is reduced, the accuracy and reliability of the data are improved, which is conducive to the result comparison and communication between different research institutions or testers. Thus, the disintegration characteristics of the soil body are studied under the relative movement state of the soil body and water.

Description

Technical Field

[0001] The present invention relates to the technical field of soil analysis, and particularly to soil analysis test equipment. Background Art

[0002] The disintegration experiment is an important means to study the water stability of soil masses. In the fields of agricultural engineering, geotechnical engineering, etc., whether the soil mass meets the construction requirements can be judged through the disintegration situation of the soil mass. Therefore, the research on the disintegration characteristics of soil masses has important practical significance.

[0003] Existing disintegration experiments are all carried out under the state where the soil mass and water are relatively static. However, under natural conditions, the soil mass will be scoured by water flow. Therefore, it is also very important to study the disintegration characteristics of the soil mass under the state where the soil mass and water are relatively moving. However, the prior art cannot realize the disintegration experiment under the state where the soil mass and water are relatively moving.

[0004] Therefore, how to study the disintegration characteristics of the soil mass under the state where the soil mass and water are relatively moving has become an urgent problem to be solved. Summary of the Invention

[0005] In view of this, the present invention provides a soil analysis test equipment to solve the problem of how to study the disintegration characteristics of the soil mass under the state where the soil mass and water are relatively moving.

[0006] In a first aspect, the present invention provides a soil analysis test equipment, which includes: a water tank, a soil disintegration test equipment, an electronic device, and a camera device. The electronic device is communicatively connected to the camera device and the soil disintegration test equipment, wherein:

[0007] The water tank is used for containing water body;

[0008] The soil disintegration test equipment is used for simulating the disintegration process of the soil sample under the state of relative movement between the soil sample and the water body, and recording the mass change data of the soil sample during the disintegration process;

[0009] The camera device is used for taking the water body image corresponding to the water body in the soil disintegration test equipment during the disintegration process of the soil sample;

[0010] The electronic device is used for identifying the mass change data to determine the soil water sensitivity corresponding to the soil sample; and identifying the water body image to determine the soil dispersion degree corresponding to the soil sample.

[0011] The soil analysis test equipment provided by this application includes a water tank for holding water. A soil disintegration test equipment is used to simulate the disintegration process of a soil sample under the relative movement state of the soil body and water, and record the mass change data of the soil sample during the disintegration process. Being able to simulate the disintegration process of the soil sample under the relative movement state of the soil body and water helps to reproduce as realistically as possible the actual situation of the soil body encountering water in the natural environment under laboratory conditions. Through this simulation, the stability and related characteristics of the soil body under the action of water can be studied more accurately. A camera device is used to take water body images corresponding to the water body in the soil disintegration test equipment during the disintegration process of the soil sample. An electronic device is used to identify the mass change data to determine the soil water sensitivity corresponding to the soil sample; and to identify the water body images to determine the soil dispersion degree corresponding to the soil sample. Thus, the soil characteristics can be analyzed from different angles. By determining the soil water sensitivity through the mass change data and the soil dispersion degree from the water body images, this multi-dimensional analysis method can comprehensively understand the properties of the soil body, providing richer and more accurate information for soil research and related engineering applications. It reduces the interference of human factors and improves the accuracy and reliability of the data. Moreover, the degree of automation of the equipment is relatively high, and it can collect and analyze data according to preset programs and algorithms, ensuring the repeatability and consistency of the test results, which is conducive to the comparison and communication of results between different research institutions or test personnel. Thus, the disintegration characteristics of the soil body are studied under the relative movement state of the soil body and water.

[0012] In an optional implementation manner, the water tank is divided into two water body areas, and the soil disintegration test equipment includes: two mesh hanging baskets, which respectively extend into the two corresponding water body areas of the water tank; one of the mesh hanging baskets is connected to a lifting rod, the lifting rod can drive the mesh hanging basket to move up and down, the lifting rod is connected to a motor, and the mesh hanging basket is connected to a first electronic scale; the other mesh hanging basket is connected to a second electronic scale, and the motor, the first electronic scale, and the second electronic scale are all communicatively connected to the electronic device, where:

[0013] Each mesh hanging basket is used to hold the soil sample;

[0014] The electronic device is used to control the motor to drive the lifting rod to move up and down at a preset speed, so that the soil sample and the water body have a preset relative movement speed, and a disintegration test can be carried out under the condition of relative movement between the soil sample and the water body;

[0015] The first electronic scale is used to measure the first initial mass of the mesh hanging basket corresponding to the first electronic scale when it is immersed in the water body, and the first target mass corresponding to the mesh hanging basket after a preset time period;

[0016] The second electronic scale is used to measure the second initial mass of the mesh hanging basket corresponding to the second electronic scale when it is immersed in the water body, and the second target mass corresponding to the mesh hanging basket after a preset time period;

[0017] An electronic device is used to calculate the soil water sensitivity of a soil sample corresponding to a preset duration at a preset relative movement speed based on a first initial mass, a first target mass, a second initial mass, and a second target mass.

[0018] The soil analysis test equipment provided in this application includes grid hanging baskets for holding soil samples. An electronic device is used to control a motor to drive a lifting rod to move up and down at a preset speed, so that the soil sample and water have a preset relative movement speed, and a disintegration test can be carried out under the condition of relative movement between the soil sample and water. This design simulates the relative movement of soil in the natural environment due to factors such as water flow scouring and groundwater fluctuation. In the actual soil around rivers and lakes, or the foundation soil affected by groundwater flow, it will be in a similar dynamic water environment. By simulating the disintegration test under such relative movement conditions, more realistic soil water sensitivity data can be obtained, making the research results more practically guiding, and providing reliable soil property references for projects such as water conservancy projects and road and bridge construction facing complex hydrogeological conditions. In addition, by adjusting the preset speed, different intensities of water flow action or relative movement scenarios between the soil and water can be simulated. A first electronic scale is used to measure the first initial mass of the grid hanging basket corresponding to the first electronic scale when it is immersed in water, and the first target mass of the grid hanging basket after a preset duration. A second electronic scale is used to measure the second initial mass of the grid hanging basket corresponding to the second electronic scale when it is immersed in water, and the second target mass of the grid hanging basket after a preset duration. An electronic device is used to calculate the soil water sensitivity of the soil sample corresponding to a preset duration at a preset relative movement speed based on the first initial mass, the first target mass, the second initial mass, and the second target mass, ensuring the accuracy of the obtained soil water sensitivity.

[0019] In an optional implementation manner, the electronic device is used to calculate a first disintegration degree by dividing the difference between the first initial mass and the first target mass by the first initial mass;

[0020] calculate a second disintegration degree by dividing the difference between the second initial mass and the second target mass by the second initial mass;

[0021] calculate the soil water sensitivity corresponding to the soil sample at the relative movement speed by dividing the first disintegration degree by the second disintegration degree.

[0022] The soil analysis test equipment provided by this application calculates the first disintegration degree by dividing the difference between the first initial mass and the first target mass by the first initial mass, so as to obtain the disintegration degree of the soil sample under the condition of relative movement. The second disintegration degree is calculated by dividing the difference between the second initial mass and the second target mass by the second initial mass, so as to obtain the disintegration degree of the soil sample in a static state. By dividing the first disintegration degree by the second disintegration degree, the soil water sensitivity corresponding to the soil sample under the relative movement speed is calculated, ensuring the accuracy of the soil water sensitivity corresponding to the soil sample at the preset relative movement speed and within the preset time duration.

[0023] In an alternative embodiment, each soil disintegration test device includes: at least one water jet and at least one wave maker, where:

[0024] The water jet is used to adjust the jet angle and flow rate under the control of the electronic device to generate water flows with different speeds, different directions, and different intensities;

[0025] The wave maker is used to generate waves with various wavelengths and wave heights under the control of the electronic device; to further simulate different water body intensities and water pressures;

[0026] The electronic device is used to calculate the soil water sensitivity corresponding to the soil sample under each relative movement speed, each water body intensity, and each water pressure.

[0027] The soil analysis test equipment provided by this application, the water jet is used to adjust the jet angle and flow rate under the control of the electronic device to generate water flows with different speeds, different directions, and different intensities; the wave maker is used to generate waves with various wavelengths and wave heights under the control of the electronic device; to further simulate different water body intensities and water pressures; the electronic device is used to calculate the soil water sensitivity corresponding to the soil sample under each relative movement speed, each water body intensity, and each water pressure, so as to be able to simulate diverse water flow conditions in the natural environment, such as the rapid flow and slow flow in rivers, and the change of the seepage direction of groundwater, etc., and simulate the disintegration process of the soil sample under waves with various wavelengths and wave heights. Furthermore, the soil water sensitivity corresponding to the soil sample under each relative movement speed, each water body intensity, and each water pressure is calculated.

[0028] In an alternative embodiment, the soil disintegration test device further includes a pH monitoring probe and an EC monitoring probe. Both the pH monitoring probe and the EC monitoring probe are communicatively connected to the electronic device, where:

[0029] The pH monitoring probe is used to monitor the pH value of the water body to obtain the change data of the water body pH value during the disintegration process of the soil sample;

[0030] An EC monitoring probe is used to monitor the water conductivity of water bodies, obtaining the variation data of the water conductivity of the soil sample during the disintegration process.

[0031] A camera device is used to capture at least one water image of the water area corresponding to the grid hanging basket connected to the second electronic scale during the disintegration process of the soil sample.

[0032] An electronic device is used to input the water image, the variation data of the water pH value, and the variation data of the water conductivity into a preset soil dispersion model, and output the soil dispersion corresponding to the soil sample.

[0033] The soil analysis test equipment provided by this application, a pH monitoring probe, is used to monitor the pH value of water bodies, obtaining the variation data of the water pH value of the soil sample during the disintegration process, thereby reflecting the dissolution, reaction, etc. of acid-base substances during the soil disintegration process. For example, if the soil contains alkaline minerals, the dissolution of alkaline substances during disintegration will increase the water pH value, and this process can be traced through the pH value variation data. The EC monitoring probe is used to monitor the water conductivity of water bodies, obtaining the variation data of the water conductivity of the soil sample during the disintegration process; it can intuitively reflect the change in the amount and type of ions released in the soil because the change in ion concentration directly affects the conductivity. These two chemical characteristic data complement the physical form information provided by the water image, providing a rich data basis for determining the soil dispersion from different dimensions and making the analysis more comprehensive and accurate. The camera device is used to capture at least one water image of the water area corresponding to the grid hanging basket connected to the second electronic scale during the disintegration process of the soil sample. The electronic device is used to input the water image, the variation data of the water pH value, and the variation data of the water conductivity into a preset soil dispersion model, and output the soil dispersion corresponding to the soil sample. Different types of data contain different information about soil dispersion, and after integration, they can make up for the limitations of single data. For example, the water image can show the macroscopic distribution and turbidity of soil particles, while the pH value and conductivity variation data can reveal the influence of the release and reaction of internal chemical components in the soil on the dispersion process. By integrating these information through the model, more in-depth and accurate soil dispersion characteristics can be mined, improving the accuracy of determining the soil dispersion.

