High and steep rock slope dangerous rock mass identification method based on unmanned aerial vehicle thermal infrared image

Abstract

The invention discloses a high and steep rock slope dangerous rock mass identification method based on an unmanned aerial vehicle thermal infrared image. The method comprises the following steps: S1, generating a flight route; s2, collecting an infrared image and a visible light aerial photography image of the heat of the target area; s3, generating a three-dimensional geographic information model based on the visible light aerial photography image, determining a labeling standard of the three-dimensional geographic information model based on the infrared image, and obtaining a data set; s4, inputting the training set, and training by using a support vector machine algorithm to obtain a dangerous rock mass recognition model; and S5, inputting the verification set into the dangerous rock mass recognition model, performing result verification in combination with manual annotation, and outputting a visual recognition result. By the adoption of the high and steep rock slope dangerous rock mass recognition method based on the unmanned aerial vehicle thermal infrared image, the visible light image and the thermal infrared image are combined, the dangerous rock mass recognition result is generated through machine learning, and support can be provided for subsequent deeper disaster prevention and reduction work decision, construction planning and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of geological disaster monitoring and prevention, and in particular to a method for identifying dangerous rock masses on high and steep rock slopes based on thermal infrared images of unmanned aerial vehicles. Background Art

[0002] With the continuous increase in investment in infrastructure construction and the continuous optimization of the distribution pattern of large-scale infrastructure such as transportation, water conservancy, industry and civil construction, more and more infrastructure is being built in plateaus, mountains, hills and other areas with complex terrain. Among them, the widely distributed high and steep slopes are key areas for breeding geological disasters such as collapse and landslides. It has become a hot issue that needs to be urgently addressed to carry out intelligent identification and rapid investigation of potential risks and hidden dangers, reduce the cost of disaster prevention and mitigation, and ensure that large-scale infrastructure distributed in plateaus, mountains and hills can give full play to their social and economic benefits.

[0003] The stability of high and steep slopes has always been a core issue in the field of geological disaster prevention and control. Especially in the fields of transportation and water conservancy, its stability is directly related to the overall safety of the project and the safety of people's lives and property. High and steep slopes often become unstable due to natural factors such as earthquakes, heavy rains, temperature changes, or human factors such as blasting and mining, resulting in geological disasters such as collapse and landslides. The uneven internal structure of the rock mass, the development of cracks, and the relative sliding of the structural surface are often the direct causes of the above disasters. Therefore, it is particularly important to quickly identify and determine the potential geological hazards of high and steep slopes.

[0004] However, the traditional methods of identifying and troubleshooting dangerous rock masses have great limitations and still rely too much on manual inspections and ground instruments. Their shortcomings are prominently reflected in the following aspects:

[0005] 1. Low efficiency: Manual identification and investigation requires a lot of manpower and time, especially when facing large areas and complex terrain, the difficulty of detection increases significantly;

[0006] 2. Limited accuracy: Instrument measurements are mostly concentrated in local areas and cannot fully cover the entire slope, making it difficult to accurately reflect the overall stability of the slope;

[0007] 3. High risk: Manual inspection requires approaching the slope, especially the dangerous area, which poses certain life safety risks;

[0008] 4. Poor continuity: Manual inspection is usually difficult to carry out continuous work and cannot capture the changes in the slope at different time periods, especially the stress adjustment caused by temperature changes. Summary of the invention

[0009] The purpose of the present invention is to provide a method for identifying dangerous rock masses on high and steep rock slopes based on thermal infrared images of unmanned aerial vehicles to solve the above technical problems.

[0010] To achieve the above object, the present invention provides a method for identifying dangerous rock masses on steep rock slopes based on thermal infrared images of unmanned aerial vehicles, comprising the following steps:

[0011] S1. construct a preliminary three-dimensional point cloud model of the target area, and generate a flight route based on the constructed preliminary three-dimensional point cloud model;

[0012] S2, controlling the drone to fly according to the flight route generated in step S1, and during the flight, using the visible light-thermal infrared multimodal high-definition camera carried by the drone to collect thermal infrared images and visible light aerial photography images of the target area;

[0013] S3, preprocessing the collected infrared images and visible light aerial photography images, generating a three-dimensional geographic information model based on the visible light aerial photography images, and determining the annotation standard of the three-dimensional geographic information model based on the infrared images to obtain a data set;

[0014] S4, dividing the data set into a training set, a test set and a validation set, inputting the training set, and using the support vector machine algorithm for training to obtain a dangerous rock mass identification model, and then using the test set to optimize the parameters of the dangerous rock mass identification model;

[0015] S5. Input the verification set into the dangerous rock mass identification model, verify the results with manual annotation, and output the visual identification results.

