Three-dimensional intelligent monitoring and early disaster identification and warning method for soft rock slope

Abstract

The present application provides a highway soft rock slope three-dimensional intelligent monitoring and early disaster identification and warning method, belonging to the technical field of slope monitoring, which comprises the following steps: using a remote sensing device to take pictures of the highway embankment, obtaining corresponding remote sensing image data, identifying the remote sensing image, inputting the image into a machine model for feature comparison learning, conducting hidden danger point screening and disaster body identification, sampling the soft rock area, establishing a soft rock body three-dimensional space model, collecting data inside the soft rock body in real time by setting a collection device, collecting the data as dynamic data of the soft rock body three-dimensional space model, inputting the remote sensing data and real-time collection data into an online warning system, judging whether the soft rock area will lose stability, issuing a warning information for the area where instability or landslide may occur, notifying the management personnel to check on site and take protective measures.

Description

Technical Field

[0001] This invention relates to the field of slope monitoring technology, and in particular to a method for three-dimensional intelligent monitoring and early disaster identification and warning of soft rock slopes. Background Technology

[0002] The geological structure of Guangxi is complex, and major geological disaster risks are often characterized by their concealment, suddenness, and uncertainty, which greatly increases the difficulty of proactive prevention and monitoring and early warning.

[0003] Traditional methods of geological disaster prevention and control primarily rely on regular on-site investigations by personnel for early identification. High-risk points are monitored manually using methods such as GPS, levels, total stations, close-range photogrammetry, and displacement gauges. While these monitoring technologies have achieved considerable success in geological disaster monitoring and prediction research and have accumulated rich practical experience, leading to breakthroughs in disaster early warning in our region, they still have some significant shortcomings under the unique conditions of our area.

[0004] (1) Manual surveys are labor-intensive, inefficient, and rely solely on visual identification of early geological hazards; (2) Landslides, debris flows, and other hazards typically occur in high-altitude, densely forested areas with harsh geological environments, making them difficult for humans to access; (3) Traditional measurement methods (leveling instruments, total stations, and close-range photogrammetry) require relevant personnel to be on-site to observe and record a large amount of field measurement data, resulting in a large workload, low efficiency, and high costs; (4) Due to the influence of observation costs and other factors, traditional measurement techniques cannot obtain continuous surface deformation information in the monitoring area, leading to divergent results, low spatial resolution, and difficulty in achieving early warning effects; (5) A complete geological hazard prevention and control system, encompassing geological hazard identification, monitoring, and remediation, has not yet been established. Therefore, it is necessary to design a three-dimensional intelligent monitoring and early warning method for soft rock slopes. Summary of the Invention

[0005] The purpose of this invention is to provide a method for three-dimensional intelligent monitoring and early disaster identification and warning of soft rock slopes, thereby solving the technical problems mentioned in the background art.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A method for three-dimensional intelligent monitoring and early disaster identification and warning of soft rock slopes, the method comprising the following steps:

[0008] Step 1: Use remote sensing equipment to photograph the highway subgrade and obtain the corresponding remote sensing image data;

[0009] Step 2: Recognize the remote sensing images and input them into the machine model for feature comparison and learning to screen for potential hazards and identify disaster bodies;

[0010] Step 3: Sample the soft rock area and establish a three-dimensional spatial model of the soft rock mass;

[0011] Step 4: Collect data from the interior of the soft rock mass in real time by setting up a data acquisition device. The collected data will be used as dynamic data for the three-dimensional spatial model of the soft rock mass.

[0012] Step 5: Input the remote sensing data and real-time acquired data into the online early warning system to determine if there are any soft rock areas that may become unstable;

[0013] Step 6: Issue early warning information for areas where instability or landslides may occur, and notify management personnel to conduct on-site inspections and take protective measures.

