Edge ai-based high slope surface deformation monitoring method and related device

Abstract

The application discloses an edge AI-based high-slope slope surface deformation monitoring method and related equipment, relates to the technical field of civil engineering safety monitoring and geological disaster early warning, and comprises the following steps: collecting a specific target image in real time, processing the specific target image based on a preset lightweight AI model, obtaining displacement data corresponding to the specific target, and sending the displacement data to a backend monitoring platform, so that the backend monitoring platform performs real-time monitoring on high-slope slope surface deformation based on the displacement data and obtains high-slope slope surface deformation monitoring results. The application cooperates the backend monitoring platform and the front-end acquisition unit to complete the high-slope slope surface deformation monitoring, the front-end acquisition unit only needs to upload displacement data instead of image data, thereby greatly reducing the data transmission amount. Meanwhile, the edge design realizes real-time processing of the collected data by the front-end acquisition unit, improves the data processing efficiency, and further efficiently and accurately performs deformation monitoring in a complex field environment.

Description

Technical Field

[0001] This application relates to the field of civil engineering safety monitoring and geological disaster early warning technology, and in particular to a method and related equipment for monitoring the deformation of high slope surfaces based on edge AI. Background Technology

[0002] In related technologies, high slope stability monitoring is a crucial link in civil engineering and geological disaster early warning. Existing automated monitoring methods mainly fall into two categories: one is contact measurement (such as GNSS, total stations, and inclinometers), which requires on-site deployment of sensors and cables, resulting in complex installation, high costs, limited monitoring point density, and susceptibility to damage in harsh environments; the other is traditional visual measurement (such as Digital Image Correlation (DIC), which calculates deformation by tracking natural textures, but faces fundamental challenges in the field: natural textures are significantly affected by lighting and seasonal changes, areas without texture or with repetition cannot be monitored, and computational demands are high, relying on high-performance computers, making high-frequency real-time analysis difficult. In recent years, while deep learning-based visual methods have been introduced, most solutions transmit images to the cloud for processing, resulting in high bandwidth pressure, high network latency, and poor real-time performance. Therefore, there is an urgent need for a non-contact, high-precision deformation monitoring solution that is adaptable to complex field environments, computationally efficient, and capable of real-time edge response. Summary of the Invention

[0003] The main purpose of this application is to provide a method and related equipment for monitoring the deformation of high slopes based on edge AI, aiming to solve the technical problem of how to perform deformation monitoring efficiently and accurately in complex field environments.

[0004] To achieve the above objectives, this application proposes a high slope surface deformation monitoring method based on edge AI, which is applied to a backend monitoring platform. The high slope surface deformation monitoring method based on edge AI includes: In response to the high slope surface deformation monitoring command, the system receives displacement data corresponding to a specific target collected in real time by the front-end acquisition unit. The specific target is fixed at a key location on the slope surface, and the key location on the slope surface includes potential slip surfaces and both sides of cracks. Based on the displacement data, the deformation of the high slope surface is monitored in real time, and the monitoring results of the high slope surface deformation are obtained.

[0005] To achieve the above objectives, this application proposes a high slope surface deformation monitoring method based on edge AI, applied to a front-end acquisition unit. The front-end acquisition unit is integrated into a protective enclosure, and the front-end monitoring unit is installed on a stable reference point. The high slope surface deformation monitoring method based on edge AI includes: Based on a pre-set intelligent gimbal, images of a specific target are acquired in real time to obtain specific target images; Based on a preset lightweight AI model, the image of the specific target is processed to obtain the displacement data corresponding to the specific target. The displacement data is sent to the backend monitoring platform so that the backend monitoring platform can monitor the deformation of the high slope surface in real time based on the displacement data and obtain the high slope surface deformation monitoring results.

[0006] In one embodiment, the step of acquiring a specific target image in real time based on a preset intelligent gimbal to obtain the specific target image further includes: Obtain environmental information; Based on the environmental information, the preset active illumination system, and the preset intelligent gimbal, a high dynamic range industrial area array camera is used to acquire images of a specific target in real time to obtain the specific target image. The preset active illumination system consists of an infrared LED array of a specific wavelength, and the preset active illumination system is triggered synchronously with the exposure of the high dynamic range industrial area array camera.