[0016] Preferably, step S1 specifically includes the following steps:

[0017] S11. Conduct preliminary on-site survey of the target area and mark typical risky rock masses;

[0018] S12. Use a visible light-thermal infrared multimodal high-definition camera carried by a drone to collect visible light aerial images of the target area in accordance with the oblique photogrammetry method;

[0019] S13, based on the visible light aerial photography image acquired in step S12, using 3D reconstruction software to generate a preliminary 3D point cloud model of the target area;

[0020] S14, importing the preliminary 3D point cloud model generated in step S13 into the DPGO route planning software, selecting the drone, lens and aerial photography area used in the DPGO route planning software, generating a flight route according to the elevation flight mode, and setting the flight route parameters. According to the accuracy requirements of the survey mission, the distance between the drone and the plane obtained by fitting the boundary points of the aerial photography area during the acquisition process, the flight overlap rate and the flight speed are set;

[0021] S15. Check the safety of the route and simulate the flight process. After completion, export the flight route file in KML format.

[0022] Preferably, in step S15, a route safety threshold is set, that is, a minimum radius between the UAV and a point in the preliminary three-dimensional point cloud model, and then the flight is simulated.

[0023] Preferably, in step S2, the flight route file in KML format is first imported into the drone remote control terminal, and then the drone takes off after checking its working status. With RTK turned on, the drone flies according to the flight route, and during the flight, the visible light-thermal infrared multimodal high-definition camera carried by the drone is used to collect thermal infrared images and visible light aerial photography images of the target area, and finally the drone lands in an open area after the collection is completed.

[0024] Preferably, step S3 specifically includes the following steps:

[0025] S31, image preprocessing: remove abnormal images until the proportion of abnormal images to total images is less than 5%, otherwise perform supplementary sampling;

[0026] S32, importing the visible light aerial photography image into the three-dimensional reconstruction software, setting reconstruction parameters according to the geographic coordinate information of the aerial photography area and the result quality requirements, and generating a three-dimensional geographic information model of the target area;

[0027] S33, performing radiation correction on the thermal infrared image, and converting the corrected thermal infrared image into a grayscale image, and removing the surface vegetation therein through image processing;

[0028] S34, using a thermal infrared image processing tool, extracting the temperature difference and gradient at the interface between the typical risky rock mass and the slope surface where it is located in the infrared image processed in step S33, and determining the marking standard.

[0029] Preferably, the 3D reconstruction software described in step S13 and step S32 is DJI Terra, the preliminary 3D point cloud model described in step S13 includes RGB information and elevation information, and the 3D geographic information model described in step S32 includes a 3D real scene model, a dense point cloud model and an elevation model.

[0030] Preferably, step S33 specifically includes the following steps:

[0031] S331. The original digital value is converted into radiant brightness by calibrating the gain and bias value parameters of the thermal imaging sensor of the visible light-thermal infrared multimodal high-definition camera. The expression is as follows:

[0032] L=G×DN+B

[0033] Where L is the radiant brightness, unit is W / (m 2 *sr*μm); G is gain, B is offset value, DN is original digital value;

[0034] S332, collecting historical atmospheric parameters of the target area, and inputting the collected historical atmospheric parameters into the MODTRAN radiation transfer model to perform atmospheric correction to eliminate the absorption and scattering changes of the thermal infrared sensor receiving signal caused by gases and particles in the atmosphere;

[0035] S333. Use Planck's radiation law to convert the radiation brightness L into the surface temperature T. The calculation formula of the processing process is as follows:

[0036]

[0037] Where K1 and K2 are the thermal infrared sensor radiation calibration constants, L s is the surface radiance after atmospheric correction;

[0038] S334. Use the empirical formula to correct the surface emissivity. The expression is as follows:

[0039]

[0040] Where, ∈ is the surface emissivity; T Ture is the corrected temperature;

[0041] S335. Verify with the help of ground measured data or reference images to evaluate the accuracy of the corrected image. If there is a deviation, adjust the atmospheric parameters or surface emissivity according to the ground measured data or reference images, and then return to step S333.

