Real-time monitoring system for slope deformation based on image recognition

Abstract

The application relates to the technical field of slope safety, and discloses a slope deformation real-time monitoring system based on image recognition, a collection unit is used for collecting slope image data, an acquisition unit is used for acquiring original images, a preprocessing unit is used for preprocessing the original images, an identification unit is used for identifying the edge features of the preprocessed images, a calculation unit is used for calculating horizontal deformation physical quantity XB1 and vertical deformation physical quantity XB2, an analysis unit is used for analyzing the slope deformation trend, a transmission unit is used for data transmission, and a power supply unit is used for providing electric energy; the application separates video encoding data streams through an FFmpeg library function and restores the video encoding data streams into one frame of original RGB image data; OpenCV library is used to perform gray scale processing and Gaussian filter processing on each frame of RGB image data; feature identification and displacement calculation are further performed; automatic extraction of the deformation quantity is realized; and automatic processing of the whole process is realized.

Description

Technical Field

[0001] This invention relates to the field of slope safety technology, and more specifically to a real-time slope deformation monitoring system based on image recognition. Background Technology

[0002] Slope deformation is an engineering geological term referring to the destructive phenomenon of slopes caused by differences in lithology, strata, geological structure, and soil structure, as well as the influence of surface water and groundwater. Its manifestations include spalling, landslides, collapses, and subsidence. The deformation mechanism involves failure modes such as strain softening of loess slopes. In engineering fields such as water conservancy, transportation, and mining, slope deformation is a major safety hazard. Its deformation mechanism is complex, with water level fluctuations, rainfall, and load changes being core contributing factors that easily lead to slope instability. Current slope deformation monitoring technologies have significant shortcomings.

[0003] Traditional contact-based equipment relies on manual labor, which is inefficient, prone to creating blind spots in monitoring, and cannot be automatically processed after deformation data collection, making the processing cumbersome. Drilling and installing inclinometers can damage the slope and is susceptible to power supply issues, thus making it impossible to operate unattended for extended periods, thereby increasing labor intensity. When issuing deformation warnings, the accuracy of the warnings is low, and it is impossible to obtain image data information in a timely manner. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present invention provide a real-time slope deformation monitoring system based on image recognition to solve the technical problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a real-time slope deformation monitoring system based on image recognition, comprising a data acquisition unit, an acquisition unit, a preprocessing unit, a recognition unit, a calculation unit, an extraction unit, an analysis unit, a display unit, a transmission unit, and a power supply unit. The data acquisition unit is used for acquiring slope image data; the acquisition unit acquires raw images; the preprocessing unit preprocesses the raw images; the recognition unit performs edge feature recognition on the preprocessed images; the calculation unit calculates the horizontal deformation physical quantity XB1 and the vertical deformation physical quantity XB2; ​​the extraction unit extracts images; the analysis unit analyzes the slope deformation trend; the display unit displays the slope deformation images; the transmission unit transmits data; and the power supply unit provides electrical energy.

[0006] When the recognition unit performs image recognition, it calls the OpenCV library to locate the target ruler in the Gaussian-filtered image. Based on the set virtual target ruler position, it locates the actual target ruler area. The Canny edge detection algorithm is applied to the located target ruler area to extract the clear edges of the scale lines on the target ruler, thus completing edge feature recognition. The recognition unit sends the image after edge feature recognition to the computing unit. When the recognition unit sends the image, it uses libx264 to compress the image data into an H.264 video stream.

[0007] In a preferred embodiment, the acquisition unit uses a high-speed network camera to acquire slope images. The high-speed network camera uses a high-frequency image capture technology of 15 frames per second to acquire image data, and the acquisition unit sends the acquired image data to the acquisition unit.

[0008] In a preferred embodiment, the acquisition unit calls FFmpeg library functions to connect to the RTSP video stream address provided by the network high-speed camera. FFmpeg separates the video encoded data stream from the received network data packets and restores the compressed video data into frames of original RGB image data to achieve image acquisition. The acquisition unit then sends the acquired RGB image data to the preprocessing unit.

[0009] In a preferred embodiment, the preprocessing unit uses the OpenCV library to perform grayscale processing on each frame of RGB image data. The grayscale processing formula is HD=0.3R+0.58G+0.12B, where HD is the grayscale value after processing, R is the red image, G is the green image, and B is the blue image, and R, G, and B are all within the grayscale value range of 0-255. The preprocessing unit then performs Gaussian filtering on the grayscale processed image.