[0014] Furthermore, the remote sensing equipment in step 1 includes geological hazard surveys based on spaceborne platforms and detailed geological hazard surveys based on airborne platforms. The specific process of geological hazard surveys based on spaceborne platforms is as follows:

[0015] By combining InSAR and optical remote sensing methods using a satellite remote sensing platform, information on landslide morphology and land cover classification is extracted using optical remote sensing, while deformation field is extracted using InSAR. The combination of these two methods yields factors such as surface deformation rate, land cover classification, ground inversion temperature, and vegetation cover, which are used to construct a geological hazard knowledge graph. This graph identifies the characteristics and relationships of three elements in landslide hazards: the disaster-inducing environment, triggering factors, and signs of change. The disaster-inducing environment includes geological structure, hydrological conditions, surface undulation, and surface cover; triggering factors include precipitation and soil freeze-thaw cycles; and signs of change include displacement and subsidence. The graph analyzes the characteristics and formation conditions of major geological hazards, screens and identifies key factors for the occurrence of geological hazards, and establishes an automatic geological hazard identification system using machine learning methods through disaster-causing factor analysis and calculation, enabling automatic or semi-automatic geological hazard extraction.

[0016] Furthermore, the specific process of detailed geological hazard investigation based on an airborne platform is as follows: using an unmanned aerial vehicle (UAV) platform, a laser scanner and a high-resolution optical camera are set up on the UAV platform to acquire ground three-dimensional coordinates and image data, realize the real-time three-dimensional changes of ground features and the reproduction of real morphological characteristics, calculate the vertical displacement, volume change, and profile before and after the change of the ground surface, filter the point cloud data of steep slopes and complex terrain with dense vegetation, and obtain a digital surface model after removing vegetation from LiDAR data. Combined with two-dimensional and three-dimensional interpretation, identify the potential geological hazards of easily occurring geological hazards such as collapsed loose deposits, ancient landslide deposits, debris flow fans, and earthquake-induced mountain cracks under vegetation cover, so as to realize a detailed investigation of geological hazard areas.

[0017] Furthermore, in step 2, the pre-processed remote sensing image data is input into a pre-trained machine model for feature comparison, and the corresponding images of areas with potential hazards are output. Then, the images of areas with potential hazards are re-input into another machine model trained based on the characteristics of the disaster body, and the disaster points are output and labeled.

[0018] Furthermore, the specific process of sampling the soft rock area in step 3 is as follows: Before construction on the soft rock mass, several samples are drilled and identified using a camera device. The camera device consists of four cameras, a circular ring, two sample fixing ends, a telescopic cylinder, and a connecting rod. The two sample fixing ends are located at both ends of the connecting rod to fix the two ends of the sample. The circular ring surrounds the outside of the sample and can slide on the connecting rod. The four cameras are installed on the side of the circular ring, facing the east, south, west, and north directions of the sample, and are positioned to align with the sample. The telescopic cylinder's telescopic end is connected to the circular ring, driving the ring to reciprocate, thus enabling the four cameras to identify the sample's periphery during movement.

[0019] Samples are extracted using a circular drilling method. If a sample breaks during drilling, it is bonded and repaired with epoxy resin. Subsequently, the sample is identified by four cameras, and the images captured by the four cameras are stitched together to obtain video data of the sample's periphery. Next, image recognition processing is performed on the video data to identify the layering of soft rock in the sample, the soft rock layer materials, and the distribution and size of pores. Finally, characteristic data related to the soft rock layer materials are automatically retrieved from the Internet.

[0020] Further, the specific process of establishing the three-dimensional spatial model of the soft rock mass in step 3 is as follows: A three-dimensional spatial model is constructed according to the required slope dimensions and proportions. The three-dimensional peripheral images of all samples are placed into the three-dimensional spatial model. Based on the different sampling locations of each sample, the samples are placed at corresponding proportional positions within the spatial model. In the three-dimensional spatial model, points at the same depth for each sample are placed at the same height. The soft rock model inside the pillars of each sample is identified, and a three-dimensional model of the soft rock for each sample is obtained through finite element analysis. In the three-dimensional spatial model, points of the same nature or the same rock layer at the same height are connected by lines, and different line types or colors are used to distinguish points at different heights to obtain the image color and void distribution in the samples. The connecting lines at the same height or of the same nature are rendered as layers, moving outwards from the sample center until the entire three-dimensional spatial model is rendered, thus obtaining the three-dimensional model of the soft rock mass.