[0042] Preferably, in step S4, the three-dimensional geographic information model is evenly divided into multiple sub-areas, and the multiple sub-areas are grouped into training set, test set and validation set in a ratio of 3:4:3, and the training set and test set are input in sequence for training and testing.

[0043] Preferably, in step S5, the verification set is input into the dangerous rock body identification model for dangerous rock body identification, and the dangerous rock body identification results are compared and verified with the manual annotation results, and the verified dangerous rock bodies are annotated into the three-dimensional geographic information model obtained in step S3, and the visual identification results are output.

[0044] Therefore, the present invention adopts the above-mentioned high steep rock slope dangerous rock mass identification method based on UAV thermal infrared image, which has the following beneficial effects:

[0045] 1. Combining visible light images with thermal infrared images and then using machine learning to generate recognition results. Compared with traditional methods, the data sources are richer and the results are more reliable. This can significantly reduce the risk of manual investigation and is suitable for monitoring large-scale, high-risk geological disaster areas;

[0046] 2. An optimized and improved machine learning algorithm is used to introduce the difference in thermal infrared temperature gradient into the identification of dangerous rock masses. Through training, more refined identification of dangerous rock masses can be achieved, eliminating repeated redundant steps and improving overall accuracy and work efficiency.

[0047] 3. The dangerous rock mass identification results are visualized through other synchronously generated geographic information products, making the identification work more intuitive and providing support for subsequent more in-depth disaster prevention and mitigation work decisions and construction plans.

[0048] In summary, the present invention can automatically identify dangerous rock masses and visualize the identification results, which is suitable for rapid identification and investigation of geological disasters such as dangerous rock masses in complex and dangerous environments in plateau mountainous areas, and provides a basis for engineering planning, disaster prevention and mitigation, and emergency management in infrastructure construction fields such as water conservancy and transportation.

[0049] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a flowchart of a method for identifying dangerous rock masses on steep rock slopes based on thermal infrared images of unmanned aerial vehicles in the present invention;

[0051] Figure 2 This is an example diagram of a thermal infrared image after radiation calibration according to the present invention;

[0052] Figure 3 An example diagram of the contour of the dangerous rock mass identified and marked according to the present invention;

[0053] Figure 4 This is an example diagram of the dangerous rock mass visual identification result described in the present invention. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical scheme and advantages disclosed in the embodiments of the present invention clearer, the embodiments of the present invention are further described in detail in combination with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the embodiments of the present invention and are not used to limit the embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application. Examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions.

[0055] It should be noted that the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or inherent to these processes, methods, products or devices.

[0056] The embodiments of the present invention are described in detail below in conjunction with the accompanying drawings.

[0057] like Figure 1 As shown in the figure, the method for identifying dangerous rock masses on steep rock slopes based on thermal infrared images of unmanned aerial vehicles includes the following steps:

[0058] S1. construct a preliminary three-dimensional point cloud model of the target area, and generate a flight route based on the constructed preliminary three-dimensional point cloud model;

[0059] Step S1 specifically includes the following steps:

[0060] S11. Conduct preliminary on-site survey of the target area and mark typical risky rock masses;

[0061] S12. Use a visible light-thermal infrared multimodal high-definition camera carried by a drone to collect visible light aerial images of the target area in accordance with the oblique photogrammetry method;

[0062] S13, based on the visible light aerial photography image acquired in step S12, using 3D reconstruction software to generate a preliminary 3D point cloud model of the target area;

[0063] S14, import the preliminary three-dimensional point cloud model generated in step S13 into the DPGO route planning software, select the drone, lens and aerial photography area used in the DPGO route planning software, generate the flight route according to the vertical flight mode, set the flight route parameters at the same time, and set the distance, flight overlap rate and flight speed of the plane obtained by fitting the drone and the boundary points of the aerial photography area during the collection process according to the accuracy requirements of the survey task; wherein, the accuracy is inversely proportional to the distance, the higher the accuracy, the closer the distance, that is, the longer the flight time; the flight overlap rate is inversely proportional to the work efficiency, and the greater the flight overlap rate, the more redundant data is collected. Taking into account the task accuracy and task cycle, the distance of this embodiment is set to 10m-20m, the flight overlap rate is set to 70%-80%, and the flight speed is set to not exceed the maximum flight speed.