[0007] In one embodiment, the step of processing the specific target image based on a preset lightweight AI model to obtain displacement data corresponding to the specific target further includes: The specific target image is preprocessed to obtain a preprocessed specific target image. The preprocessing includes automatic white balance and HDR synthesis. The preprocessing is used to suppress the influence of uneven illumination on the specific target image. Based on a pre-defined lightweight AI model, the pre-processed image of a specific target is processed to obtain the displacement data corresponding to the specific target.

[0008] In one embodiment, the step of processing the preprocessed image of a specific target based on a preset lightweight AI model to obtain displacement data corresponding to the specific target further includes: A preset lightweight AI model is used to perform pixel-level segmentation on the preprocessed specific target image to accurately extract the pixel region of the specific target. The pixel region is a binary region. The preset lightweight AI model is obtained after training on a large dataset of specific target images in multiple outdoor scenarios. Perform morphological filtering on the pixel region to remove noise, and obtain the pixel region after noise removal; Based on the pixel regions after noise removal, displacement data corresponding to a specific target is obtained.

[0009] In one embodiment, the step of obtaining displacement data corresponding to a specific target based on the noise-removed pixel region further includes: Based on the specific target, the high dynamic range industrial area array camera is calibrated to obtain the intrinsic and extrinsic parameters of the high dynamic range industrial area array camera. The intrinsic parameters include focal length and distortion coefficient, and the extrinsic parameters include spatial position and orientation relative to the monitoring area. The physical dimensions of the specific target are known. Based on the noise-removed pixel region and the preset centroid calculation algorithm, the sub-pixel level coordinates of the centroid of the noise-removed pixel region in the image coordinate system are calculated. The preset centroid calculation algorithm includes gray-level weighted centroid method and ellipse fitting. Based on the sub-pixel level coordinates of the centroid in the image coordinate system corresponding to the current frame and the sub-pixel level coordinates of the centroid in the image coordinate system corresponding to the previous frame, the two-dimensional displacement of the specific target is determined. Based on the intrinsic parameters, the extrinsic parameters, the physical dimensions of the specific target, and the preset perspective projection geometry model, the two-dimensional displacement is processed to obtain the displacement data corresponding to the specific target. The displacement data includes three-dimensional displacement and displacement change trend. The three-dimensional displacement is the actual displacement of the specific target on the plane perpendicular to the optical axis.

[0010] In one embodiment, the step of processing the two-dimensional displacement based on the intrinsic parameters, the extrinsic parameters, the physical dimensions of the specific target, and a preset perspective projection geometry model to obtain the displacement data corresponding to the specific target further includes: Based on the intrinsic parameters, the extrinsic parameters, the physical dimensions of the specific target, and the preset perspective projection geometric model, the two-dimensional displacement is calculated to obtain the three-dimensional displacement corresponding to the specific target; Based on the three-dimensional displacement, the displacement time series data corresponding to the specific target is constructed; A filtering operation is performed on the displacement time series data to smooth out random errors, resulting in filtered displacement time series data; Based on the filtered displacement time series data, the displacement change trend is extracted.

[0011] Furthermore, to achieve the above objectives, this application also proposes a high slope surface deformation monitoring device based on edge AI, applied to a backend monitoring platform. The edge AI-based high slope surface deformation monitoring device includes: The receiving module is used to respond to the high slope surface deformation monitoring command and receive the displacement data corresponding to a specific target collected in real time by the front-end acquisition unit. The specific target is fixed at a key position on the slope surface, and the key position on the slope surface includes the potential slip surface and both sides of the crack. The monitoring module is used to monitor the deformation of the high slope surface in real time based on the displacement data, and obtain the monitoring results of the high slope surface deformation.

[0012] Furthermore, to achieve the above objectives, this application also proposes a high slope surface deformation monitoring device based on edge AI, applied to a front-end acquisition unit. The edge AI-based high slope surface deformation monitoring device includes: The acquisition module is used to acquire images of a specific target in real time based on a preset intelligent gimbal, and obtain the specific target image. An image processing module is used to process the image of the specific target based on a preset lightweight AI model to obtain displacement data corresponding to the specific target. The sending module is used to send the displacement data to the backend monitoring platform, so that the backend monitoring platform can monitor the deformation of the high slope surface in real time based on the displacement data and obtain the high slope surface deformation monitoring results.

[0013] In addition, to achieve the above objectives, this application also proposes a high slope surface deformation monitoring device based on edge AI. The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. The computer program is configured to implement the steps of the high slope surface deformation monitoring method based on edge AI as described above.