[0021] Furthermore, in step 4, a detection device is installed on all the samples taken out, and the samples are put back into the soft rock mass for real-time detection. Multiple humidity sensors, stress sensors and displacement sensors are installed on the side of the sampled samples, and the transmission line is connected to the outside. An external acquisition device is set up to collect data in real time and transmit it back to the data processing system. In the initial acquisition stage, the data collected represents the initial state of the soft rock mass, that is, its stable state data.

[0022] During data collection, if the data collected is collected in an area where no potential hazards were found during the remote sensing survey, and the collected data does not exceed the set value within the set time period, the collection rate will be reduced. The adaptive frequency conversion intelligent collection method adopts a low-frequency collection mode to avoid the accumulation of a large amount of redundant data.

[0023] Before construction, internal data of the soft rock mass is collected, and the self-weight of the soft rock is calculated based on its material properties, including hardness and density. Humidity data is detected by a humidity sensor, and the pore structure is identified to determine the structural stress of the soft rock in its initial state. The self-weight and structural stress are superimposed to obtain the initial stress state of the soft rock mass. Subsequently, the relevant data of the initial stress state are fused with the three-dimensional model of the soft rock mass, so that the stress data of the three-dimensional model of the soft rock mass can be updated in real time according to the changes in the detection data. Users can observe the stress status in the three-dimensional model of the soft rock mass in real time through the display screen.

[0024] Further, in step 4, based on the material properties of the soft rock at different depths, a strength reduction method is used to gradually reduce the shear strength parameters of the material, including cohesion and internal friction angle, until the slope reaches a critical state, i.e., failure occurs. This is the slope instability condition. According to this instability condition, instability failure boundary conditions are applied, and the three-dimensional model of the soft rock mass is backed up as a second three-dimensional model of the soft rock mass. Failure conditions are simulated in the second three-dimensional model of the soft rock mass, causing the soft rock mass to gradually approach the critical state, thereby determining the area where instability occurs fastest and marking the key areas of instability. Finally, the simulated key areas of instability are integrated into the original three-dimensional model of the soft rock mass to form a region-labeled three-dimensional model of the soft rock mass.

[0025] Furthermore, in step 5, remote sensing data and collected data need to be considered simultaneously. When one of them shows a potential hazard, the other needs to conduct a detailed verification of the corresponding area. When remote sensing data indicates a potential hazard, and data changes are detected in the soft rock area on-site, it is determined that the corresponding area is at risk of landslide or instability. The relevant personnel are notified to monitor the corresponding area. When remote sensing shows no potential hazard, but on-site soft rock detection shows a potential hazard, it is determined that the soft rock area under construction has become unstable.

[0026] Among them, the online early warning system is used to display disaster early warning and monitoring data for the entire region, zoom in on each monitored area at any time, and generate corresponding three-dimensional dynamic models for specific soft rock areas. Managers can intuitively see the changes in the three-dimensional model from normal data to the entire unstable dynamic, and perform real-time analysis based on the changes in the three-dimensional model.

[0027] Furthermore, in step 6, if the instability data is due to increased humidity, then drilling and pumping are performed. The bottom hole is used for vacuum pumping, and the top hole is used to introduce nitrogen gas. After the humidity returns to the original level, both the pumping hole and the nitrogen gas inlet are sealed. If there is movement or increased stress, it is determined that it is caused by vibration or internal oxidation and loosening. In this case, anchor rods are used for reinforcement, and cement grout is injected into the anchor rods for reinforcement until the data collected again returns to the error range set in the initial data.

[0028] If both the remote sensing data and the real-time monitoring 3D dynamic model show potential risks, the corresponding area should be isolated, and the occurrence of potential risks should be monitored regularly and manual intervention blasting should be used to intervene in the time when the risks occur.