[0064] S15. Check the safety of the route and simulate the flight process. After completion, export the flight route file in KML format.

[0065] In step S15, a route safety threshold is set, that is, the minimum radius between the UAV and the midpoint of the preliminary three-dimensional point cloud model, to avoid unexpected situations such as collisions during image acquisition, and then simulate the flight.

[0066] S2, controlling the drone to fly according to the flight route generated in step S1, and during the flight, using the visible light-thermal infrared multimodal high-definition camera carried by the drone to collect thermal infrared images and visible light aerial photography images of the target area;

[0067] In step S2, the flight route file in KML format is first imported into the drone remote control terminal, and then the drone takes off after checking its working status. With RTK (real-time kinematic) turned on to ensure the working accuracy is within the centimeter range, the drone flies according to the flight route. During the flight, the visible light-thermal infrared multimodal high-definition camera carried by the drone is used to collect thermal infrared images and visible light aerial photography images of the target area. Finally, the drone lands in an open area after the collection is completed.

[0068] S3, preprocessing the collected infrared images and visible light aerial photography images, generating a three-dimensional geographic information model based on the visible light aerial photography images, and determining the annotation standard of the three-dimensional geographic information model based on the infrared images to obtain a data set;

[0069] Step S3 specifically includes the following steps:

[0070] S31, image preprocessing: remove abnormal images until the proportion of abnormal images to total images is less than 5%, otherwise perform supplementary sampling;

[0071] S32, importing the visible light aerial photography image into the three-dimensional reconstruction software, setting reconstruction parameters according to the geographic coordinate information of the aerial photography area and the result quality requirements, and generating a three-dimensional geographic information model of the target area;

[0072] The 3D reconstruction software described in step S13 and step S32 is DJI Terra. The preliminary 3D point cloud model described in step S13 includes RGB information and elevation information. The 3D geographic information model described in step S32 includes a 3D real scene model, a dense point cloud model and an elevation model.

[0073] S33. Perform radiation correction on the thermal infrared image to remove the influence of the instrument, environment and atmosphere, ensure that the image data can accurately reflect the thermal radiation characteristics of the ground objects, and convert the corrected thermal infrared image into a grayscale image, while removing the surface vegetation through image processing;

[0074] S34. Since the temperature of the typical dangerous rock masses manually identified and marked in the previous investigation is significantly higher than that of the slope where they are located, that is, there is an obvious temperature difference at the junction of the dangerous rock mass and the slope surface where it is located, which can clearly show the outline of the dangerous rock mass at the moment of imaging, the thermal infrared image processing tool is used to extract the temperature difference and gradient at the junction of the typical dangerous rock mass and the slope surface where it is located in the infrared image processed by step S33, and determine the marking standard.

[0075] Step S33 specifically includes the following steps:

[0076] S331. The original digital value is converted into radiant brightness by calibrating the gain and bias value parameters of the thermal imaging sensor of the visible light-thermal infrared multimodal high-definition camera. The expression is as follows:

[0077] L=G×DN+B

[0078] Where L is the radiant brightness, unit is W / (m 2 *sr*μm); G is gain, B is offset value, DN is original digital value;

[0079] S332, collecting historical atmospheric parameters of the target area, and inputting the collected historical atmospheric parameters into the MODTRAN radiation transfer model to perform atmospheric correction to eliminate the absorption and scattering changes of the thermal infrared sensor receiving signal caused by gases and particles in the atmosphere;

[0080] S333. Use Planck's radiation law to convert the radiation brightness L into the surface temperature T. The calculation formula of the processing process is as follows:

[0081]

[0082] Where K1 and K2 are the thermal infrared sensor radiation calibration constants, L s is the surface radiance after atmospheric correction;

[0083] S334. Use the empirical formula to correct the surface emissivity. The expression is as follows:

[0084]

[0085] Where, ∈ is the surface emissivity, which can be obtained by looking up a table or measuring on site; T Ture is the corrected temperature;

[0086] S335. Verify with the help of ground measured data or reference images to evaluate the accuracy of the corrected image. If there is a deviation, adjust the atmospheric parameters or surface emissivity according to the ground measured data or reference images, and then return to step S333.