[0010] In a preferred embodiment, the computing unit receives image data sent by the recognition unit, compares the pixel coordinates of the same feature point in the current frame with the initial reference frame, calculates the pixel-level displacement, establishes a pixel-to-actual-millimeters conversion scale based on the known physical scale spacing on the target ruler, and multiplies the pixel displacement by the scale to obtain the horizontal deformation physical quantity XB1 and the vertical deformation physical quantity XB2 of the slope.

[0011] In a preferred embodiment, the extraction unit receives the horizontal deformation physical quantity XB1 and the vertical deformation physical quantity XB2 calculated by the calculation unit and compares them with the internal horizontal threshold Y1 and vertical threshold Y2. When the horizontal deformation physical quantity XB1 > the horizontal threshold Y1, the extraction unit retrieves the images of the three frames when the horizontal deformation physical quantity XB1 appears, before its appearance, and after its appearance from the acquisition unit, and sends the horizontal deformation physical quantity XB1 to the analysis unit and the display unit to issue an early warning. When the horizontal deformation physical quantity XB1 ≤ the horizontal threshold Y1, the extraction unit is in a standby state.

[0012] In a preferred embodiment, when the vertical deformation physical quantity XB2 in the extraction unit is greater than the vertical threshold Y2, the extraction unit retrieves three frames of images from the acquisition unit—the frames when XB2 appears, before it appears, and after it appears—as well as the vertical deformation physical quantity XB2 itself, and sends them to the analysis unit and the display unit to issue an early warning. When the vertical deformation physical quantity XB2 is less than or equal to the vertical threshold Y2, the extraction unit is in standby mode, and all images sent by the extraction unit are timestamped.

[0013] In a preferred embodiment, the analysis unit receives all the data sent by the extraction unit and performs trend analysis. The analysis unit calculates the displacement change per unit time to obtain the rate of change BS and calculates the change of the rate of change per unit time to obtain the acceleration JS. When the rate of change BS approaches zero, the slope is in a stable state. When the rate of change BS converges to a fixed value and the acceleration JS approaches zero, the slope is in a uniform deformation state. When the rate of change BS increases and the acceleration JS is positive, the slope is in an accelerated deformation state. When the rate of change BS decreases and the acceleration JS is negative, the slope is in a decelerated deformation state.

[0014] In a preferred embodiment, the display unit receives the data sent by the extraction unit and displays the image data. The acquisition unit, preprocessing unit, recognition unit, calculation unit, and extraction unit are all located in a micro industrial computer. The transmission unit performs 4G communication network data transmission through a 4G router.

[0015] In a preferred embodiment, the power supply unit includes a battery, a solar charge controller, and a solar panel. The solar panel converts solar energy into electrical energy, the battery stores electrical energy and provides power for operation, and the solar charge controller controls the flow of electrical energy between the solar panel and the battery.

[0016] The technical effects and advantages of this invention are as follows:

[0017] 1. This invention uses FFmpeg library functions to separate the video encoded data stream and then restore it into frames of original RGB image data. The OpenCV library is used to perform grayscale processing and Gaussian filtering on each frame of RGB image data, followed by feature recognition and displacement calculation to achieve automatic extraction of deformation. When handling alarms or storage needs, the libx264 library is used to perform efficient compression encoding on key video frames to achieve fully automated processing.

[0018] 2. This invention uses a target ruler as the basis for deformation monitoring, eliminating the need for drilling and damaging the slope structure. This reduces deployment costs while ensuring the integrity of monitoring coverage. Furthermore, it uses solar power to solve the problems of remote slopes lacking stable mains power, high costs of traditional power supply, and short battery life, enabling long-term unattended operation and reducing labor intensity.