[0029] The present invention, by adopting the above-described technical solution, has the following beneficial effects:

[0030] This invention utilizes remote sensing data and real-time detection within soft rock masses to better monitor slope areas, ensuring no potential hazards are missed. It first screens a large spatial area, providing better detection points for subsequent accurate identification. Based on this pre-screening, it reduces the need for real-time data collection at each detection point, avoiding the accumulation of redundant data. Samples are collected from different points in the soft rock mass, and the samples are analyzed using graphic recognition and finite element analysis to obtain a three-dimensional model of the soft rock mass. All samples are combined, and a model of the entire soft rock mass is generated through recognition and rendering. The samples are then placed back into the soft rock mass, and data is collected in real-time. This collected data is then placed into the soft rock mass model, allowing for real-time observation of changes in the model and a more direct view of structural changes within the soft rock mass. The data is more accurate, and appropriate protective measures are implemented based on these changes. Attached Figure Description

[0031] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and preferred embodiments. However, it should be noted that many details listed in the specification are merely to provide the reader with a thorough understanding of one or more aspects of the present invention, and these aspects of the invention can be implemented even without these specific details.

[0033] like Figure 1 As shown, a three-dimensional intelligent monitoring and early disaster identification and warning method for soft rock slopes includes the following steps:

[0034] Step 1: Use remote sensing equipment to photograph the highway subgrade and acquire corresponding remote sensing image data. Remote sensing equipment includes space-based geological hazard surveys and airborne detailed geological hazard surveys. Fully utilize the advantages of space, air, and ground platforms to establish a geological hazard identification and risk assessment system that combines the strengths of all three.

[0035] The specific process of geological hazard survey based on a satellite-based platform is as follows: Using a satellite remote sensing platform, InSAR combined with optical remote sensing is employed. Optical remote sensing extracts information on landslide morphology and land cover classification, while InSAR extracts the deformation field. The combined data, including surface deformation rate, land cover classification, ground inversion temperature, and vegetation cover, are used to construct a geological hazard knowledge graph. This graph identifies the characteristics and relationships of three elements in landslide hazards: the disaster-inducing environment, triggering factors, and signs of change. The disaster-inducing environment includes geological structure, hydrological conditions, surface undulation, and surface cover; triggering factors include precipitation and soil freeze-thaw cycles; and signs of change include displacement and subsidence. The characteristics and formation conditions of major geological hazards are analyzed, and key factors for geological hazard occurrence are screened and identified. Through disaster-causing factor analysis and calculation, machine learning methods are used to establish an automatic geological hazard identification system, enabling automatic or semi-automatic geological hazard extraction.

[0036] The specific process of detailed geological hazard investigation based on an airborne platform is as follows: Using an unmanned aerial vehicle (UAV) platform, a laser scanner and a high-resolution optical camera are set up on the UAV platform to acquire three-dimensional coordinates and image data of the ground, realize the real-time three-dimensional changes of ground features and the reproduction of real morphological characteristics, calculate the vertical displacement, volume change, and profile before and after the change of the ground surface, filter the point cloud data of steep slopes and complex terrain with dense vegetation, and obtain a digital surface model after removing vegetation from LiDAR data. Combined with two-dimensional and three-dimensional interpretation, identify the potential geological hazards of prone geological disasters such as collapsed loose deposits, ancient landslide deposits, debris flow fans, and earthquake-induced mountain cracks under vegetation cover, so as to realize a detailed investigation of high-risk areas of geological disasters.

[0037] Step 2: Recognize the remote sensing images and input them into a machine learning model for feature comparison learning, conducting hazard screening and disaster identification. The pre-processed remote sensing image data is input into a pre-trained machine learning model for feature comparison, outputting images of areas with potential hazards. These images are then re-inputted into another machine learning model trained based on disaster characteristics, outputting and labeling the disaster points. The machine learning model uses a convolutional neural network, suitable for processing image data. This two-stage recognition reduces the amount of image processing data and makes feature recognition more targeted.