[0034] In an alternative embodiment, the preset soil dispersion model includes a feature extraction network and a soil dispersion discriminator, where:

[0035] An electronic device is used to input the water image, the variation data of the water pH value, and the variation data of the water conductivity into the feature extraction network in the preset soil dispersion model.

[0036] The feature extraction network extracts the RGB features and image features corresponding to the water body image; the image features include at least one of the water body turbidity feature and the water body texture feature;

[0037] The feature extraction network captures the front and back features of the water body pH value change data and the water body conductivity change data from the forward and reverse directions, and obtains the water body pH value forward feature, the water body pH value backward feature, the water body conductivity forward feature, and the water body conductivity backward feature;

[0038] The feature extraction network determines the key time series in the water body pH value change data and the water body conductivity change data based on the attention mechanism, and obtains the water body pH value key time series and the water body conductivity key time series;

[0039] Identify the water body pH value key time series and the water body conductivity key time series to determine the water body pH value key features and the water body conductivity key features;

[0040] Perform feature fusion on the RGB features, image features, water body pH value forward features, water body pH value backward features, water body conductivity forward features, water body conductivity backward features, water body pH value key features, and water body conductivity key features to generate initial fusion features;

[0041] Input the initial fusion features into the soil dispersion discriminator, and the soil dispersion discriminator outputs the soil dispersion corresponding to the soil sample.

[0042] The soil analysis test equipment provided by this application is an electronic device used to input water body images, water body pH value change data, and water body conductivity change data into the feature extraction network of a preset soil dispersion model. This multi-source data input method can obtain soil dispersion-related information from different perspectives. The water body image intuitively reflects the physical distribution state of the soil in water, while the water body pH value and conductivity change data reveal the chemical characteristics during the interaction between the soil and water. The comprehensive utilization of multi-source data avoids the limitations of a single data source and provides a more comprehensive and rich basis for accurately judging the soil dispersion degree in the subsequent process. The feature extraction network extracts the RGB features and image features corresponding to the water body image. By extracting the RGB features, these color changes can be captured, providing a visually intuitive basis for judging the soil dispersion degree. Turbidity is closely related to the dispersion degree of the soil sample in water. By extracting the water body turbidity feature, the dispersion situation of the soil can be directly quantified. The texture feature describes the microscopic structure and distribution pattern of the water body surface. Different soil dispersion states will make the water body surface present different textures. Extracting the water body texture feature helps to discover some details of soil dispersion that are difficult to detect by the naked eye and improves the accuracy of judging the soil dispersion degree. The feature extraction network captures the front and back features of the water body pH value change data and the water body conductivity change data from both the forward and reverse directions, obtaining the water body pH value forward feature, water body pH value backward feature, water body conductivity forward feature, and water body conductivity backward feature, so as to comprehensively understand the changing trends of the water body pH value and water body conductivity over time. The feature extraction network determines the key time series in the water body pH value change data and the water body conductivity change data based on the attention mechanism, obtaining the water body pH value key time series and the water body conductivity key time series, so as to focus on the time periods that have a greater impact on the soil sample dispersion degree. Identify the water body pH value key time series and the water body conductivity key time series to determine the water body pH value key feature and the water body conductivity key feature; further refine the information closely related to the soil dispersion degree. By accurately identifying the key features, it can provide more valuable input for subsequent feature fusion and discrimination. Perform feature fusion on the RGB features, image features, water body pH value forward feature, water body pH value backward feature, water body conductivity forward feature, water body conductivity backward feature, water body pH value key feature, and water body conductivity key feature to generate initial fusion features, so as to integrate the advantages of different types of data and give full play to the synergistic effect of multi-source data. There may be an interrelated and complementary relationship between different features. Through fusion, more in-depth and comprehensive information can be mined. Input the initial fusion features into the soil dispersion discriminator, and the soil dispersion discriminator outputs the soil dispersion degree corresponding to the soil sample. The soil dispersion discriminator can accurately output the soil dispersion degree by comprehensively analyzing and judging this information.

[0043] In an alternative embodiment, the feature extraction network includes dilated convolutions with multiple different receptive fields; wherein:

[0044] The feature extraction network identifies the water body image, determines the importance of features at different scales, and assigns weight information to the dilated convolutions of each receptive field;

[0045] Identify the water body image to determine the complexity corresponding to different regions in the water body image;

[0046] According to the complexity corresponding to different regions in the water body image, determine the sampling frequency corresponding to the dilated convolutions of each different receptive field;

[0047] Based on each dilated convolution, perform feature extraction on the water body image to determine the image features corresponding to the water body image.

[0048] For the soil analysis test equipment provided in this application, the feature extraction network identifies the water body image, determines the importance of features at different scales, and assigns weight information to the dilated convolutions of each receptive field, so that the soil dispersion discriminator can automatically adjust the attention to features at different scales according to the characteristics of the input water body image. Identify the water body image to determine the complexity corresponding to different regions in the water body image, so that potential features in the water body image can be highlighted through complexity analysis. According to the complexity corresponding to different regions in the water body image, determine the sampling frequency corresponding to the dilated convolutions of each different receptive field; it is possible to achieve optimized allocation of computing resources. For regions with high complexity, increase the sampling frequency so that the dilated convolution can collect the feature information of this region more carefully and fully capture details such as complex textures and particle distributions; for regions with low complexity, reduce the sampling frequency, and on the premise of ensuring the acquisition of basic features, reduce unnecessary computational volume. In this way, it can not only comprehensively extract the key information of the image, but also avoid waste of computing resources caused by over-sampling of simple regions, and improve the computational efficiency of the feature extraction process. Based on each dilated convolution, perform feature extraction on the water body image to determine the image features corresponding to the water body image. At different scales, the dilated convolution can capture multi-scale features from the macroscopic distribution of water body turbidity regions to the microscopic soil particle textures, etc., and make adaptive adjustments according to the image region complexity and scale feature importance, so as to obtain richer and more accurate image features, providing more comprehensive information support for judging soil dispersion. In addition, through the dilated convolution feature extraction process after the above series of optimization operations, the obtained image features have stronger representativeness and discriminability.

[0049] In an alternative embodiment, an electronic device is configured to construct a feature map with RGB features, image features, forward water body pH value features, backward water body pH value features, forward water body conductivity features, backward water body conductivity features, key water body pH value features, and key water body conductivity features as nodes, and the correlation between pairwise features as the weight of the edge.

[0050] Use a graph neural network to process the feature map; in each layer of message passing, each node updates its own feature representation according to the features of its neighbor nodes and the weight of the edge.

[0051] After message passing and feature updating for a preset number of rounds, the features of all nodes in the feature map are weighted and summed to obtain an initial fusion feature.

[0052] The soil analysis test equipment provided in this application uses RGB features, image features, forward water body pH value features, backward water body pH value features, forward water body conductivity features, backward water body conductivity features, key water body pH value features, and key water body conductivity features as nodes, so as to comprehensively present the relationships between different types of features. The correlation between pairwise features is used as the weight of the edge to construct a feature map, which can uncover some potential feature associations that are not easily observable directly. There may be complex non-linear relationships between some features, and these relationships can be discovered through correlation analysis and reflected in the feature map. Use a graph neural network to process the feature map; in each layer of message passing, each node updates its own feature representation according to the features of its neighbor nodes and the weight of the edge. This method can effectively capture the complex interactions between features. In the study of soil dispersion, the mutual influence between different features is not a simple linear relationship, but there is a complex coupling effect. The graph neural network can simulate this complex interaction process through the message passing mechanism. For example, the change in the water body pH value may affect the surface charge of soil particles, and then affect their dispersion state in the water body, and this influence will be transmitted to other relevant feature nodes through the edges in the feature map, enabling the nodes to update their own feature representations according to this information, so as to more accurately reflect the actual situation of soil dispersion. After message passing and feature updating for a preset number of rounds, the features of all nodes in the feature map are weighted and summed to obtain an initial fusion feature. This weighted summation method can comprehensively consider the contribution of each feature to soil dispersion. In the feature map, the weight of the edge already reflects the correlation between features. Through weighted summation, features with stronger correlation will occupy a larger proportion in the fusion feature, while features with weaker correlation will be relatively smaller. This can ensure that the fusion feature can more accurately reflect the comprehensive influence of different features on soil dispersion, avoiding problems such as information loss or inaccuracy that may be caused by simple averaging or equal-weight fusion.

[0053] In an alternative embodiment, an electronic device is configured to input an initial fusion feature into a soil dispersion discriminator; at the initial layer of the soil dispersion discriminator, different types of features are extracted through different convolutional layer branches, and an attention mechanism is introduced to fuse the features extracted by each convolutional layer branch to generate a target fusion feature;

[0054] The soil dispersion discriminator makes a preliminary evaluation of the target fusion feature. By calculating the correlation score between each sub-feature in the fusion feature and the known soil dispersion label, key features that contribute more than a preset threshold to the discrimination of soil dispersion are screened out;

[0055] Feature enhancement technology is adopted to enhance each key feature to obtain enhanced features;

[0056] Based on the enhanced features, the soil dispersion corresponding to the soil sample is output.

[0057] The soil analysis test equipment provided in this application, an electronic device, is configured to input an initial fusion feature into a soil dispersion discriminator; at the initial layer of the soil dispersion discriminator, different types of features are extracted through different convolutional layer branches, which can deeply explore the initial fusion feature from multiple dimensions. For example, some branches focus on extracting the spatial structure information in the image features, while some branches are good at capturing the change trend information in the features related to pH value and conductivity. This multi-dimensional feature extraction method fully utilizes various information in the initial fusion feature, avoids the omission of some features by a single convolutional layer, and provides a richer feature basis for accurately judging the soil dispersion subsequently. And an attention mechanism is introduced to fuse the features extracted by each convolutional layer branch to generate a target fusion feature. The attention mechanism can automatically learn the importance of different features and fuse the features according to these importance levels. For example, when judging the soil dispersion, if the turbidity feature in the image and the key change feature of the pH value have a greater impact on the result, the attention mechanism will assign higher weights to these two features, making the key information more prominent in the fused features. This efficient feature fusion method improves the quality and discrimination ability of the features, and helps the discriminator to more accurately grasp the feature combination related to the soil dispersion. The soil dispersion discriminator makes a preliminary evaluation of the target fusion feature. By calculating the correlation score between each sub-feature in the fusion feature and the known soil dispersion label, key features that contribute more than a preset threshold to the discrimination of soil dispersion are screened out, so as to further highlight the information related to the soil dispersion in the key features. Feature enhancement technology is adopted to enhance each key feature to obtain enhanced features, which can make full use of the high-quality feature information after screening and enhancement, and greatly improve the accuracy of judging the soil dispersion. Based on the enhanced features, the soil dispersion corresponding to the soil sample is output, ensuring the accuracy of the output soil dispersion corresponding to the soil sample.