[0019] 3. This invention uses horizontal deformation physical quantity XB1 and vertical deformation physical quantity XB2, and compares them with their respective thresholds to ensure the accuracy of deformation monitoring and early warning. In addition, it adopts high-frequency image capture technology of 15 frames per second, so that image data can be obtained in time when jade crystals appear, which is convenient for analysis and thus improves the accuracy of slope deformation analysis. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the overall system composition of the present invention. Detailed Implementation

[0021] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. In addition, the forms of the various structures described in the following embodiments are merely illustrative. The real-time slope deformation monitoring system based on image recognition involved in the present invention is not limited to the structures described in the following embodiments. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Example 1

[0023] Reference Figure 1This invention provides a real-time slope deformation monitoring system based on image recognition, comprising a data acquisition unit, an acquisition unit, a preprocessing unit, a recognition unit, a calculation unit, an extraction unit, an analysis unit, a display unit, a transmission unit, and a power supply unit. The data acquisition unit is used for acquiring slope image data; the acquisition unit acquires raw images; the preprocessing unit preprocesses the raw images; the recognition unit performs edge feature recognition on the preprocessed images; the calculation unit calculates the horizontal deformation physical quantity XB1 and the vertical deformation physical quantity XB2; ​​the extraction unit extracts images; the analysis unit analyzes the slope deformation trend; the display unit displays the slope deformation images; the transmission unit transmits data; and the power supply unit provides electrical energy.

[0024] In this embodiment, the present application can automatically perform tasks such as image acquisition, image recognition, and slope deformation calculation and analysis, thereby achieving fully automatic slope deformation monitoring, reducing labor intensity and improving deformation monitoring accuracy.

[0025] Reference Figure 1 When the recognition unit performs image recognition, it calls the OpenCV library to locate the target ruler in the image after Gaussian filtering. Based on the set virtual target ruler position, it locates the actual target ruler area. The Canny edge detection algorithm is applied to the located target ruler area to extract the clear edges of the scale lines on the target ruler, thus completing edge feature recognition. The recognition unit sends the image after edge feature recognition to the computing unit. When the recognition unit sends the image, it uses libx264 to compress the image data into an H.264 video stream.

[0026] In this embodiment, image recognition is performed using the computer vision library OpenCV and the video coding library libx264. The OpenCV library can accurately identify edges, thereby improving the overall deformation monitoring accuracy. libx264 compresses image data into H.264 video streams, achieving a compression ratio of several hundred times, greatly saving storage space and backhaul bandwidth, and thus improving system reliability and processing efficiency.

[0027] Reference Figure 1 The acquisition unit uses a high-speed network camera to acquire slope images. The high-speed network camera uses a high-frequency image capture technology of 15 frames per second to acquire image data. The acquisition unit sends the acquired image data to the acquisition unit.

[0028] In this embodiment, the acquisition frequency is 15 frames per second to ensure the continuity and high frequency of image data. The acquisition frequency of 15 frames per second can capture small deformations of the slope, overcoming the shortcomings of traditional monitoring with low frequency and high risk of missed detection. High-speed cameras usually have better optical performance, which is beneficial to subsequent image processing and recognition.

[0029] Reference Figure 1 The acquisition unit calls FFmpeg library functions to connect to the RTSP video stream address provided by the high-speed network camera. FFmpeg separates the video encoded data stream from the received network data packets and restores the compressed video data into frames of original RGB image data to achieve image acquisition. The acquisition unit sends the acquired RGB image data to the preprocessing unit. The preprocessing unit uses the OpenCV library to perform grayscale processing on each frame of RGB image data. The grayscale processing formula is HD=0.3R+0.58G+0.12B, where HD is the grayscale value after processing, R is the red image, G is the green image, and B is the blue image, and R, G, and B are all within the grayscale value range of 0-255. The preprocessing unit then performs Gaussian filtering on the grayscale processed image.

[0030] In this embodiment, the acquisition unit connects to the RTSP video stream through the FFmpeg library to achieve video data decoding and image extraction. FFmpeg has efficient multimedia processing capabilities, ensuring the real-time performance of image extraction. Grayscale conversion reduces the amount of data and improves processing speed. Gaussian filtering effectively smooths image noise and improves the accuracy of subsequent edge detection and recognition.

[0031] Reference Figure 1 The calculation unit receives the image data sent by the recognition unit, compares the pixel coordinates of the same feature point in the current frame with the initial reference frame, calculates the pixel-level displacement, establishes a pixel-to-actual-millimeters conversion scale based on the known physical scale spacing on the target ruler, and multiplies the pixel displacement by the scale to obtain the physical deformation XB of the slope.