[0087] S4, dividing the data set into a training set, a test set and a validation set, inputting the training set, and using the support vector machine algorithm for training to obtain a dangerous rock mass identification model, and then using the test set to optimize the parameters of the dangerous rock mass identification model;

[0088] In step S4, the three-dimensional geographic information model is evenly divided into multiple sub-areas, and the multiple sub-areas are grouped into a training set, a test set, and a validation set in a ratio of 3:4:3. The training set and the test set are input in sequence for training and testing.

[0089] S5. Input the verification set into the dangerous rock mass identification model, verify the results with manual annotation, and output the visual identification results.

[0090] In step S5, the verification set is input into the dangerous rock mass identification model for dangerous rock mass identification (identify and mark the edge contours of all potential dangerous rock masses that meet the annotation standards to achieve the division of dangerous rock masses), and the dangerous rock mass identification results are compared and verified with the manual annotation results. The verified dangerous rock masses are annotated to the three-dimensional geographic information model obtained in step S3. The marking on the three-dimensional real-scene model can intuitively present the spatial distribution characteristics of the dangerous rock masses. The marking on the dense point cloud model and the elevation model can provide data support for the subsequent dangerous rock mass segmentation, stability assessment and destruction process deduction, provide a basis for engineering planning, disaster prevention and mitigation, and output visual identification results.

[0091] Example: The method for identifying dangerous rock masses on steep rock slopes based on thermal infrared images of unmanned aerial vehicles of the present invention is used to collect and process images of a mountain. The results obtained are as follows: Figure 2-Figure 4 As shown, the effectiveness of the present invention is confirmed.

[0092] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for identifying dangerous rock masses on steep rock slopes based on thermal infrared images from unmanned aerial vehicles, characterized in that: The following steps are involved: S1. construct a preliminary three-dimensional point cloud model of the target area, and generate a flight route based on the constructed preliminary three-dimensional point cloud model; S2, controlling the drone to fly according to the flight route generated in step S1, and during the flight, using the visible light-thermal infrared multimodal high-definition camera carried by the drone to collect thermal infrared images and visible light aerial photography images of the target area; S3, preprocessing the collected infrared images and visible light aerial photography images, generating a three-dimensional geographic information model based on the visible light aerial photography images, and determining the annotation standard of the three-dimensional geographic information model based on the infrared images to obtain a data set; S4, dividing the data set into a training set, a test set and a validation set, inputting the training set, and using the support vector machine algorithm for training to obtain a dangerous rock mass identification model, and then using the test set to optimize the parameters of the dangerous rock mass identification model; S5. Input the verification set into the dangerous rock mass identification model, verify the results with manual annotation, and output the visual identification results.

2. The method for identifying dangerous rock masses on steep rock slopes based on thermal infrared images of unmanned aerial vehicles according to claim 1 is characterized by: Step S1 specifically includes the following steps: S11. Conduct preliminary on-site survey of the target area and mark typical risky rock masses; S12. Use a visible light-thermal infrared multimodal high-definition camera carried by a drone to collect visible light aerial images of the target area in accordance with the oblique photogrammetry method; S13, based on the visible light aerial photography image acquired in step S12, using 3D reconstruction software to generate a preliminary 3D point cloud model of the target area; S14, importing the preliminary 3D point cloud model generated in step S13 into the DPGO route planning software, selecting the drone, lens and aerial photography area used in the DPGO route planning software, generating a flight route according to the elevation flight mode, and setting the flight route parameters. According to the accuracy requirements of the survey mission, the distance between the drone and the plane obtained by fitting the boundary points of the aerial photography area during the acquisition process, the flight overlap rate and the flight speed are set; S15. Check the safety of the route and simulate the flight process. After completion, export the flight route file in KML format.