[0058] In an alternative embodiment, an electronic device is configured to obtain soil physical and chemical property data and geographic information data corresponding to each soil sample; the soil physical and chemical property data includes at least one of soil particle characteristics, three-phase ratio indexes of soil, physical and chemical state indexes of soil, and soil permeability indexes; the geographic information data includes at least one of longitude and latitude, altitude, and geological structure;

[0059] Generate a soil water sensitivity training data set based on the soil physical and chemical property data, geographic information data, and soil water sensitivity corresponding to each soil sample; the soil water sensitivity in the soil water sensitivity training data set is used as label data; train a soil water sensitivity prediction model based on the soil water sensitivity training data set;

[0060] Generate a soil dispersion training data set based on the soil physical and chemical property data, geographic information data, and soil water sensitivity corresponding to each soil sample; the soil dispersion in the soil dispersion training data set is used as label data; train a soil dispersion prediction model based on the soil dispersion training data set.

[0061] The soil analysis test equipment provided by this application obtains the soil physical and chemical property data and geographic information data corresponding to each soil sample, and generates a soil water sensitivity training data set based on the soil physical and chemical property data, geographic information data, and soil water sensitivity corresponding to each soil sample; the soil water sensitivity in the soil water sensitivity training data set is used as label data; train a soil water sensitivity prediction model based on the soil water sensitivity training data set, so that the soil water sensitivity can be predicted by using various characteristics of the soil. Generate a soil dispersion training data set based on the soil physical and chemical property data, geographic information data, and soil water sensitivity corresponding to each soil sample; the soil dispersion in the soil dispersion training data set is used as label data; train a soil dispersion prediction model based on the soil dispersion training data set. Thus, the soil dispersion can be predicted by using various characteristics of the soil. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0063] Figure 1 FIG. 1 is a schematic structural diagram of a first soil analysis test equipment according to an embodiment of the present invention;

[0064] Figure 2It is a schematic structural diagram of a second soil analysis test device according to an embodiment of the present invention;

[0065] Figure 3 It is a schematic structural diagram of a third soil analysis test device according to an embodiment of the present invention;

[0066] Figure 4 It is a schematic structural diagram of a fourth soil analysis test device according to an embodiment of the present invention. Detailed implementation manners

[0067] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0068] The serial numbers assigned to the components herein, such as "first", "second", etc., are only used to distinguish the described objects and do not have any sequential or technical meanings. In the description of the present application, it should be understood that the orientation or positional relationships indicated by the terms "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", etc. are based on the orientation or positional relationships shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as a limitation to the present application. In the present application, unless otherwise clearly specified and limited, the first feature being "on" or "under" the second feature may be that the first and second features are in direct contact, or the first and second features are in indirect contact through an intermediate medium. Moreover, the first feature being "above", "over" and "on" the second feature may be that the first feature is directly above or obliquely above the second feature, or simply means that the first feature has a higher horizontal height than the second feature. The first feature being "under", "beneath" and "under" the second feature may be that the first feature is directly below or obliquely below the second feature, or simply means that the first feature has a lower horizontal height than the second feature.

[0069] An embodiment of the present application provides a soil analysis test device, as Figure 1 shown. The soil analysis test device includes: a water tank 1, a soil disintegration test device 2, an electronic device 3, and a camera device 4. The electronic device 3 is communicatively connected to the camera device 4 and the soil disintegration test device 2, where:

[0070] The water tank 1 is used for holding water;

[0071] The soil disintegration test device 2 is used to simulate the disintegration process of the soil sample under the relative movement state of the soil body sample and the water body, and record the mass change data of the soil body sample during the disintegration process;

[0072] The camera device 4 is used to take the water body image corresponding to the water body in the soil disintegration test device 2 during the disintegration process of the soil body sample;

[0073] The electronic device 3 is used to identify the mass change data to determine the soil water sensitivity corresponding to the soil body sample; and identify the water body image to determine the soil dispersion degree corresponding to the soil body sample.

[0074] Specifically, the water tank 1 can hold the water body. The soil disintegration test device 2 simulates the disintegration process of the soil body sample under the relative movement state of the soil body sample and the water body, and records the mass change data of the soil body sample during the disintegration process, and then transmits the mass change data to the electronic device 3 based on the communication connection with the electronic device 3.

[0075] During the disintegration process of the soil body sample, the camera device 4 takes the water body image corresponding to the water body in the soil disintegration test device 2, and then transmits the water body image to the electronic device 3 based on the communication connection with the electronic device 3. The electronic device 3 identifies the mass change data to determine the soil water sensitivity corresponding to the soil body sample; and identifies the water body image to determine the soil dispersion degree corresponding to the soil body sample.

[0076] The soil analysis test equipment provided by the embodiments of the present application includes a water tank 1 for holding water. A soil disintegration test device 2 is used to simulate the disintegration process of a soil sample under the relative movement state of the soil body and water, and record the mass change data of the soil sample during the disintegration process. It can simulate the disintegration process of the soil sample under the relative movement state of the soil body and water, which helps to reproduce as realistically as possible the actual situation of the soil body encountering water in the natural environment under laboratory conditions. Through this simulation, the stability and related characteristics of the soil body under the action of water can be studied more accurately. A camera device 4 is used to capture the water body image corresponding to the water body in the soil disintegration test device 2 during the disintegration process of the soil sample. An electronic device 3 is used to identify the mass change data to determine the soil water sensitivity corresponding to the soil sample; and identify the water body image to determine the soil dispersion degree corresponding to the soil sample. Thus, the soil characteristics can be analyzed from different angles. By determining the soil water sensitivity through the mass change data and the soil dispersion degree from the water body image, this multi-dimensional analysis method can comprehensively understand the properties of the soil body, providing richer and more accurate information for soil research and related engineering applications. It reduces the interference of human factors and improves the accuracy and reliability of the data. Moreover, the degree of automation of the equipment is relatively high, and it can collect and analyze data according to preset programs and algorithms, ensuring the repeatability and consistency of the test results, which is conducive to the result comparison and communication between different research institutions or test personnel. Thus, the disintegration characteristics of the soil body are studied under the relative movement state of the soil body and water.

[0077] In an alternative embodiment of the present application, as Figure 2 shown, the water tank 1 is divided into two water body areas. The soil disintegration test device 2 includes: two grid hanging baskets 21, and the two grid hanging baskets 21 respectively extend into the two corresponding water body areas of the water tank 1; one of the grid hanging baskets 21 is connected to a lifting rod 22, and the lifting rod 22 can drive the grid hanging basket 21 to move up and down. The lifting rod 22 is connected to a motor 23, and the grid hanging basket 21 is connected to a first electronic scale 24; the other grid hanging basket 21 is connected to a second electronic scale 25. The motor 23, the first electronic scale 24, and the second electronic scale 25 are all communicatively connected to the electronic device 3, where:

[0078] Each grid hanging basket 21 is used to hold the soil sample;

[0079] The electronic device 3 is used to control the motor 23 to drive the lifting rod 22 to move up and down at a preset speed, so that the soil sample and the water body have a preset relative movement speed, and a disintegration test can be carried out under the condition of relative movement between the soil sample and the water body;

[0080] The first electronic scale 24 is used to measure the first initial mass of the grid hanging basket 21 corresponding to the first electronic scale 24 when it is immersed in the water body, and the first target mass corresponding to the grid hanging basket 21 after a preset time;

[0081] A second electronic scale 25, configured to measure a second initial mass of the grid hanging basket 21 corresponding to the second electronic scale 25 when immersed in water, and a second target mass of the grid hanging basket 21 after a preset time period;

[0082] An electronic device 3, configured to calculate the soil water sensitivity of the soil sample at a preset relative motion speed corresponding to a preset time period based on the first initial mass, the first target mass, the second initial mass, and the second target mass.

[0083] In an alternative embodiment of the present application, the electronic device 3 is configured to calculate a first disintegration degree by dividing the difference between the first initial mass and the first target mass by the first initial mass;

[0084] calculate a second disintegration degree by dividing the difference between the second initial mass and the second target mass by the second initial mass;

[0085] calculate the soil water sensitivity of the soil sample at the relative motion speed by dividing the first disintegration degree by the second disintegration degree.

[0086] Specifically, the water tank 1 is partitioned into two water regions, and one grid hanging basket 21 extends into each water region. The grid hanging basket 21 has a length, width, and height of 10 cm, and the grids on the front, back, left, right, and lower sides are all 5 mm. The material is stainless steel, and there is no grid on the upper side. One of the grid hanging baskets 21 is connected to the lifting rod 22. The specific connection method may be that the steel fork at the front end of the lifting rod 22 is located between the grid hanging basket 21 and the upper vertical rod of the grid hanging basket 21. The lifting rod 22 can drive the grid hanging basket 21 to move up and down, and the lifting rod 22 is connected to the motor 23. In addition, the upper hook of the upper vertical rod of the grid hanging basket 21 is connected to the lower hook of the first electronic scale 24. The upper hook of the upper vertical rod of the other grid hanging basket 21 is connected to the lower hook of the second electronic scale 25. The first electronic scale 24 and the second electronic scale 25 are installed on the cross bar. In addition, the motor 23, the first electronic scale 24, and the second electronic scale 25 are all communicatively connected to the electronic device 3.

[0087] During the experiment, soil samples of the same mass and the same volume are placed in the two grid hanging baskets 21. Then, the motor 23 is started, and the motor 23 controls the lifting rod 22 to perform vertical up and down movements.