[0032] In this embodiment, the calculation unit converts pixel displacement with the target scale to obtain the physical quantities of horizontal and vertical deformation. Based on the known scale of the target scale, it performs pixel-to-actual length conversion to improve measurement accuracy. At the same time, it calculates horizontal and vertical deformation to comprehensively reflect the slope deformation state.

[0033] When calculating physical deformation, first calculate the horizontal and vertical movement distances H and C of the same feature point in the image coordinate system between the current frame and the reference frame. The formula for calculating the horizontal movement distance H is as follows: The formula for calculating the vertical movement distance C is: In the formula, Ht is the horizontal coordinate of the current frame pixel coordinate, H0 is the horizontal coordinate of the reference frame pixel coordinate, Ct is the vertical coordinate of the current frame pixel coordinate, and C0 is the vertical coordinate of the reference frame pixel coordinate. Then, the conversion scale S is calculated using the following formula: In the formula, WL is the known physical length between two graduations on the target ruler, XS is the pixel distance between the two graduation lines on the target ruler in the current image, and the calculation formulas for the horizontal deformation physical quantity XB1 and the vertical deformation physical quantity XB2 are respectively... and .

[0034] Reference Figure 1 The extraction unit receives the horizontal deformation physical quantity XB1 and the vertical deformation physical quantity XB2 calculated by the calculation unit and compares them with the internal horizontal threshold Y1 and vertical threshold Y2. When the horizontal deformation physical quantity XB1 > the horizontal threshold Y1, the extraction unit retrieves the images of the three frames when the horizontal deformation physical quantity XB1 appears, before its appearance, and after its appearance from the acquisition unit, and sends the horizontal deformation physical quantity XB1 to the analysis unit and the display unit to issue an early warning. When the horizontal deformation physical quantity XB1 ≤ the horizontal threshold Y1, the extraction unit is in a standby state. When the vertical deformation physical quantity XB2 in the extraction unit > the vertical threshold Y2, the extraction unit retrieves the images of the three frames when the vertical deformation physical quantity XB2 appears, before its appearance, and after its appearance from the acquisition unit, and sends the vertical deformation physical quantity XB2 to the analysis unit and the display unit to issue an early warning. When the vertical deformation physical quantity XB2 ≤ the vertical threshold Y2, the extraction unit is in a standby state. All images sent by the extraction unit are timestamped.

[0035] In this embodiment, the extraction unit determines whether to extract an image and trigger an early warning based on the comparison results of the horizontal deformation physical quantity XB1 and the vertical deformation physical quantity XB2 with the horizontal threshold Y1 and the vertical threshold Y2. The image is extracted only when the deformation exceeds the threshold, reducing the data storage and transmission burden. The extraction of three frames of images before, during and after the deformation facilitates post-event analysis and tracing. Furthermore, the separate judgment of horizontal and vertical deformation in this application improves the monitoring flexibility. The timestamp is added to the extracted images to facilitate time series analysis and event review.

[0036] Reference Figure 1 The analysis unit receives all the data sent by the extraction unit and performs trend analysis. The analysis unit calculates the displacement change per unit time to obtain the rate of change BS and the change in the rate of change per unit time to obtain the acceleration JS. When the rate of change BS approaches zero, the slope is in a stable state. When the rate of change BS converges to a fixed value and the acceleration JS approaches zero, the slope is in a state of uniform deformation. When the rate of change BS increases and the acceleration JS is positive, the slope is in a state of accelerated deformation. When the rate of change BS decreases and the acceleration JS is negative, the slope is in a state of decelerated deformation.

[0037] In this embodiment of the application, the method of trend analysis of deformation data by the analysis unit of this application determines the deformation state of the slope by calculating the rate of change BS and the acceleration JS. The rate of change BS is the current deformation of the slope, and the acceleration JS can reflect the deformation trend of the slope. Therefore, it not only monitors the deformation, but also analyzes its changing trend and provides early warning of potential risks. After analysis, the slope deformation is divided into four states: stable, uniform, accelerating, and decelerating, which is convenient for engineering judgment.

[0038] Reference Figure 1 The display unit receives data sent by the extraction unit and displays the image data. The acquisition unit, preprocessing unit, recognition unit, calculation unit, and extraction unit are all housed within a micro industrial computer. The transmission unit performs 4G communication network data transmission via a 4G router.