[0038] Step 3: Sample the soft rock area and establish a three-dimensional spatial model of the soft rock mass. Several samples are drilled from the soft rock mass before construction. A camera device is used for identification. The camera device consists of four cameras, a circular ring, two sample fixing ends, a telescopic cylinder, and a connecting rod. The two sample fixing ends are located at both ends of the connecting rod to fix the two ends of the sample. The circular ring surrounds the outside of the sample and can slide on the connecting rod. The four cameras are installed on the side of the circular ring, facing the east, south, west, and north directions of the sample, and are set to be aligned with the sample. The telescopic cylinder's telescopic end is connected to the circular ring, driving the circular ring to reciprocate, thus allowing the four cameras to identify the sample's perimeter during movement.

[0039] Samples are extracted using a circular drilling method. If a sample breaks during drilling, it is bonded and repaired with epoxy resin. Subsequently, the sample is identified by four cameras, and the images captured by the four cameras are stitched together to obtain video data of the sample's periphery. Next, image recognition processing is performed on the video data to identify the layering of soft rock in the sample, the soft rock layer materials, and the distribution and size of pores. Finally, characteristic data related to the soft rock layer materials are automatically retrieved from the Internet.

[0040] A three-dimensional spatial model was constructed according to the required slope dimensions and scale. The three-dimensional peripheral images of all samples were placed into this model, with each sample positioned at a corresponding proportional location within the model based on its sampling location. Within the three-dimensional model, points at the same depth for each sample were placed at the same height. The soft rock model inside the column for each sample was identified, and a three-dimensional model of the soft rock for each sample was obtained using finite element analysis. Points of the same properties or rock layers at the same height were connected by lines, using different line types or colors to distinguish points at different heights, thus obtaining the image color and void distribution within the sample. The connecting lines at the same height or of the same properties were rendered as layers, moving outwards from the sample center until the entire three-dimensional spatial model was rendered, thereby obtaining the three-dimensional model of the soft rock mass.

[0041] Step 4: Data from the interior of the soft rock mass is collected in real time using a data acquisition device. This data serves as the dynamic data for the three-dimensional spatial model of the soft rock mass. Detection devices are installed on all extracted samples. The samples are then placed back into the soft rock mass for real-time monitoring. Multiple humidity sensors, stress sensors, and displacement sensors are installed on the sides of the samples, and transmission lines are connected to an external data acquisition device to collect data in real time and transmit it back to the data processing system. In the initial acquisition phase, the collected data represents the initial state of the soft rock mass, i.e., its stable state data.

[0042] During data collection, if the data collected is collected in an area where no potential hazards were found during the remote sensing survey, and the collected data does not exceed the set value within the set time period, the collection rate will be reduced. The adaptive frequency conversion intelligent collection method adopts a low-frequency collection mode to avoid the accumulation of a large amount of redundant data.

[0043] Before construction, internal data of the soft rock mass is collected, and the self-weight of the soft rock is calculated based on its material properties, including hardness and density. Humidity data is detected by a humidity sensor, and the pore structure is identified to determine the structural stress of the soft rock in its initial state. The self-weight and structural stress are superimposed to obtain the initial stress state of the soft rock mass. Subsequently, the relevant data of the initial stress state are fused with the three-dimensional model of the soft rock mass, so that the stress data of the three-dimensional model of the soft rock mass can be updated in real time according to the changes in the detection data. Users can observe the stress status in the three-dimensional model of the soft rock mass in real time through the display screen.

[0044] Based on the material properties of soft rock at different depths, a strength reduction method is used to gradually decrease the shear strength parameters of the material, including cohesion and internal friction angle, until the slope reaches a critical state, i.e., failure occurs. This is the slope instability condition. According to this instability condition, instability failure boundary conditions are applied, and a second 3D model of the soft rock mass is created. Failure conditions are simulated in this second 3D model, gradually bringing the soft rock mass towards a critical state. This identifies the region where instability will occur fastest and marks the key instability areas. Finally, the simulated key instability areas are integrated into the original 3D soft rock mass model to form a region-labeled 3D soft rock mass model.

[0045] Step 5: Input remote sensing data and real-time acquired data into the online early warning system to determine if any soft rock areas are likely to become unstable. Both remote sensing and acquired data need to be considered simultaneously. If one indicates a potential hazard, the other requires detailed verification of the corresponding area. If remote sensing data indicates a hazard, and on-site soft rock detection shows changes in data, then the corresponding area is considered to have a landslide risk. Relevant personnel should be notified promptly to monitor the area. If both remote sensing and on-site soft rock detection indicate a hazard, then the soft rock area under construction is determined to be unstable.