[0088] Exemplarily, the lifting rod 22 is initially at a position 10 cm above the upper side of the water tank 1. The motor 23 can control the lifting rod 22 to move within a vertical range of 6 cm up and down, and the moving speed is 1 cm / s. That is, at the beginning of the test, the lifting rod 22 moves downward from a position 10 cm above the upper side of the water tank 1 at a speed of 1 cm / s under the control of the motor 23. At this time, the upper hook of the hanging basket hook is hooked to the hook of the first electronic scale 24, and the grid hanging basket 21 will not move downward with the lifting rod 22. At this time, the reading can be taken through the first electronic scale 24. After the lifting rod 22 moves downward by 3 cm, it will move upward in the reverse direction. After moving upward by 3 cm at a speed of 1 cm / s, the fork at the front end of the lifting rod 22 will fork the cross bar on the hanging basket hook, so that the grid hanging basket 21 moves upward with the lifting rod 22. After moving 3 cm, it moves downward again at a speed of 1 cm / s. After moving 3 cm, the upper hook of the hanging basket hook is hooked to the hook of the first electronic scale 24. When the lifting rod 22 continues to move downward, the grid hanging basket 21 will not move downward with the lifting rod 22. At this time, the reading can be taken again through the first electronic scale 24, and so on until the disintegration test ends. And the grid hanging basket 21 not connected to the lifting rod 22 extends into the water body until the disintegration test ends.

[0089] The first electronic scale 24 can measure the first initial mass of the grid hanging basket 21 corresponding to the first electronic scale 24 when it is immersed in the water body, and the first target mass of the grid hanging basket 21 after a preset time.

[0090] The second electronic scale 25 can measure the second initial mass of the grid hanging basket 21 corresponding to the second electronic scale 25 when it is immersed in the water body, and the second target mass of the grid hanging basket 21 after a preset time.

[0091] Then, based on the communication connection between the first electronic scale 24 and the second electronic scale 25, the electronic device 3 obtains the first initial mass, the first target mass, the second initial mass, and the second target mass.

[0092] The electronic device 3 calculates the first disintegration degree by dividing the difference between the first initial mass and the first target mass by the first initial mass. The electronic device 3 calculates the second disintegration degree by dividing the difference between the second initial mass and the second target mass by the second initial mass. Then, the electronic device 3 calculates the soil water sensitivity corresponding to the soil sample at the relative movement speed by dividing the first disintegration degree by the second disintegration degree.

[0093] Exemplarily, after the electronic device 3 reads the data of the second electronic scale 25, it will compare with the reading of the second electronic scale 25 after the net plate is immersed in water when no soil body is placed. When the data of the second electronic scale 25 read by the electronic device 3 is the same as the reading of the second electronic scale 25 after the net plate is immersed in water when no soil body is placed, it indicates that the soil body is completely disintegrated in this group of disintegration experiments. When the reading of the second electronic scale 25 in each group of disintegration experiments is the same as the reading of the electronic scale after the net plate is immersed in water when no soil body is placed, the electronic device 3 automatically stops the motor 23 from working, and the disintegration test ends.

[0094] Exemplarily, for the disintegration test carried out under the relative motion state of the soil sample and the water body, first fill the water tank 1 with water to a depth of 30 cm, hang the hanging basket hook on the hook of the first electronic scale 24, and ensure that the cross bar above the hanging basket hook is on the upper side of the fork at the front end of the lifting rod 22 and does not touch. The electronic device 3 records the reading Me of the first electronic scale 24 at this time. Subsequently, take out the grid hanging basket 21 connected to the lifting rod 22 from the water tank 1, place the soil sample in the middle part of the grid hanging basket 21, keep it balanced, and slowly put it into the water. Hang the hanging basket hook on the hook of the first electronic scale 24, and ensure that the cross bar above the hanging basket hook is on the upper side of the fork at the front end of the lifting rod 22 and does not touch each other. The electronic device 3 starts to record the reading M0 of the first electronic scale 24. Set the moving speed of the lifting rod 22 on the electronic device 3 to 1 cm / s, and the grid hanging basket 21 will move along with the lifting rod 22. It should be noted that when moving downward, after the hanging basket hook is hung on the hook of the first electronic scale 24, the grid hanging basket 21 will not move downward along with the lifting rod 22; when moving upward, after the fork at the front end of the lifting rod 22 forks the cross bar on the hanging basket hook, the hanging basket hook will separate from the hook of the first electronic scale 24, and the grid hanging basket 21 will move upward along with the lifting rod 22. Therefore, the first electronic scale 24 has 6 seconds to stabilize and read the value, ensuring the accuracy of the reading. The reading of the first electronic scale 24 read by the electronic device 3 is the reading when it is stable. If the reading of the first electronic scale 24 at time t is Mt, then the first disintegration amount Dt of the soil sample at the preset duration time t is:

[0095] Dt = (M0 - Mt) / (M0 - Me) X 100%

[0096] A disintegration test is carried out in a state where the soil sample and the water are relatively stationary. The hanging hook of another grid hanging basket 21 is hung on the hanging hook of the second electronic scale 25, and the electronic device 3 records the reading Re of the electronic scale. Subsequently, the grid hanging basket 21 not connected to the lifting rod 22 is taken out of the water tank 1, the soil sample is placed in the middle of the grid hanging basket 21, kept in balance, and slowly placed into the water. The hanging hook of the hanging basket is hung on the hanging hook of the second electronic scale 25, and the electronic device 3 starts to record the reading R0 of the second electronic scale 25. When the reading of the second electronic scale 25 at the preset time t is Rt, the second disintegration amount St of the soil sample at time t in the state where the soil sample and the water are relatively stationary can be calculated. The calculation formula is as follows:

[0097] St = (R0 - Rt) / (R0 - Re) X 100%

[0098] Subsequently, the soil water sensitivity Wt can be calculated. The calculation formula is as follows:

[0099] Wt = Dt / St

[0100] Therefore, the soil water sensitivity is equal to the ratio of the disintegration speed of the soil sample within time t under the relative motion state of the soil and water to the disintegration speed of the soil sample within time t under the relative stationary condition of the soil and water, which can clearly reflect the change of the disintegration speed after the relative motion of the soil and water. This gives practical meaning to the concept of soil water sensitivity. The larger the soil water sensitivity value, the greater the influence of the water flow movement on the soil stability; the smaller the soil water sensitivity value, the smaller the influence of the water flow movement on the soil stability.

[0101] The soil analysis test equipment provided by the embodiments of the present application, each grid hanging basket 21 is used to hold the soil body sample. The electronic device 3 is used to control the motor 23 to drive the lifting rod 22 to move up and down at a preset speed, so that the soil body sample and the water body have a preset relative movement speed, and the disintegration test can be carried out under the condition of relative movement between the soil body sample and the water body. This design simulates the relative movement of the soil body in the natural environment due to factors such as water flow scouring and groundwater fluctuation. In the actual soil around rivers and lakes, or the foundation soil affected by groundwater flow, it will be in a similar dynamic water environment. By simulating the disintegration test under such relative movement conditions, more realistic soil body water sensitivity data can be obtained, making the research results more practically guiding, and providing reliable soil body characteristic references for projects such as water conservancy projects and road and bridge construction facing complex hydrogeological conditions. In addition, by adjusting the preset speed, different intensities of water flow action or relative movement scenarios between the soil body and water can be simulated. The first electronic scale 24 is used to measure the first initial mass of the grid hanging basket 21 corresponding to the first electronic scale 24 when it is immersed in the water body, and the first target mass corresponding to the grid hanging basket 21 after a preset time period. The second electronic scale 25 is used to measure the second initial mass of the grid hanging basket 21 corresponding to the second electronic scale 25 when it is immersed in the water body, and the second target mass corresponding to the grid hanging basket 21 after a preset time period. The electronic device 3 is used to calculate the first disintegration degree by dividing the difference between the first initial mass and the first target mass by the first initial mass, so as to obtain the disintegration degree of the soil body sample under the condition of relative movement. The second disintegration degree is calculated by dividing the difference between the second initial mass and the second target mass by the second initial mass, so as to obtain the disintegration degree of the soil body sample in the static state. By dividing the first disintegration degree by the second disintegration degree, the soil body water sensitivity corresponding to the soil body sample under the relative movement speed is calculated, ensuring the accuracy of the soil body water sensitivity corresponding to the soil body sample at the preset relative movement speed and within the preset time period.

[0102] In an alternative embodiment of the present application, as Figure 3 shown, each soil disintegration test device 2 includes: at least one water jet 26 and at least one wave maker 27, wherein:

[0103] The water jet 26 is used to adjust the jet angle and flow rate under the control of the electronic device 3 to generate water flows with different speeds, different directions and different intensities;

[0104] The wave maker 27 is used to generate waves with various wavelengths and wave heights under the control of the electronic device 3; to further simulate different water body intensities and water body pressures;

[0105] The electronic device 3 is used to calculate the soil body water sensitivity corresponding to the soil body sample under each relative movement speed, each water body intensity and each water body pressure.

[0106] Specifically, the water ejector 26 generally consists of a nozzle, a flow regulating valve, and an angle adjusting mechanism. During operation, water is ejected through the nozzle under pressure to form a water flow. The flow regulating valve controls the water flow rate, thereby adjusting the water flow intensity. The angle adjusting mechanism is driven by the motor 23 or a hydraulic device to change the ejection angle of the nozzle, achieving the generation of water flows in different directions. Multiple water ejectors 26 can be used in combination. By independently controlling the flow rate and angle of each ejector, complex multi-directional water flows can be generated. Optionally, the water ejector 26 can also be a telescopic and bendable ejector, which can quickly change its shape according to different test requirements to generate more complex and diverse water flows.

[0107] There are two common types of wave generators 27: mechanical and pneumatic. The mechanical wave generator 27 drives mechanical structures such as cams and cranks through the motor 23 to make the wave-making plate move up and down or back and forth, pushing the water body to generate waves. By adjusting the rotation speed of the motor 23 and the movement amplitude of the wave-making plate, the wavelength and wave height of the waves can be controlled. The pneumatic wave generator 27 utilizes compressed air released underwater to form a bubble curtain, disturbing the water body to generate waves, and adjusts the wave parameters by controlling the flow rate and pressure of the compressed air. Optionally, the wave generator 27 can also be a telescopic and bendable wave generator 27, which can quickly change its shape according to different test requirements to generate more complex and diverse waves.

[0108] Through each water ejector 26 and each wave generator 27, different water body intensities and water pressures can be simulated. Thus, based on the above test process, the soil water sensitivity corresponding to the soil sample can be calculated under various relative movement speeds, water body intensities, and water pressures.