[0039] In this embodiment, some units are integrated into a micro industrial computer and data is transmitted via a 4G network. The micro industrial computer is small in size and has low power consumption, making it suitable for field deployment, while the 4G network has wide coverage and is suitable for slope sites without wired network environments.

[0040] Reference Figure 1 The power supply unit includes a storage battery, a solar charge controller, and a solar panel. The solar panel converts solar energy into electrical energy, the storage battery stores electrical energy and provides power for operation, and the solar charge controller controls the flow of electrical energy between the solar panel and the storage battery.

[0041] In this embodiment, the application adopts an independent power supply mode for the power supply unit, which completely solves the problems of no stable mains power supply on remote slopes, high cost of traditional power supply, and short battery life, and realizes long-term unattended operation.

[0042] Example 2

[0043] When performing actual deformation monitoring, this application requires logging in first. Open the provided URL using a modern browser. If you are not logged in, you will be automatically redirected to the login page. Please enter your username and password to log in and use the subsequent functions.

[0044] After logging in, configure the camera settings. If it is an uninitialized system, you need to configure and connect the camera on the first login. Please click Settings → Camera in the left menu, and fill in the camera's IP address, port number (8000 by default for Hikvision cameras), username, password, channel number (usually 1), channel name (customizable), and identification bitstream (values ​​are 0: main bitstream, 1: sub-bitstream, 3: third bitstream).

[0045] After setting up the camera, configure the target ruler. In the Settings → Target Ruler Settings interface, you can configure the target ruler's positioning and recognition. The process is as follows:

[0046] (1) Add a target ruler. Click the "+" button to add a target ruler. Generally, only one corresponding target ruler setting needs to be added for a physical target ruler. In order to improve the recognition accuracy and precision, if there is an extra-long (more than 2 meters) target ruler, it can be divided into multiple segments and multiple scale segments can be added for segment recognition.

[0047] (2) Target ruler basic information settings, which can set the target ruler name, basic water level, and scale value range (decimeter);

[0048] (3) Positioning area setting: Use the camera pan-tilt function below to move the camera to the middle of the image and adjust the zoom to ensure that all the scale areas to be identified appear in the image and reserve a certain space. Set the positioning area. Because the image effect is very different between day and night, daytime positioning needs to be set during the day and nighttime positioning needs to be set at night. Save again. Under normal circumstances, the positioning area only needs to select the upper half of the target ruler.

[0049] (4) Target ruler setting: Use the move and stretch tools to make the virtual target ruler coincide with the physical target ruler in the image as much as possible (numbers and scale).

[0050] (5) Save settings. After completing the settings adjustment, click the save button in the editing area to save the settings.

[0051] (6) After completing the above settings, enable the target ruler in the target ruler list on the right. The system will automatically start recognition within 1 minute.

[0052] After setting the target ruler, configure the central station. Go to Settings → Central Station and click the "+" icon to add a central station (up to 5). After adding a central station, edit its information and click the save button below. Once settings are complete, turn on the enable switch at the top of the corresponding central station to start automatic data reporting.

[0053] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. The units and algorithm steps of the various examples described in the embodiments can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0054] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0055] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0056] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A real-time monitoring system for slope deformation based on image recognition, characterized in that: The system includes a data acquisition unit, an acquisition unit, a preprocessing unit, a recognition unit, a calculation unit, an extraction unit, an analysis unit, a display unit, a transmission unit, and a power supply unit. The data acquisition unit is used for acquiring slope image data. The acquisition unit acquires raw images. The preprocessing unit preprocesses the raw images. The recognition unit performs edge feature recognition on the preprocessed images. The calculation unit calculates the horizontal deformation physical quantity XB1 and the vertical deformation physical quantity XB2. The extraction unit extracts images. The analysis unit analyzes the slope deformation trend. The display unit displays the slope deformation image. The transmission unit is used for data transmission. The power supply unit provides electrical energy. When the recognition unit performs image recognition, it calls the OpenCV library to locate the target ruler in the Gaussian-filtered image. Based on the set virtual target ruler position, it locates the actual target ruler area. The Canny edge detection algorithm is applied to the located target ruler area to extract the clear edges of the scale lines on the target ruler, thus completing edge feature recognition. The recognition unit sends the image after edge feature recognition to the computing unit. When the recognition unit sends the image, it uses libx264 to compress the image data into an H.264 video stream.