[0014] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the high slope deformation monitoring method based on edge AI as described above.

[0015] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the edge AI-based high slope deformation monitoring method described above.

[0016] One or more technical solutions proposed in this application have at least the following technical effects: This application discloses a method and related equipment for monitoring the deformation of high slopes based on edge AI, relating to the fields of civil engineering safety monitoring and geological disaster early warning technology. Compared to other related technologies, firstly, contact measurement (such as GNSS, total stations, and inclinometers) requires on-site deployment of sensors and cables, which is complex, costly, and has limited monitoring point density, making it prone to damage in harsh environments; secondly, traditional visual measurement (such as Digital Image Correlation (DIC)) calculates deformation by tracking natural textures, but faces fundamental challenges in the field: natural textures are significantly affected by lighting and seasonal changes, areas without texture or with repetition cannot be monitored, and computation is large, relying on high-performance computers, making high-frequency real-time analysis difficult. In this application, firstly, based on a preset intelligent gimbal, images of specific targets are acquired in real time to obtain specific target images; then, based on a preset lightweight AI model, the specific target images are processed to obtain displacement data corresponding to the specific targets; finally, the displacement data is sent to a backend monitoring platform, which then monitors the deformation of the high slope in real time based on the displacement data to obtain the high slope deformation monitoring results.

[0017] Understandably, this application relies on a backend monitoring platform and a frontend acquisition unit to collaboratively monitor high slope deformation. The frontend acquisition unit only needs to upload displacement data, not image data, significantly reducing data transmission volume, improving data transmission efficiency, and lowering communication costs and cloud storage pressure. Simultaneously, the edge computing design enables the frontend acquisition unit to process the acquired data in real time, improving data processing efficiency and supporting high-frequency monitoring. Therefore, deformation monitoring can be performed efficiently and with high precision in complex field environments. Attached Figure Description

[0018] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating an embodiment of the high slope deformation monitoring method based on edge AI in this application. Figure 2 This is a flowchart illustrating Embodiment 2 of the high slope deformation monitoring method based on edge AI in this application; Figure 3 This is a flowchart illustrating Embodiment 3 of the high slope deformation monitoring method based on edge AI in this application; Figure 4 This is a schematic diagram of the equipment structure of the hardware operating environment involved in the edge AI-based high slope deformation monitoring method in this application embodiment.

[0021] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0022] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0023] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0024] The main solution in this application embodiment is: In this embodiment, for ease of description, the following description uses an edge AI-based high slope deformation monitoring device as the execution subject.

[0025] The relevant technologies are problematic because: firstly, contact measurement (such as GNSS, total station, and inclinometer) requires the on-site deployment of sensors and cables, which is complex, costly, and has a limited density of monitoring points, making it prone to damage in harsh environments; secondly, traditional visual measurement (such as digital image correlation (DIC)) calculates deformation by tracking natural textures, but faces fundamental challenges in the field: natural textures are significantly affected by lighting and seasonal changes, areas without texture or with repetition cannot be monitored, and the computational load is large, relying on high-performance computers, making it difficult to achieve high-frequency real-time analysis.

[0026] This application provides a solution in which: first, based on a preset intelligent gimbal, a specific target image is acquired in real time to obtain the specific target image; then, based on a preset lightweight AI model, the specific target image is processed to obtain displacement data corresponding to the specific target; finally, the displacement data is sent to a backend monitoring platform so that the backend monitoring platform can monitor the deformation of the high slope surface in real time based on the displacement data to obtain the high slope surface deformation monitoring result.

[0027] Understandably, this application relies on a backend monitoring platform and a frontend acquisition unit to collaboratively monitor high slope deformation. The frontend acquisition unit only needs to upload displacement data, not image data, significantly reducing data transmission volume, improving data transmission efficiency, and lowering communication costs and cloud storage pressure. Simultaneously, the edge computing design enables the frontend acquisition unit to process the acquired data in real time, improving data processing efficiency and supporting high-frequency monitoring. Therefore, deformation monitoring can be performed efficiently and with high precision in complex field environments.

[0028] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or an electronic device capable of performing the above functions, such as an edge AI-based high slope deformation monitoring device. The following description uses an edge AI-based high slope deformation monitoring device as an example to illustrate this embodiment and the subsequent embodiments.

[0029] Based on this, this application provides a method for monitoring the deformation of high slope surfaces based on edge AI, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the high slope deformation monitoring method based on edge AI in this application.