[0046] Among them, the online early warning system is used to display disaster early warning and monitoring data for the entire region, zoom in on each monitored area at any time, and generate corresponding three-dimensional dynamic models for specific soft rock areas. Managers can intuitively see the changes in the three-dimensional model from normal data to the entire unstable dynamic, and perform real-time analysis based on the changes in the three-dimensional model.

[0047] Step 6: Issue early warning information for areas prone to instability or landslides, and notify management personnel to conduct on-site inspections and implement protective measures. Based on the retrieved data indicating potential instability, if the cause is increased humidity, drilling and pumping will be performed. Holes will be drilled at the bottom for vacuum pumping, and at the top for nitrogen gas injection. Once the humidity returns to its original level, both the pumping holes and the nitrogen gas inlet will be sealed. If movement or increased stress occurs, indicating vibration or internal oxidation causing loosening, anchor bolts will be used for reinforcement, and cement grout will be injected into the anchor bolts for reinforcement. This process will continue until the collected data returns to the initial error range.

[0048] If potential risks are detected in remote sensing data and real-time monitoring of the 3D dynamic model, the corresponding area should be isolated, and the occurrence of potential risks should be monitored regularly and manual intervention blasting should be used to intervene.

[0049] Effectively identify areas prone to geological instability or landslides, propose treatment plans for geological hazards at different landslide stages, and improve the ability of geological hazards to cope with complex working conditions. For soft rock geological hazards that may exceed the disaster warning value based on deformation monitoring data, conduct pre-reinforcement protection; for geological hazards that have already exceeded the disaster warning value, carry out emergency rescue and treatment. In the early stages, cost-effective small-scale rapid rescue methods such as anchor piles, steel pipe piles, or steel light rails are used for treatment; in the later stages, anchor retaining walls and anti-slide piles are combined for treatment, forming a comprehensive approach of prevention first, combined prevention and control, emergency rescue, and integrated management. Optimize the design of geological hazard retaining structures; combine stress and deformation data of anchor bolts, anti-slide piles, and other retaining structures obtained from on-site monitoring, conduct numerical simulation analysis, and optimize the design of retaining structure types, dimensions, and reinforcement under different working conditions, making geological hazard management more economical, safe, and environmentally friendly.

[0050] Matters not covered in this invention are common knowledge.

[0051] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A three-dimensional intelligent monitoring and early disaster identification and warning method for soft rock slopes, characterized by: The method includes the following steps: Step 1: Use remote sensing equipment to photograph the highway subgrade and obtain the corresponding remote sensing image data; Step 2: Recognize the remote sensing images and input them into the machine model for feature comparison and learning to screen for potential hazards and identify disaster bodies; Step 3: Sample the soft rock area and establish a three-dimensional spatial model of the soft rock mass; Step 4: Collect data from the interior of the soft rock mass in real time by setting up a data acquisition device. The collected data will be used as dynamic data for the three-dimensional spatial model of the soft rock mass. Step 5: Input the remote sensing data and real-time acquired data into the online early warning system to determine if there are any soft rock areas that may become unstable; Step 6: Issue early warning information for areas where instability or landslides may occur, and notify management personnel to conduct on-site inspections and take protective measures; The specific process of sampling the soft rock area in step 3 is as follows: Before construction of the soft rock mass, several samples are drilled and identified using a camera device. The camera device consists of four cameras, a ring, two sample fixing ends, a telescopic cylinder, and a connecting rod. The two sample fixing ends are located at both ends of the connecting rod and are used to fix the two ends of the sample. The ring surrounds the outside of the sample and can slide on the connecting rod. The four cameras are installed on the side of the ring, facing the east, south, west, and north directions of the sample, and are set to be aligned with the sample. The telescopic end of the telescopic cylinder is connected to the ring and drives the ring to move back and forth, so that the four cameras can identify the outer perimeter of the sample during the movement. Samples are extracted using a circular drilling method. If a sample breaks during drilling, it is bonded and repaired with epoxy resin. Then, the sample is identified by four cameras, and the images captured by the four cameras are stitched together to obtain video data of the sample's periphery. Next, image recognition processing is performed on the video data to identify the layering of soft rock in the sample, the soft rock layer materials, and the distribution and size of pores. Finally, characteristic data related to the soft rock layer materials are automatically retrieved from the Internet. The specific process of establishing a three-dimensional spatial model of soft rock mass in step 3 is as follows: A three-dimensional spatial model is constructed according to the required slope dimensions and proportions. The three-dimensional peripheral images of all samples are placed into the three-dimensional spatial model. Based on the different sampling locations of each sample, the samples are placed at corresponding proportional positions within the spatial model. In the three-dimensional spatial model, points at the same depth for each sample are placed at the same height. The soft rock model inside the pillars of each sample is identified, and a three-dimensional model of the soft rock mass for each sample is obtained through finite element analysis. In the three-dimensional spatial model, points of the same nature or the same rock layer at the same height are connected by lines, and different line types or colors are used to distinguish points at different heights to obtain the image color and void distribution in the samples. The connecting lines at the same height or of the same nature are rendered as layers, moving outwards from the sample center until the entire three-dimensional spatial model is rendered, thus obtaining the three-dimensional model of the soft rock mass.