2. The real-time slope deformation monitoring system based on image recognition according to claim 1, characterized in that: The acquisition unit uses a high-speed network camera to acquire slope images. The high-speed network camera uses a high-frequency image capture technology of 15 frames per second to acquire image data. The acquisition unit sends the acquired image data to the acquisition unit.

3. The real-time slope deformation monitoring system based on image recognition according to claim 1, characterized in that: The acquisition unit calls FFmpeg library functions to connect to the RTSP video stream address provided by the network high-speed camera. FFmpeg separates the video encoded data stream from the received network data packets and restores the compressed video data into frames of original RGB image data to achieve image acquisition. The acquisition unit then sends the acquired RGB image data to the preprocessing unit.

4. The real-time slope deformation monitoring system based on image recognition according to claim 3, characterized in that: The preprocessing unit uses the OpenCV library to perform grayscale processing on each frame of RGB image data. The grayscale processing formula is HD=0.3R+0.58G+0.12B, where HD is the grayscale value after processing, R is the red image, G is the green image, and B is the blue image. R, G, and B are all within the grayscale value range of 0-255. The preprocessing unit then performs Gaussian filtering on the grayscale processed image.

5. The real-time slope deformation monitoring system based on image recognition according to claim 1, characterized in that: The computing unit receives image data sent by the recognition unit, compares the pixel coordinates of the same feature point in the current frame with the initial reference frame, calculates the pixel-level displacement, establishes a conversion scale from pixel to actual millimeter based on the known physical scale spacing on the target ruler, and multiplies the pixel displacement by the scale to obtain the horizontal deformation physical quantity XB1 and the vertical deformation physical quantity XB2 of the slope.

6. The real-time slope deformation monitoring system based on image recognition according to claim 1, characterized in that: The extraction unit receives the horizontal deformation physical quantity XB1 and the vertical deformation physical quantity XB2 calculated by the calculation unit and compares them with the internal horizontal threshold Y1 and vertical threshold Y2. When the horizontal deformation physical quantity XB1 > the horizontal threshold Y1, the extraction unit retrieves the images of the three frames when the horizontal deformation physical quantity XB1 appears, before it appears, and after it appears from the acquisition unit, and sends the horizontal deformation physical quantity XB1 to the analysis unit and the display unit and issues an early warning. When the horizontal deformation physical quantity XB1 ≤ the horizontal threshold Y1, the extraction unit is in standby mode.

7. The real-time slope deformation monitoring system based on image recognition according to claim 6, characterized in that: When the vertical deformation physical quantity XB2 in the extraction unit is greater than the vertical threshold Y2, the extraction unit retrieves the images of the three frames when the vertical deformation physical quantity XB2 appears, before it appears, and after it appears from the acquisition unit, and sends the vertical deformation physical quantity XB2 to the analysis unit and the display unit to issue an early warning. When the vertical deformation physical quantity XB2 is less than or equal to the vertical threshold Y2, the extraction unit is in standby mode, and the images sent by the extraction unit are all timestamped.

8. The real-time slope deformation monitoring system based on image recognition according to claim 1, characterized in that: The analysis unit receives all the data sent by the extraction unit and performs trend analysis. The analysis unit calculates the displacement change per unit time to obtain the rate of change BS and calculates the change of the rate of change per unit time to obtain the acceleration JS. When the rate of change BS approaches zero, the slope is in a stable state. When the rate of change BS converges to a fixed value and the acceleration JS approaches zero, the slope is in a uniform deformation state. When the rate of change BS increases and the acceleration JS is positive, the slope is in an accelerated deformation state. When the rate of change BS decreases and the acceleration JS is negative, the slope is in a decelerated deformation state.

9. The real-time slope deformation monitoring system based on image recognition according to claim 1, characterized in that: The display unit receives the data sent by the extraction unit and displays the image data. The acquisition unit, preprocessing unit, recognition unit, calculation unit and extraction unit are all set in a micro industrial control computer. The transmission unit performs 4G communication network data transmission through a 4G router.

10. The real-time slope deformation monitoring system based on image recognition according to claim 1, characterized in that: The power supply unit includes a storage battery, a solar charge controller, and a solar panel. The solar panel converts solar energy into electrical energy, the storage battery stores electrical energy and provides power for operation, and the solar charge controller controls the flow of electrical energy between the solar panel and the storage battery.

Images