[0109] The soil analysis test equipment provided by the embodiments of the present application includes a water ejector 26, which is used to adjust the ejection angle and flow rate under the control of the electronic device 3 to generate water flows with different speeds, directions, and intensities; a wave generator 27, which is used to generate waves with various wavelengths and wave heights under the control of the electronic device 3 to further simulate different water body intensities and water pressures; and an electronic device 3, which is used to calculate the soil water sensitivity corresponding to the soil sample under various relative movement speeds, water body intensities, and water pressures, so as to be able to simulate diverse water flow conditions in the natural environment, such as rapid water flow and slow water flow in rivers, and changes in the seepage direction of groundwater, etc., and simulate the disintegration process of the soil sample under waves with various wavelengths and wave heights. Furthermore, the soil water sensitivity corresponding to the soil sample can be calculated under various relative movement speeds, water body intensities, and water pressures.

[0110] In an alternative embodiment of the present application, as Figure 4 shown, the soil disintegration test equipment 2 further includes a pH monitoring probe 28 and an EC monitoring probe 29. Both the pH monitoring probe 28 and the EC monitoring probe 29 are communicatively connected to the electronic device 3, where:

[0111] The pH monitoring probe 28 is used to monitor the pH value of the water body and obtain the change data of the pH value of the water body during the disintegration process of the soil sample.

[0112] The EC monitoring probe 29 is used to monitor the electrical conductivity of the water body and obtain the change data of the electrical conductivity of the water body during the disintegration process of the soil sample.

[0113] The imaging device 4 is used to capture at least one water body image of the water body area corresponding to the grid hanging basket 21 connected to the second electronic scale 25 during the disintegration process of the soil sample.

[0114] The electronic device 3 is used to input the water body image, the change data of the water body pH value, and the change data of the water body electrical conductivity into a preset soil dispersion model and output the soil dispersion corresponding to the soil sample.

[0115] Specifically, the pH monitoring probe generally works based on the principle of the Nernst equation. There is a glass membrane or other sensitive material sensitive to hydrogen ions inside the probe. When the probe is immersed in the water body, the hydrogen ions in the water body will exchange with the surface of the sensitive membrane, forming a potential difference on both sides of the membrane. This potential difference is related to the activity of hydrogen ions in the water body. Through the Nernst equation, the potential difference can be converted into the corresponding pH value, thereby realizing the measurement of the pH value of the water body. During the disintegration process of the soil sample, the change of the water body pH value is monitored in real time to obtain continuous change data of the water body pH value. These data can reflect the possible chemical reactions during the soil disintegration process. For example, alkaline or acidic substances in the soil dissolve into the water body, causing the pH value to change, which helps to analyze the chemical properties of the soil and the disintegration mechanism.

[0116] The EC monitoring probe generally determines the electrical conductivity by measuring the conductivity of the water body. The probe usually consists of two or more electrodes. When a certain voltage is applied across the electrodes, the ions in the water body will move under the action of the electric field, forming an electric current. The electrical conductivity is related to factors such as the concentration, type, and mobility of ions in the water body. By measuring the magnitude of the current and combining known parameters such as the electrode constant, the electrical conductivity of the water body can be calculated. During the disintegration process of the soil sample, the change of the water body electrical conductivity is continuously monitored to obtain the change data of the water body electrical conductivity. The change of the water body electrical conductivity can reflect the dissolution and release of electrolytes in the soil because different ions will enter the water body when the soil disintegrates, thus changing the electrical conductivity of the water body. These data help to understand the ionic composition of the soil and the migration of substances during the disintegration process, providing an important basis for further analyzing the characteristics of the soil.

[0117] The imaging device 4 is used to capture at least one water body image of the water body area corresponding to the grid hanging basket 21 connected to the second electronic scale 25 during the disintegration process of the soil sample.

[0118] In an alternative embodiment of the present application, the preset soil dispersion model includes a feature extraction network and a soil dispersion discriminator. The above "inputting the water body image, the water body pH value change data, and the water body conductivity change data into the preset soil dispersion model and outputting the soil dispersion corresponding to the soil sample" may include the following steps:

[0119] Step S101: Input the water body image, the water body pH value change data, and the water body conductivity change data into the feature extraction network in the preset soil dispersion model.

[0120] Specifically, the electronic device 3 may input the water body image, the water body pH value change data, and the water body conductivity change data into the feature extraction network in the preset soil dispersion model.

[0121] Step S102: The feature extraction network extracts the RGB features and image features corresponding to the water body image. Among them, the image features include at least one of the water body turbidity feature and the water body texture feature.

[0122] Specifically, the feature extraction network includes dilated convolutions with multiple different receptive fields. The above "the feature extraction network extracts the image features corresponding to the water body image" in step S102 may include the following steps:

[0123] Step a1: The feature extraction network identifies the water body image, determines the importance of features at different scales, and assigns weight information to the dilated convolutions of each receptive field.

[0124] Specifically, when processing the water body image, the importance of features at different scales for describing the image content and reflecting the soil dispersion situation is different. Large-scale features may reflect macroscopic information such as the overall contour of the water body and large-area turbid regions; small-scale features can capture microscopic information such as the fine texture of soil particles and local concentration changes. The feature extraction network can dynamically adjust the evaluation method of scale importance according to the input water body image. For example, when there are obvious large-scale turbid regions in the image, the network can automatically increase the attention to large-scale features; when there are some fine particle textures in the image, the network pays more attention to small-scale features.

[0125] For example, output feature maps through the intermediate layer of a convolutional neural network (CNN), and then perform global average pooling or global max pooling operations on these feature maps to obtain statistical information of features at different scales. Then, use a fully connected layer to map these statistical information to a weight vector, where each element in the weight vector corresponds to the importance score of a scale feature. Finally, according to these scores, assign weight information to the dilated convolutions of each receptive field.

[0126] Step a2: identify the water body image and determine the complexity corresponding to different areas in the water body image.

[0127] Specifically, the complexity of different regions in a water body image reflects the information richness of the region. Regions with high complexity may contain more details and changes, such as areas where soil particles are concentrated and areas where water flows rapidly, while regions with low complexity are relatively uniform and flat, such as the edge of a water body.

[0128] Optionally, the feature extraction network can calculate the local variance of each region in the water body image. The larger the local variance, the more drastic the pixel value change in the region, and the higher the complexity.

[0129] Optionally, the feature extraction network can also use texture analysis techniques, such as gray-level co-occurrence matrix (GLCM), to evaluate the complexity of the region by calculating the texture features of the image region, such as contrast, correlation, etc.

[0130] Step a3, according to the complexity corresponding to different areas in the water body image, determine the sampling frequency corresponding to the hole convolution of each different receptive field.

[0131] Specifically, dilated convolution is a technique that introduces holes (i.e. skips some pixels) in the convolution operation, which can expand the receptive field without increasing the number of parameters. The sampling frequency determines the sampling interval of dilated convolution on the image. The higher the sampling frequency, the denser the sampling of the image by dilated convolution, and it can capture more detailed information; the lower the sampling frequency, the sparser the sampling, and the main focus is on the macroscopic features of the image.

[0132] In the feature extraction process, as the number of network layers increases, the feature representation of the image gradually changes from micro to macro. Therefore, the sampling frequency can be dynamically adjusted according to the number of network layers and regional complexity. For example, in the shallow layer of the network, a higher sampling frequency is used for the area with high complexity. As the number of network layers increases, the sampling frequency is gradually reduced to balance the computational efficiency and the accuracy of feature extraction.

[0133] Step a4: extract features of the water body image based on each dilated convolution to determine image features corresponding to the water body image.

[0134] Specifically, after assigning weight information and sampling frequencies to the dilated convolutions of each receptive field, feature extraction of the water body image can be performed based on these dilated convolutions. Dilated convolutions extract various features of the image by performing convolution operations on the image at different scales and different sampling frequencies. These features include, but are not limited to, edge features, texture features, shape features, etc. For example, edge features can reflect the boundaries of soil particles in the water body, texture features can describe the distribution patterns of soil particles, and shape features can represent the morphology of turbid regions in the water body. Finally, a feature fusion method based on an attention mechanism is adopted. By learning the correlations between different features, different weights are automatically assigned to each feature, and then weighted fusion is performed. This can better retain important features and improve the expressive ability of the features. The extracted features are combined and fused to obtain the image features corresponding to the water body image.

[0135] Step S103, the feature extraction network captures the front and back features of the water body pH value change data and the water body conductivity change data from the front and back, and obtains the water body pH value forward feature, the water body pH value backward feature, the water body conductivity forward feature, and the water body conductivity backward feature.

[0136] Specifically, for the water body pH value change data, the data arranged in chronological order are sequentially input into the forward RNN or LSTM, GRU units. Starting from the starting time point, the data at each time step (for example, the pH value measured every hour or every day) is used as the input. The network updates the current hidden layer state based on the current input and the state of the previous hidden layer. During this process, the network gradually learns the change characteristics of the pH value on the forward time axis. For example, if the pH value shows a gradually increasing trend, the hidden layer state of the network will gradually record this trend information, including characteristics such as the rising rate and the changing amplitude. These characteristics will be encoded in the weights and activation values of the hidden layer and form part of the forward features. The forward features may include information such as the initial value of the pH value, the slope during the rising or falling stage, and the time period of continuous rising or falling.

[0137] Reverse feature extraction is to input the time series data in reverse order into the network. For the water body pH value change data, starting from the data at the last time point, it is input into the network unit in reverse. Similarly, the network updates the hidden layer state based on the current input and the state of the previous hidden layer (here, the previous hidden layer in the reverse time order). When inputting the data in reverse, the network will notice the process of the pH value gradually decreasing from a high value and record the characteristics of this reverse change, such as the starting time of recovery and the recovery rate. Reverse features can provide a different perspective from the forward. In some cases, the later changes in the data may correct the previous trends. Reverse features help to discover hidden patterns in the data and comprehensively understand the dynamic process of pH value changes.

[0138] When the water body conductivity change data is input in the forward direction, the network captures the characteristics of the forward change of conductivity over time, such as the starting point of conductivity increase or decrease, the change amplitude, the periodic change characteristics (if any), etc. Reverse feature extraction: Input the conductivity data in the reverse direction, and the network focuses on the changes of conductivity on the reverse time axis.

[0139] Step S104, the feature extraction network determines the key time series in the water body pH value change data and the water body conductivity change data based on the attention mechanism, and obtains the key time series of the water body pH value and the key time series of the water body conductivity.

[0140] Specifically, the feature extraction network will perform preliminary processing on the water body pH value change data and the water body conductivity change data, and convert the data of each time step into a suitable feature representation. Usually, recurrent neural networks (RNNs) and their variants, such as long short-term memory networks (LSTMs) or gated recurrent units (GRUs), are used to complete this task. These network structures can process time series data and learn the temporal features in the data. Taking the water body pH value change data as an example, a series of pH value measurement values arranged in chronological order are input into the LSTM network, and the LSTM network will output a hidden state vector at each time step, which contains the historical information before this time step and the input information of the current time step. The same process is also applied to the water body conductivity change data.