2. The method for three-dimensional intelligent monitoring and early disaster identification and warning of soft rock slopes according to claim 1, characterized in that: Step 1 involves remote sensing equipment including a general geological hazard survey based on a spaceborne platform and a detailed geological hazard survey based on an airborne platform. The specific process of the general geological hazard survey based on the spaceborne platform is as follows: By combining InSAR and optical remote sensing methods using a satellite remote sensing platform, information on landslide morphology and land cover classification is extracted using optical remote sensing, while deformation field is extracted using InSAR. The combined data, including surface deformation rate, land cover classification, ground inversion temperature, and vegetation cover factors, are used to construct a geological hazard knowledge graph. This graph identifies the characteristics and relationships of three elements in landslide hazards: the disaster-inducing environment, triggering factors, and signs of change. The disaster-inducing environment includes geological structure, hydrological conditions, surface undulation, and surface cover; triggering factors include precipitation and soil freeze-thaw cycles; and signs of change include displacement and subsidence. The characteristics and formation conditions of major geological hazards are analyzed, and key factors for their occurrence are screened and identified. Through disaster-causing factor analysis and calculation, machine learning methods are employed to establish an automatic geological hazard identification system, enabling automatic or semi-automatic geological hazard extraction.

3. The method for three-dimensional intelligent monitoring and early disaster identification and warning of soft rock slopes according to claim 2, characterized in that: The specific process of detailed geological hazard investigation based on an airborne platform is as follows: Using an unmanned aerial vehicle (UAV) platform, a laser scanner and a high-resolution optical camera are set up on the UAV platform to acquire ground three-dimensional coordinates and image data, realize the real-time three-dimensional changes of ground features and the reproduction of real morphological characteristics, calculate the vertical displacement, volume change, and profile before and after the change of the ground surface, filter the point cloud data of steep slopes and complex terrain with dense vegetation, and obtain a digital surface model after removing vegetation from LiDAR data. Combined with two-dimensional and three-dimensional interpretation, potential geological hazards such as collapsed loose deposits, ancient landslide deposits, debris flow fans, and earthquake-induced cracks under vegetation cover are identified, thus realizing a detailed investigation of geological hazard areas.

4. The method for three-dimensional intelligent monitoring and early disaster identification and warning of soft rock slopes according to claim 1, characterized in that: In step 2, the pre-processed remote sensing image data is input into a pre-trained machine model for feature comparison, and the corresponding images of areas with potential hazards are output. Then, the images of areas with potential hazards are re-input into another machine model trained based on the characteristics of the disaster body, and the disaster points are output and labeled.