[0030] In this embodiment, the method is applied to a front-end acquisition unit, which is integrated into a protective enclosure. The front-end monitoring unit is installed on a stable reference point. The high slope deformation monitoring method based on edge AI includes steps A100~A300: Step A100: Based on the preset intelligent gimbal, acquire specific target images in real time to obtain specific target images; It should be noted that the preset intelligent gimbal refers to an electrically controlled mechanical rotating device integrated into the front-end monitoring unit, possessing two degrees of freedom: pitch and horizontal rotation. This gimbal can automatically adjust its attitude according to a preset program or remote command, aligning the camera's optical axis with different targets or scanning areas, achieving flexible coverage of multiple areas and targets.

[0031] It should be noted that a specific target image refers to a digital image containing one or more specially designed optical targets, captured by a high-resolution image acquisition module (industrial area array camera). This image has high dynamic range characteristics, and the target area exhibits stable and bright optical features due to retroreflective or high-reflective coatings, with a natural slope scene as the background.

[0032] In this embodiment, after the front-end monitoring unit is powered on and initialized, the edge AI image processing module reads the preset gimbal scanning strategy (e.g., sequentially aligning with several pre-deployed target areas, or aligning only with a specified target according to instructions). The intelligent gimbal performs pitch and / or horizontal rotation movements according to the strategy instructions, aligning the camera's optical axis center with the target location. Subsequently, the high-resolution image acquisition module acquires images at a preset frequency (e.g., once per minute) or through an event-triggered method (e.g., vibration triggering). During acquisition, the active illumination system (infrared LED array) is triggered synchronously with the camera exposure, providing uniform illumination. The camera outputs a high dynamic range image, which is the "specific target image." The acquired image can be directly sent to the edge AI image processing module for further processing without storage or uploading.

[0033] Understandably, this step enables automated and controllable image acquisition of multiple monitoring areas on the slope. Its purpose is to utilize the mobility of the intelligent pan-tilt unit, allowing a single front-end monitoring unit to cover multiple dispersed targets, thus avoiding the need to configure a separate camera for each target.

[0034] Specifically, the step of acquiring a specific target image in real time based on a preset intelligent gimbal to obtain the specific target image further includes steps A110~A120: Step A110: Obtain environmental information; It should be noted that environmental information refers to data reflecting the on-site lighting conditions, weather conditions, and temporal characteristics of the slope. Specifically, this includes parameters such as ambient light levels (e.g., daytime, dusk, nighttime), the presence of rain or fog, and whether the area is in backlight or shade. This information can be obtained from ambient light sensors and temperature / humidity sensors integrated into the front-end monitoring unit, or by analyzing the brightness histogram of camera preview images.

[0035] Understandably, this step provides the data foundation for adaptive adjustment of the active lighting system and camera parameters.

[0036] Step A120: Based on the environmental information, the preset active illumination system, and the preset intelligent gimbal, a high dynamic range industrial area array camera is used to acquire a specific target image in real time to obtain the specific target image. The preset active illumination system consists of an infrared LED array of a specific wavelength, and the preset active illumination system is triggered synchronously with the exposure of the high dynamic range industrial area array camera.

[0037] It should be noted that a pre-set active illumination system refers to an illumination device composed of an infrared LED array of specific wavelengths, used to actively provide uniform illumination when ambient light is insufficient. The core feature of this system is that the wavelength of the emitted infrared light is selected (such as 850 nm or 940 nm), which is invisible or weakly visible to the human eye, and is matched with the retroreflective microprism film or high reflectivity coating on the target surface to improve the contrast of the target in the image.

[0038] It should be noted that a specific wavelength infrared LED array refers to a lighting source composed of multiple infrared light-emitting diodes arranged in a certain manner (such as a rectangular or circular array). The "specific wavelength" refers to an optimized single or narrow-band infrared wavelength that can be effectively reflected by the retroreflective material of the target while reducing infrared interference from ambient light.

[0039] It should be noted that exposure synchronization triggering means that the emission time of the active fill light system is precisely aligned with the exposure start time of the high dynamic range industrial area scan camera, so that the fill light is lit only during the period when the camera sensor accumulates photocharge and is turned off during non-exposure periods.

[0040] It should be noted that high dynamic range industrial area scan cameras have wide dynamic range imaging capabilities, which can simultaneously preserve details of the bright target area and the darker background area in the same frame image.