[0141] After obtaining the feature representation of each time step, the attention mechanism calculates the attention weights of each time step. The attention weights reflect the importance of the data at this time step in the entire time series.

[0142] Specifically, an attention function will be defined. This function usually uses a learnable weight matrix to map the feature representation of each time step to a scalar value, and this scalar value is the attention score of this time step. Then, through a normalization operation (such as the softmax function), these scores are converted into a probability distribution, that is, the attention weights. Taking the water body pH value change data as an example, assuming that the pH value changes sharply at a certain time step, then the attention score of this time step will be relatively high. After being processed by the softmax function, its corresponding attention weight will also be large, indicating that the data at this time step is more important for subsequent analysis.

[0143] According to the calculated attention weights, select the time steps with attention weights greater than a certain preset threshold as the key time steps, and the data corresponding to these key time steps constitutes the key time series.

[0144] For the water body pH value change data, screen out the time steps with higher attention weights, and arrange the pH value data of these time steps in chronological order to obtain the key time series of the water body pH value. The same method also applies to the water body conductivity change data to obtain the key time series of the water body conductivity.

[0145] Step S105: Identify the key time series of the water body pH value and the key time series of the water body conductivity, and determine the key features of the water body pH value and the key features of the water body conductivity.

[0146] Specifically, to identify the key time series of the water body pH value and the key time series of the water body conductivity, calculate basic statistics such as mean, median, and standard deviation. Then, based on the obtained basic statistics, determine the key features of the water body pH value and the key features of the water body conductivity.

[0147] Step S106: Perform feature fusion on the RGB features, image features, forward features of the water body pH value, backward features of the water body pH value, forward features of the water body conductivity, backward features of the water body conductivity, key features of the water body pH value, and key features of the water body conductivity to generate initial fusion features.

[0148] Specifically, the above step S106 may include the following steps:

[0149] Step b1: Using the RGB features, image features, forward features of the water body pH value, backward features of the water body pH value, forward features of the water body conductivity, backward features of the water body conductivity, key features of the water body pH value, and key features of the water body conductivity as nodes, and the correlation between two features as the weight of the edge, construct a feature graph.

[0150] Specifically, RGB features, image features (such as water turbidity features, water texture features, etc.), forward features of water body pH value, backward features of water body pH value, forward features of water body conductivity, backward features of water body conductivity, key features of water body pH value, and key features of water body conductivity are selected as nodes. These features reflect the characteristics of the soil sample during the disintegration process from different perspectives. For example, RGB features and image features visually present the appearance and physical state of the water body, while pH value and conductivity features reveal the dynamic changes in the chemical properties of the water body, covering multi-dimensional information. Then, the correlation between pairwise features is calculated to determine the edge weights. Multiple methods can be used for correlation analysis, such as Pearson correlation coefficient, Spearman correlation coefficient, etc. Taking the key features of pH value and conductivity as an example, in some experiments, when the pH value changes sharply, the conductivity also shows a significant upward or downward trend. By calculating the correlation coefficient, a relatively high absolute value will be obtained, indicating a strong correlation between the two, and the corresponding edge weight will be large; conversely, if there is almost no synchronous change pattern between the two features and the correlation coefficient is close to 0, the edge weight will be small. In this way, the internal relationship between features is reflected in the feature map in the form of edge weights.

[0151] Step b2, use a graph neural network to process the feature map; in each layer of message passing, each node updates its own feature representation according to the features of its neighbor nodes and the edge weights.

[0152] Specifically, in each layer of message passing of the graph neural network, a node updates its own feature representation according to the features of its neighbor nodes and the edge weights. For example, assume that the RGB feature node is adjacent to the water turbidity feature node, and the edge weight between them is large, indicating a strong correlation. During the message passing process, the RGB feature node will receive information from the water turbidity feature node, and this information will be weighted according to the edge weight. If the edge weight is 0.8, then the information passed from the water turbidity feature node to the RGB feature node will be given a higher proportion. The RGB feature node will combine its own original features and the information received from the neighbor node, and update its own feature representation according to the update rules of the graph neural network (such as through specific activation functions and weight matrix operations).

[0153] As messages are passed layer by layer in the graph, nodes continuously update their own features. In this process, the graph neural network can learn complex interaction relationships between features. For example, after the forward feature node of the water body pH value updates its own feature by receiving information from neighbor nodes, the updated feature will be passed to other neighbor nodes, enabling the continuous propagation and fusion of feature information throughout the graph. This propagation mechanism allows distant nodes to indirectly influence each other through multi-step message passing, uncovering potential long-range dependencies between features. For instance, the backward feature node of conductivity may influence the final feature representation of the RGB feature node through multiple message passes with intermediate nodes, even if they are not directly connected in the initial feature map.

[0154] Step b3, after message passing and feature update for a preset number of rounds, perform a weighted sum of the features of all nodes in the feature map to obtain the initial fusion feature.

[0155] Specifically, the preset number of rounds of message passing and feature update is a key parameter setting. If the number of rounds is too small, the features may not be fully fused, and insufficient connections between features may not be uncovered. If the number of rounds is too large, overfitting may occur, and the model may over-adapt to the noise in the training data. For example, in some experiments, after multiple attempts, it is found that setting 5 - 10 rounds of message passing can fully fuse the features while maintaining the generalization ability of the model. Within this preset number of rounds, the node features in the feature map are continuously optimized and integrated.

[0156] After the preset number of rounds, perform a weighted sum of the features of all nodes in the feature map to obtain the initial fusion feature. The weight of each node can be determined based on its importance in the entire feature map. Importance can be comprehensively measured by factors such as the degree of the node (the number of edges connected to the node) and the sum of edge weights. For example, a node connected to multiple other nodes with large edge weights indicates that it plays an important role in feature interaction and can be assigned a higher weight during the weighted sum. Through this weighted sum method, features of different types and importance levels are fused to generate an initial fusion feature containing rich information from multi-source data, providing a better feature input for subsequent tasks such as accurately judging the dispersion degree of soil.

[0157] Step S107, input the initial fusion feature into the soil dispersion discriminator, and the soil dispersion discriminator outputs the soil dispersion corresponding to the soil sample.

[0158] Specifically, the above step S107 may include the following steps:

[0159] Step c1: Input the initial fusion feature into the soil dispersion discriminator. At the initial layer of the soil dispersion discriminator, extract different types of features through different convolutional layer branches, and introduce an attention mechanism to fuse the features extracted by each convolutional layer branch to generate a target fusion feature.

[0160] Specifically, input the initial fusion feature into the soil dispersion discriminator. At the initial layer of the soil dispersion discriminator, set different convolutional layer branches. Each branch has specific parameters such as the convolutional kernel size and stride, which are used to extract different types of features. For example, one branch uses a larger convolutional kernel and is good at extracting global and macroscopic features, perhaps paying attention to the overall distribution pattern of the turbid areas in the water body; another branch uses a smaller convolutional kernel and focuses on capturing local and subtle features, such as the edge details of soil particles or local texture features. This multi-branch design can comprehensively mine the initial fusion feature from multiple scales and angles, making full use of various information contained therein.

[0161] Then, introduce an attention mechanism to fuse the features extracted by each convolutional layer branch. The attention mechanism determines the contribution degree of each feature in the fusion process by calculating the importance weight of each feature. In specific implementation, usually, the features extracted by each convolutional layer branch are input into an attention module. This module will calculate the correlation between features, such as using dot product operation or other similarity measurement methods, to obtain the association degree of each feature with other features. For features closely related to soil dispersion, the attention mechanism will assign a higher weight, so that these key features can more prominently affect the final result during fusion. For example, if in some experiments it is found that the feature reflecting the texture of the soil particle aggregation area is closely related to soil dispersion, the attention mechanism will assign a larger weight to the output of the convolutional layer branch that extracts this feature, thereby generating a more discriminative target fusion feature.

[0162] Step c2: The soil dispersion discriminator makes a preliminary evaluation of the target fusion feature. By calculating the correlation score between each sub-feature in the fusion feature and the known soil dispersion label, key features whose contribution to discriminating soil dispersion is greater than a preset threshold are selected.

[0163] Specifically, the soil dispersion discriminator makes a preliminary evaluation of the target fusion features. A large number of sample data with known soil dispersion labels are pre-stored in the discriminator. For each sub-feature in the target fusion features, the soil dispersion discriminator calculates the correlation score between it and these known labels. Multiple methods can be used to calculate the correlation, such as the Pearson correlation coefficient, which can measure the linear correlation degree between two variables (here the sub-feature value and the soil dispersion label value). For example, for a sub-feature reflecting the change trend of water conductivity, if in the dataset, when the value of this sub-feature increases, the value of the soil dispersion label also shows an obvious upward or downward trend, then the absolute value of the calculated Pearson correlation coefficient will be relatively large, indicating a strong correlation between this sub-feature and soil dispersion.

[0164] According to the calculated correlation scores, key features that contribute more to the discrimination of soil dispersion than a preset threshold are selected. A correlation score threshold can be set, and only sub-features with scores higher than this threshold are recognized as key features. For example, if the threshold is set to 0.6, then sub-features with correlation scores greater than 0.6 will be retained. These key features contain the most important information for judging soil dispersion. By screening, sub-features with weak correlation with soil dispersion that may interfere with the judgment are removed, improving the accuracy and efficiency of subsequent analysis.

[0165] Step c3, using feature enhancement techniques, enhance each key feature to obtain enhanced features.

[0166] Specifically, use feature enhancement techniques to enhance the selected key features. Common feature enhancement techniques include but are not limited to data transformation, feature synthesis, etc. For example, for a key feature reflecting the particle size distribution of the soil, data transformation methods such as logarithmic transformation or power transformation can be used to highlight the detailed information in the feature, making the differences in particle sizes that were originally difficult to distinguish more obvious. In terms of feature synthesis, according to domain knowledge, multiple related key features can be combined and calculated to generate new features. For example, by weighted summing two key features reflecting the change range and change rate of the water pH value, an enhanced feature that comprehensively reflects the dynamic change of the pH value is obtained. This feature may have a stronger indicative effect on the judgment of soil dispersion. Through the application of the above feature enhancement techniques, enhanced features are obtained. These enhanced features, while retaining the core information of the original key features, further strengthen the information that has an important impact on soil dispersion and suppress noise and irrelevant information. For example, the enhanced feature reflecting the soil texture feature after enhancement processing may have higher contrast and clarity, and can more accurately reflect the arrangement and aggregation mode of soil particles, providing more powerful support for accurately judging soil dispersion.