3. The method for identifying dangerous rock masses on steep rock slopes based on thermal infrared images of unmanned aerial vehicles according to claim 2 is characterized by: In step S15, a route safety threshold is set, that is, the minimum radius between the UAV and the midpoint of the preliminary three-dimensional point cloud model, and then the flight is simulated.

4. The method for identifying dangerous rock masses on steep rock slopes based on thermal infrared images of unmanned aerial vehicles according to claim 3 is characterized by: In step S2, the flight route file in KML format is first imported into the drone remote control terminal, and then the drone takes off after checking its working status. With RTK turned on, the drone flies according to the flight route. During the flight, the visible light-thermal infrared multimodal high-definition camera carried by the drone is used to collect thermal infrared images and visible light aerial photography images of the target area. Finally, the drone lands in an open area after the collection is completed.

5. The method for identifying dangerous rock masses on steep rock slopes based on thermal infrared images of unmanned aerial vehicles according to claim 4 is characterized by: Step S3 specifically includes the following steps: S31, image preprocessing: remove abnormal images until the proportion of abnormal images to total images is less than 5%, otherwise perform supplementary sampling; S32, importing the visible light aerial photography image into the three-dimensional reconstruction software, setting reconstruction parameters according to the geographic coordinate information of the aerial photography area and the result quality requirements, and generating a three-dimensional geographic information model of the target area; S33, performing radiation correction on the thermal infrared image, and converting the corrected thermal infrared image into a grayscale image, and removing the surface vegetation therein through image processing; S34, using a thermal infrared image processing tool, extracting the temperature difference and gradient at the interface between the typical risky rock mass and the slope surface where it is located in the infrared image processed in step S33, and determining the marking standard.

6. The method for identifying dangerous rock masses on steep rock slopes based on thermal infrared images of unmanned aerial vehicles according to claim 5 is characterized by: The 3D reconstruction software described in step S13 and step S32 is DJI Terra. The preliminary 3D point cloud model described in step S13 includes RGB information and elevation information. The 3D geographic information model described in step S32 includes a 3D real scene model, a dense point cloud model and an elevation model.

7. The method for identifying dangerous rock masses on steep rock slopes based on thermal infrared images of unmanned aerial vehicles according to claim 6 is characterized by: Step S33 specifically includes the following steps: S331. The original digital value is converted into radiant brightness by calibrating the gain and bias value parameters of the thermal imaging sensor of the visible light-thermal infrared multimodal high-definition camera. The expression is as follows: L=G×DN+B Where L is the radiant brightness, unit is W / (m 2 *sr*μm); G is gain, B is offset value, DN is original digital value; S332, collecting historical atmospheric parameters of the target area, and inputting the collected historical atmospheric parameters into the MODTRAN radiation transfer model to perform atmospheric correction to eliminate the absorption and scattering changes of the thermal infrared sensor receiving signal caused by gases and particles in the atmosphere; S333. Use Planck's radiation law to convert the radiation brightness L into the surface temperature T. The calculation formula of the processing process is as follows: Where K1 and K2 are the thermal infrared sensor radiation calibration constants, L s is the surface radiance after atmospheric correction; S334. Use the empirical formula to correct the surface emissivity. The expression is as follows: Where, ∈ is the surface emissivity; T Ture is the corrected temperature; S335. Verify with the help of ground measured data or reference images to evaluate the accuracy of the corrected image. If there is a deviation, adjust the atmospheric parameters or surface emissivity according to the ground measured data or reference images, and then return to step S333.

8. The method for identifying dangerous rock masses on steep rock slopes based on thermal infrared images of unmanned aerial vehicles according to claim 7 is characterized by: In step S4, the three-dimensional geographic information model is evenly divided into multiple sub-areas, and the multiple sub-areas are grouped into a training set, a test set, and a validation set in a ratio of 3:4:

3. The training set and the test set are input in sequence for training and testing.

9. The method for identifying dangerous rock masses on steep rock slopes based on thermal infrared images of unmanned aerial vehicles according to claim 8 is characterized by: In step S5, the verification set is input into the dangerous rock mass identification model for dangerous rock mass identification, and the dangerous rock mass identification results are compared and verified with the manual annotation results. The verified dangerous rock masses are annotated into the three-dimensional geographic information model obtained in step S3, and the visual identification results are output.

Images