[0041] In this embodiment, after acquiring the environmental information in step A110, the edge AI image processing module determines whether the current ambient brightness is lower than a preset threshold (e.g., illuminance less than 10 lux). If it is lower than the threshold, a supplementary lighting start command is generated. Before starting image acquisition, the system configures the camera's exposure parameters (exposure time, gain, etc.) to adapt to the current environment. Subsequently, the camera sends an exposure start signal, which is simultaneously sent as a trigger pulse to the driving circuit of the infrared LED array. Upon receiving the trigger signal, the driving circuit illuminates the infrared LED array with a preset current and duration, the illumination duration matching the camera's exposure time (usually slightly longer than the exposure time to ensure complete coverage). During exposure, the camera receives reflected infrared light from the target while suppressing ambient stray light. After exposure, the supplementary lighting system automatically turns off. The camera outputs a high dynamic range image, which is the "specific target image." If the ambient brightness is sufficient, the supplementary lighting system does not start, and the camera acquires images normally using ambient light.

[0042] Understandably, this step achieves adaptive, low-interference, high-quality target image acquisition. It automatically determines whether to provide supplemental lighting based on ambient light, and maximizes supplemental lighting efficiency through synchronous triggering technology. Infrared supplemental lighting of a specific wavelength is invisible to the human eye, avoiding light pollution to personnel or the surrounding environment. The synchronous triggering mechanism ensures that all supplemental lighting energy is used for effective exposure, saving significant energy compared to constant illumination, making it suitable for solar-powered outdoor scenarios. Infrared supplemental lighting, combined with the target's retroreflective properties, ensures the target exhibits stable, high-brightness characteristics in the image, effectively overcoming the effects of strong shadows during the day, low illumination at night, and scattering from rain and fog. High dynamic range imaging ensures that both the target and background are clearly rendered across a wide brightness range, improving the robustness of subsequent AI recognition.

[0043] Step A200: Based on a preset lightweight AI model, process the specific target image to obtain displacement data corresponding to the specific target; It should be noted that the pre-trained lightweight AI model refers to a lightweight semantic segmentation neural network model that is pre-trained and deployed in the edge AI image processing module. This model employs a DeepLabV3+ model based on an improved MobileNet-V3 backbone network. This model has been trained on a large dataset of target images from various outdoor scenarios (sunny, rainy, foggy, dawn / dusk, and partially occluded targets) and utilizes data augmentation techniques such as simulated dirt and motion blur. The model has a small parameter size, making it suitable for real-time operation on embedded neural network processors.

[0044] In this embodiment, after receiving the specific target image output in step A100, the edge AI image processing module first preprocesses the image (automatic white balance, high dynamic range synthesis) to suppress uneven illumination. The preprocessed image is directly input into a preset lightweight AI model. The model performs pixel-level semantic segmentation on the image, accurately extracting the pixel region of each target and segmenting the background. Subsequently, morphological filtering is performed on the segmented binarized regions to remove noise. For each target region, the sub-pixel-level coordinates of its centroid in the image coordinate system are calculated using the gray-scale weighted centroid method or ellipse fitting method, while simultaneously decoding the identity information embedded in the target pattern. Then, the difference in centroid coordinates between the current frame and the initial reference frame (or the previous frame) for the same identity target is compared to obtain the pixel displacement. Using pre-calibrated camera intrinsic parameters (focal length, distortion coefficients) and extrinsic parameters (spatial position and attitude), combined with the known physical dimensions of the target, pixel displacement is converted into actual spatial displacement on a plane perpendicular to the optical axis through a perspective projection geometry model (for 3D displacement requirements, it can be estimated through the intersection of multiple cameras or gimbal rotation information). Finally, the displacement data corresponding to each target is output.

[0045] Understandably, this step enables intelligent conversion from image to displacement directly at the data acquisition end. Its purpose is to replace the traditional method of uploading images to the cloud for processing, significantly reducing data transmission volume and latency. Simultaneously, the lightweight AI model exhibits strong robustness to outdoor interference such as changes in lighting, rain, fog, dust, and cluttered backgrounds, overcoming the fundamental shortcomings of traditional digital image correlation methods; edge processing enables the system to operate autonomously without human intervention.

[0046] Step A300: The displacement data is sent to the backend monitoring platform so that the backend monitoring platform can monitor the deformation of the high slope surface in real time based on the displacement data and obtain the monitoring results of the high slope surface deformation.