[0167] Step c4: Output the soil dispersion degree corresponding to the soil sample based on the enhanced features.

[0168] Specifically, based on the obtained enhanced features, the soil dispersion discriminator outputs the soil dispersion degree corresponding to the soil sample. Inside the soil dispersion discriminator, there usually contains a classification or regression model. For example, a support vector machine (SVM) is used for classification tasks (judging which grade category the soil dispersion degree belongs to), or a linear regression model is used for regression tasks (predicting the specific soil dispersion degree value). Taking SVM as an example, it will find an optimal classification hyperplane according to the distribution of the enhanced features in the feature space to distinguish soil samples with different dispersion degrees. As the input of the model, the rich and prominent information of the enhanced features enables the model to make more accurate decisions and improve the accuracy of judging the soil dispersion degree.

[0169] The finally output soil dispersion degree results can be applied to multiple fields. In the agricultural field, it helps to evaluate the soil stability, guide reasonable irrigation and fertilization, and avoid soil erosion and fertility decline caused by soil dispersion; in civil engineering, it provides important soil property parameters for foundation treatment, road and bridge construction, etc., helps to design a reasonable foundation structure, and ensures the stability and safety of the project; in environmental science research, it is used to analyze the soil erosion risk and formulate effective soil and water conservation and environmental protection measures.

[0170] The soil analysis test equipment provided by the embodiments of this application, the pH monitoring probe 28 is used to monitor the pH value of the water body, and obtain the change data of the pH value of the water body during the disintegration process of the soil sample, so as to reflect the dissolution, reaction, etc. of acid-base substances during the soil disintegration process. For example, if the soil contains alkaline minerals, the dissolution of alkaline substances during disintegration will increase the pH value of the water body, and this process can be traced through the pH value change data. The EC monitoring probe 29 is used to monitor the water body conductivity of the water body and obtain the change data of the water body conductivity during the disintegration process of the soil sample; it can intuitively reflect the change of the ion release amount and type in the soil because the ion concentration change will directly affect the conductivity. These two chemical property data complement the physical form information provided by the water body image, providing a rich data basis for determining the soil dispersion degree from different dimensions and making the analysis more comprehensive and accurate.

[0171] Then, the electronic device 3 is used to input the water body image, the water body pH value change data, and the water body conductivity change data into the feature extraction network in the preset soil dispersion model. This multi-source data input method can obtain soil dispersion-related information from different perspectives. The feature extraction network extracts the RGB features and image features corresponding to the water body image. Specifically, the feature extraction network identifies the water body image, determines the importance of features at different scales, and assigns weight information to the dilated convolutions in each receptive field, so that the soil dispersion discriminator can automatically adjust the attention to features at different scales according to the characteristics of the input water body image. By identifying the water body image, the complexity corresponding to different regions in the water body image is determined, so that potential features in the water body image can be highlighted through complexity analysis. According to the complexity corresponding to different regions in the water body image, the sampling frequency corresponding to the dilated convolution in each different receptive field is determined; the optimal allocation of computing resources can be achieved. For regions with high complexity, the sampling frequency is increased so that the dilated convolution can collect the feature information of this region more carefully and fully capture details such as complex textures and particle distributions; for regions with low complexity, the sampling frequency is reduced to reduce unnecessary computational volume while ensuring the acquisition of basic features. In this way, it can not only comprehensively extract the key information of the image but also avoid wasting computing resources caused by over-sampling simple regions, improving the computational efficiency of the feature extraction process. Based on each dilated convolution, feature extraction is performed on the water body image to determine the image features corresponding to the water body image. At different scales, the dilated convolution can capture multi-scale features from the macroscopic distribution of water body turbidity regions to the microscopic soil particle textures, and make adaptive adjustments according to the image region complexity and scale feature importance, so as to obtain richer and more accurate image features, providing more comprehensive information support for judging soil dispersion. The feature extraction network captures the front and back features of the water body pH value change data and the water body conductivity change data from the forward and reverse directions, obtaining the water body pH value forward feature, the water body pH value backward feature, the water body conductivity forward feature, and the water body conductivity backward feature, so as to comprehensively understand the change trends of the water body pH value and the water body conductivity over time. The feature extraction network determines the key time series in the water body pH value change data and the water body conductivity change data based on the attention mechanism, obtaining the water body pH value key time series and the water body conductivity key time series, so as to focus on the time periods that have a greater impact on the dispersion degree of the soil sample. By identifying the water body pH value key time series and the water body conductivity key time series, the water body pH value key features and the water body conductivity key features are determined; further refining the information closely related to soil dispersion, and by accurately identifying the key features, it can provide more valuable input for subsequent feature fusion and discrimination.Next, using RGB features, image features, forward water body pH features, backward water body pH features, forward water body conductivity features, backward water body conductivity features, key water body pH features, and key water body conductivity features as nodes, the relationships between different types of features can be comprehensively presented. The correlation between pairwise features is used as the weight of the edge to construct a feature graph, which can uncover some potential feature associations that are not easily observable directly. There may be complex non-linear relationships between some features, and these relationships can be discovered through correlation analysis and reflected in the feature graph. Use a graph neural network to process the feature graph; in each layer of message passing, each node updates its own feature representation based on the features of its neighbor nodes and the weight of the edge. This way can effectively capture the complex interactions between features. In the study of soil dispersion degree, the mutual influence between different features is not a simple linear relationship, but there is a complex coupling effect. The graph neural network can simulate this complex interaction process through the message passing mechanism. After a preset number of rounds of message passing and feature update, the features of all nodes in the feature graph are weighted and summed to obtain the initial fusion feature. This way of weighted summation can comprehensively consider the contribution of each feature to the soil dispersion degree. In the feature graph, the weight of the edge already reflects the correlation between features. Through weighted summation, features with stronger correlation will occupy a larger proportion in the fusion feature, while features with weaker correlation will be relatively smaller. This can ensure that the fusion feature can more accurately reflect the comprehensive influence of different features on the soil dispersion degree, avoiding problems such as information loss or inaccuracy that may be caused by simple averaging or equal-weight fusion. Input the initial fusion feature into the soil dispersion discriminator, and the soil dispersion discriminator outputs the soil dispersion degree corresponding to the soil sample. The soil dispersion discriminator can accurately output the soil dispersion degree by comprehensively analyzing and judging this information.

[0172] In an alternative embodiment of the present application, the electronic device 3 is used to obtain the soil physical and chemical property data and geographical information data corresponding to each soil sample; the soil physical and chemical property data includes at least one of soil particle characteristics, three-phase proportion indexes of soil, physical and chemical state indexes of soil, and permeability indexes of soil; the geographical information data includes at least one of longitude and latitude, altitude, and geological structure;

[0173] Generate a soil water sensitivity training data set based on the soil physical and chemical property data, geographical information data, and soil water sensitivity corresponding to each soil sample; the soil water sensitivity in the soil water sensitivity training data set is the label data; based on the soil water sensitivity training data set, train to obtain a soil water sensitivity prediction model;

[0174] Generate a soil dispersion training data set based on the soil physical and chemical property data, geographical information data, and soil water sensitivity corresponding to each soil sample; the soil dispersion in the soil dispersion training data set is the labeled data; based on the soil dispersion training data set, train a soil dispersion prediction model.

[0175] Among them, soil particle characteristics include information such as the size, shape, and gradation of soil particles. The size of soil particles directly affects the pore size and distribution of the soil, and thus affects the permeability and stability of the soil. For example, sandy soil particles are larger, with large pores and strong permeability; clay particles are fine, with small pores, weak permeability, and strong cohesion. The three-phase ratio index of soil: It involves the proportional relationship of the solid phase (soil particles), liquid phase (water), and gas phase (air) in the soil. Such as the void ratio (the ratio of the pore volume in the soil to the soil particle volume), water content (the ratio of the mass of water in the soil to the mass of soil particles), etc. These indexes reflect the degree of water saturation and compaction state of the soil mass, and have an important impact on the mechanical properties and water sensitivity of the soil mass. Soils with high water content are more likely to soften, disintegrate, etc. when encountering water, showing a higher water sensitivity. The physical and chemical state indexes of soil: For example, the relative density of sandy soil, the liquidity index of cohesive soil, the pH index, the redox potential index, the cation exchange capacity index, etc. The relative density reflects the degree of compaction of sandy soil in its natural state. Sandy soil with high density has better stability when subjected to the action of water; the liquidity index is used to judge the physical state of cohesive soil (such as hard, plastic, flowing, etc.). The sensitivity of cohesive soil to water varies significantly under different states. The permeability index of soil: Such as the permeability coefficient, which measures the ability of the soil mass to allow water to pass through. For soil masses with a large permeability coefficient, the water flows fast in it, which may cause the fine particles in the soil mass to be washed away, aggravating the dispersion of the soil mass, and is closely related to the soil dispersion. Geographical information data, including longitude and latitude: Determine the geographical location of the soil sample. Soils in different geographical locations may be affected by different factors such as climate and vegetation, and thus affect their characteristics. For example, due to high temperature and heavy rainfall in tropical regions, the chemical weathering is strong, and the soil may have different mineral compositions and physical and chemical properties from those in cold regions, and the sensitivity to water is also different. Altitude: The altitude affects climate conditions such as temperature and precipitation, and thus affects the formation and characteristics of the soil. In high-altitude areas, the temperature is low and the soil freezing period is long. The physical state and water sensitivity of the soil mass may be different from those in low-altitude areas. In addition, altitude may also affect the groundwater level, indirectly affecting the water saturation state and permeability of the soil mass. Geological structure: The geological structure determines the origin and distribution of the soil mass. For example, the soil mass near the fault, due to the action of geological stress, its structure may be more broken, the connection between soil particles is weak, and it is more likely to disperse and deform under the action of water, and the water sensitivity and dispersion may be higher.

[0176] Based on the obtained soil physical and chemical property data, geographical information data, and soil water sensitivity data obtained through experiments or other means, a soil water sensitivity training dataset is generated. In this dataset, the soil water sensitivity is used as the label data, while the soil physical and chemical property data and geographical information data are used as the feature data. For example, for a group of soil samples located in a certain mountainous area (the geographical location is determined by longitude, latitude, and altitude, and the geological structure is a fold belt), the soil particle characteristics are relatively coarse particles and well-graded, the three-phase ratio index shows a low water content, the physical state index indicates a dense state, and the permeability coefficient is large. The corresponding soil water sensitivity is determined to be a low value through a soil disintegration experiment. These data are sorted into a record and incorporated into the soil water sensitivity training dataset. A large number of such records constitute the complete training dataset, which is used to train the model to learn the relationship between soil characteristics and water sensitivity.