5. The method for three-dimensional intelligent monitoring and early disaster identification and warning of soft rock slopes according to claim 1, characterized in that: In step 4, a detection device is installed on all the samples taken out, and the samples are put back into the soft rock mass for real-time detection. Several humidity sensors, stress sensors and displacement sensors are installed on the side of the sampled samples, and the transmission line is connected to the outside. An external acquisition device is set up to collect data in real time and transmit it back to the data processing system. In the initial acquisition stage, the data collected represents the initial state of the soft rock mass, that is, its stable state data. During data collection, if the data collected is collected in an area with potential hazards not identified in the remote sensing survey, and the collected data does not exceed the set value within the set time period, the collection rate will be reduced, and an adaptive frequency conversion intelligent collection method will be adopted. A low-frequency collection mode will be used to avoid the accumulation of a large amount of redundant data. Before construction, internal data of the soft rock mass is collected, and the self-weight of the soft rock is calculated based on its material properties, including hardness and density. Humidity data is detected by a humidity sensor, and the pore structure is identified to determine the structural stress of the soft rock in its initial state. The self-weight and structural stress are superimposed to obtain the initial stress state of the soft rock mass. Subsequently, the relevant data of the initial stress state are fused with the three-dimensional model of the soft rock mass, so that the stress data of the three-dimensional model of the soft rock mass can be updated in real time according to the changes in the detection data. Users can observe the stress status in the three-dimensional model of the soft rock mass in real time through the display screen.

6. The method for three-dimensional intelligent monitoring and early disaster identification and warning of soft rock slopes according to claim 1, characterized in that: In step 4, based on the material properties of soft rock at different depths, a strength reduction method is used to gradually reduce the shear strength parameters of the material, including cohesion and internal friction angle, until the slope reaches a critical state, i.e., failure occurs. This is the instability condition of the slope. According to this instability condition, instability failure boundary conditions are applied, and the three-dimensional model of the soft rock mass is backed up as a second three-dimensional model of the soft rock mass. The failure conditions are simulated in the second three-dimensional model of the soft rock mass, so that the soft rock mass gradually approaches the critical state, thereby determining the area where instability occurs fastest and marking the key areas of instability. Finally, the simulated key areas of instability are integrated into the original three-dimensional model of the soft rock mass to form a region-labeled three-dimensional model of the soft rock mass.

7. The method for three-dimensional intelligent monitoring and early disaster identification and warning of soft rock slopes according to claim 1, characterized in that: In step 5, remote sensing data and collected data need to be considered simultaneously. When one of them shows a potential hazard, the other needs to be carefully verified in the corresponding area. When remote sensing data indicates a potential hazard, and data changes are detected in the soft rock area on-site, it is determined that the corresponding area is at risk of landslide or instability. The relevant personnel are notified to monitor the corresponding area. When remote sensing shows no potential hazard, but on-site soft rock mass detection shows a potential hazard, it is determined that the soft rock mass area under construction has become unstable. Among them, the online early warning system is used to display disaster early warning and monitoring data for the entire region, zoom in on each monitored area at any time, and generate corresponding three-dimensional dynamic models for specific soft rock areas. Managers can intuitively see the changes in the three-dimensional model from normal data to the entire unstable dynamic, and perform real-time analysis based on the changes in the three-dimensional model.

8. The method for three-dimensional intelligent monitoring and early disaster identification and warning of soft rock slopes according to claim 1, characterized in that: In step 6, if the instability data is due to increased humidity, then drilling and pumping are performed. The bottom hole is used for vacuum pumping, and the top hole is used to introduce nitrogen gas. After the humidity returns to the original level, both the pumping hole and the nitrogen gas inlet are sealed. If there is movement or increased stress, it is determined that it is caused by vibration or internal oxidation and loosening. In this case, anchor rods are used for reinforcement, and cement grout is injected into the anchor rods for reinforcement until the data collected again returns to the error range set in the initial data. If both the remote sensing data and the real-time monitoring 3D dynamic model show potential risks, the corresponding area should be isolated, and the occurrence time of potential risks should be observed and addressed using blasting intervention at regular intervals.

Images