[0177] Similarly, a soil dispersivity training dataset is generated. Again, the soil dispersivity is used as the label data, and the soil physical and chemical property data and geographical information data are used as the feature data. For example, another group of soil samples located in a river alluvial plain (specific geographical information), the soil particle characteristics are relatively fine particles and good sorting, the porosity ratio is large in the three-phase ratio index, the physical state index shows a plastic state, the permeability coefficient is moderate, and its soil dispersivity is determined to be at a medium level through methods such as water body image analysis. These data are formed into a record in the training dataset, and numerous such records build the soil dispersivity training dataset, providing data support for training related prediction models.

[0178] Use the constructed soil water sensitivity training dataset to train the model. Multiple machine learning models can be used, such as decision trees, random forests, neural networks, etc. Taking the neural network as an example, the feature data in the training dataset is input into the input layer of the neural network. Through a series of neurons in the hidden layer, the data is subjected to feature extraction and transformation, and finally the predicted soil water sensitivity value is obtained at the output layer. During the training process, by comparing the predicted value with the true soil water sensitivity label value, the loss function (such as the mean square error loss function) is calculated, and the backpropagation algorithm is used to adjust the weights and biases of the neural network, so that the loss function gradually decreases, and the model can predict the soil water sensitivity more and more accurately.

[0179] For the soil dispersion prediction model, the corresponding training dataset is also used for training. If the random forest model is selected, the random forest consists of multiple decision trees, and each decision tree is trained based on a random subset of the training dataset. During the training process, the decision tree classifies or regresses the input feature data (physical and chemical properties of the soil and geographical information) to predict the soil dispersion. By integrating the prediction results of multiple decision trees (such as using the voting method in classification tasks and the averaging method in regression tasks), the final prediction value is obtained. The parameters of the random forest (such as the number of decision trees, maximum depth, etc.) are continuously adjusted to optimize the prediction performance of the model for soil dispersion.

[0180] The soil analysis test equipment provided by the embodiments of the present application obtains the physical and chemical property data and geographical information data of each soil sample, and generates a soil water sensitivity training dataset based on the physical and chemical property data, geographical information data, and soil water sensitivity of each soil sample; the soil water sensitivity in the soil water sensitivity training dataset is the labeled data; based on the soil water sensitivity training dataset, a soil water sensitivity prediction model is trained, so that the soil water sensitivity can be predicted by using various features of the soil. A soil dispersion training dataset is generated based on the physical and chemical property data, geographical information data, and soil dispersion of each soil sample; the soil dispersion in the soil dispersion training dataset is the labeled data; based on the soil dispersion training dataset, a soil dispersion prediction model is trained. Thus, the soil dispersion can be predicted by using various features of the soil.

[0181] Although the embodiments of the present 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 present invention, and such modifications and variations fall within the scope defined by the appended claims.

Claims

1. A soil analysis test device, characterized in that, The soil analysis test equipment includes: A water tank, a soil disintegration test equipment, an electronic device, and a camera device. The electronic device is communicatively connected to the camera device and the soil disintegration test equipment. Among them: The water tank is used to hold water. The soil disintegration test equipment is used to simulate the disintegration process of the soil sample under the relative movement state with water, and record the mass change data of the soil sample during the disintegration process. The camera device is used to capture the water body image corresponding to the water in the soil disintegration test equipment during the disintegration process of the soil sample. The electronic device is used to identify the mass change data to determine the soil water sensitivity corresponding to the soil sample; and identify the water body image to determine the soil dispersion degree corresponding to the soil sample. Among them, the water tank is divided into two water body areas. The soil disintegration test equipment includes: two grid hanging baskets, and the two grid hanging baskets respectively extend into the two corresponding water body areas of the water tank. One of the grid hanging baskets is connected to a lifting rod, the lifting rod can drive the grid hanging basket to move up and down, the lifting rod is connected to a motor, and the grid hanging basket is connected to a first electronic scale. The other grid hanging basket is connected to a second electronic scale. The motor, the first electronic scale, and the second electronic scale are all communicatively connected to the electronic device. Among them: Each grid hanging basket is used to hold the soil sample. The electronic device is used to control the motor to drive the lifting rod to move up and down at a preset speed, so that the soil sample and the water have a preset relative movement speed, and the disintegration test can be carried out under the condition of relative movement between the soil sample and the water. The first electronic scale is used to measure the first initial mass of the grid hanging basket corresponding to the first electronic scale when it is immersed in water, and the first target mass of the grid hanging basket after a preset time. The second electronic scale is used to measure the second initial mass of the grid hanging basket corresponding to the second electronic scale when it is immersed in water, and the second target mass of the grid hanging basket after a preset time. The electronic device calculates the first disintegration degree by dividing the difference between the first initial mass and the first target mass by the first initial mass. Calculates the second disintegration degree by dividing the difference between the second initial mass and the second target mass by the second initial mass. Calculates the soil water sensitivity corresponding to the soil sample under the relative movement speed by dividing the first disintegration degree by the second disintegration degree.

2. The soil analysis test equipment according to claim 1, characterized in that Each of the soil disintegration test equipment includes: at least one water jet and at least one wave maker. Among them: The water jet is used to adjust the jet angle and flow rate under the control of the electronic device to generate water flows with different speeds, different directions, and different intensities. The wave maker is used to generate waves with various wavelengths and wave heights under the control of the electronic device; to further simulate different water body intensities and water pressures. The electronic device is used to calculate the soil water sensitivity corresponding to the soil sample under each of the relative movement speeds, each of the water body intensities, and each of the water pressures.

3. The soil analysis test equipment according to claim 1, characterized in that The soil disintegration test equipment further includes a pH monitoring probe and an EC monitoring probe. The pH monitoring probe and the EC monitoring probe are both communicatively connected to the electronic device. Among them: The pH monitoring probe is used to monitor the pH value of the water body to obtain the water body pH value change data of the soil sample during the disintegration process. The EC monitoring probe is used to monitor the water conductivity of the water body, and obtain the water conductivity change data of the soil sample during the disintegration process; The imaging device is used to capture at least one water image of the water area corresponding to the grid hanging basket connected to the second electronic scale during the disintegration process of the soil sample; The electronic device is used to input the water image, the water pH value change data, and the water conductivity change data into a preset soil dispersion model, and output the soil dispersion corresponding to the soil sample.

4. The soil analysis test equipment according to claim 3, characterized in that The preset soil dispersion model includes a feature extraction network and a soil dispersion discriminator, where: The electronic device is used to input the water image, the water pH value change data, and the water conductivity change data into the feature extraction network in the preset soil dispersion model; The feature extraction network extracts the RGB features and image features corresponding to the water image; the image features include at least one of the water turbidity feature and the water texture feature; The feature extraction network captures the front and back features of the water pH value change data and the water conductivity change data from the front and back, and obtains the water pH value forward feature, the water pH value backward feature, the water conductivity forward feature, and the water conductivity backward feature; The feature extraction network determines the key time series in the water pH value change data and the water conductivity change data based on the attention mechanism, and obtains the water pH value key time series and the water conductivity key time series; Identify the water pH value key time series and the water conductivity key time series to determine the water pH value key feature and the water conductivity key feature; Perform feature fusion on the RGB features, the image features, the water pH value forward feature, the water pH value backward feature, the water conductivity forward feature, the water conductivity backward feature, the water pH value key feature, and the water conductivity key feature to generate an initial fusion feature; Input the initial fusion feature into the soil dispersion discriminator, and the soil dispersion discriminator outputs the soil dispersion corresponding to the soil sample.

5. The soil analysis test equipment according to claim 4, characterized in that The feature extraction network includes dilated convolutions with multiple different receptive fields; where: The feature extraction network identifies the water image, determines the importance of different scale features, and assigns weight information to the dilated convolutions of each receptive field; Identify the water image to determine the complexity corresponding to different regions in the water image; According to the complexity corresponding to different regions in the water image, determine the sampling frequency corresponding to the dilated convolutions of each different receptive field; Based on each dilated convolution, perform feature extraction on the water image to determine the image features corresponding to the water image.

6. The soil analysis test equipment according to claim 4, wherein The electronic device is used to construct a feature map with the RGB feature, the image feature, the forward feature of the water body pH value, the backward feature of the water body pH value, the forward feature of the water body conductivity, the backward feature of the water body conductivity, the key feature of the water body pH value, and the key feature of the water body conductivity as nodes, and the correlation between two features as the weight of the edge. Process the feature map using a graph neural network; in each layer of message passing, each node updates its own feature representation according to the features of its neighbor nodes and the weight of the edge. After a preset number of rounds of message passing and feature update, perform a weighted sum of the features of all nodes in the feature map to obtain the initial fusion feature.

7. The soil analysis test equipment according to claim 4, characterized in that, The electronic device is used to input the initial fusion feature into the soil dispersion discriminator; in the initial layer of the soil dispersion discriminator, extract different types of features through different convolutional layer branches, and introduce an attention mechanism to fuse the features extracted by each convolutional layer branch to generate a target fusion feature. The soil dispersion discriminator makes a preliminary evaluation of the target fusion feature, and by calculating the correlation score between each sub-feature in the fusion feature and the known soil dispersion label, filters out the key features that contribute more than a preset threshold to the discrimination of soil dispersion. Adopt a feature enhancement technique to enhance each of the key features to obtain enhanced features. Based on the enhanced features, output the soil dispersion corresponding to the soil sample.

8. The soil analysis test equipment according to claim 1, characterized in that The electronic device is used to obtain the soil physical and chemical property data and geographical information data corresponding to each soil sample; the soil physical and chemical property data includes at least one of soil particle characteristics, soil three-phase ratio indicators, soil physical and chemical state indicators, and soil permeability indicators; the geographical information data includes at least one of longitude and latitude, altitude, and geological structure. Generate a soil water sensitivity training data set based on the soil physical and chemical property data, the geographical information data, and the soil water sensitivity corresponding to each soil sample; the soil water sensitivity in the soil water sensitivity training data set is the label data; based on the soil water sensitivity training data set, train a soil water sensitivity prediction model. Generate a soil dispersion training data set based on the soil physical and chemical property data, the geographical information data, and the soil water sensitivity corresponding to each soil sample; the soil dispersion in the soil dispersion training data set is the label data; based on the soil dispersion training data set, train a soil dispersion prediction model.

Images