[0047] It should be noted that the backend monitoring platform refers to the software system deployed in the monitoring center or cloud, consisting of a data aggregation and storage server, a digital twin and early warning system, and remote control clients. It is responsible for receiving and storing all data uploaded by the front-end units, performing 3D visualization, early warning judgment, data querying, and report generation.

[0048] In this embodiment, after completing step A200, the edge AI image processing module packages the calculated displacement data (typically only a few tens of bytes, containing target identification, displacement value, timestamp, device status, etc.) and sends it to the backend monitoring platform via the data communication module (supporting 4G / 5G, wired, or LoRa). The transmission can be done via real-time push or batch upload, depending on the communication strategy. The backend monitoring platform's data aggregation and storage server receives the data, verifies, decodes, and stores it. Subsequently, the digital twin and early warning system maps the real-time displacement data of each target onto a pre-constructed three-dimensional geological model of the slope, dynamically updating the deformation curves and three-dimensional deformation fields at each point. The system calculates the cumulative displacement and deformation rate in real time and compares them with preset multi-level early warning thresholds (rate threshold, cumulative displacement threshold). When the threshold is exceeded, a graded alarm is automatically triggered (e.g., SMS, platform pop-up). All data and early warning records together constitute the high slope surface deformation monitoring results, which users can view, query, or export reports through a remote control client (Web or App interface).

[0049] Understandably, this step completes the collaboration between the front-end and back-end, transforming the results of edge computing into monitoring information usable by the project. Its purpose is to achieve centralized data management, visualization, and automatic early warning, providing decision support for management personnel.

[0050] This application discloses a method and related equipment for monitoring the deformation of high slopes based on edge AI, relating to the fields of civil engineering safety monitoring and geological disaster early warning technology. Compared to other related technologies, firstly, contact measurement (such as GNSS, total stations, and inclinometers) requires on-site deployment of sensors and cables, which is complex, costly, and has limited monitoring point density, making it prone to damage in harsh environments; secondly, traditional visual measurement (such as Digital Image Correlation (DIC)) calculates deformation by tracking natural textures, but faces fundamental challenges in the field: natural textures are significantly affected by lighting and seasonal changes, areas without texture or with repetition cannot be monitored, and computation is large, relying on high-performance computers, making high-frequency real-time analysis difficult. In this application, firstly, based on a preset intelligent gimbal, images of specific targets are acquired in real time to obtain specific target images; then, based on a preset lightweight AI model, the specific target images are processed to obtain displacement data corresponding to the specific targets; finally, the displacement data is sent to a backend monitoring platform, which then monitors the deformation of the high slope in real time based on the displacement data to obtain the high slope deformation monitoring results.

[0051] Understandably, this application relies on a backend monitoring platform and a frontend acquisition unit to collaboratively monitor high slope deformation. The frontend acquisition unit only needs to upload displacement data, not image data, significantly reducing data transmission volume, improving data transmission efficiency, and lowering communication costs and cloud storage pressure. Simultaneously, the edge computing design enables the frontend acquisition unit to process the acquired data in real time, improving data processing efficiency and supporting high-frequency monitoring. Therefore, deformation monitoring can be performed efficiently and with high precision in complex field environments.

[0052] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in Embodiment 1 above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 2 The step of processing the specific target image based on a preset lightweight AI model to obtain the displacement data corresponding to the specific target further includes steps B100~B200: Step B100: Preprocess the specific target image to obtain a preprocessed specific target image. The preprocessing includes automatic white balance and HDR synthesis. The preprocessing is used to suppress the influence of uneven illumination on the specific target image. It should be noted that preprocessing refers to a series of correction and enhancement operations performed on the original image before it is input into the lightweight AI model. The purpose is to improve image quality and unify image features, thereby improving the accuracy and robustness of subsequent recognition and localization.

[0053] It should be noted that automatic white balance is an image color correction technique used to eliminate the influence of different light source color temperatures on image color reproduction. Specifically, the automatic white balance algorithm automatically detects areas in an image that should originally be neutral colors (such as white or gray) and adjusts the gain of the red, green, and blue channels to restore these areas to neutral colors, thus ensuring that the true colors of the target (such as the black, white, and gray of a high-contrast pattern) present a consistent appearance under different lighting conditions.

[0054] It's important to note that HDR compositing is an abbreviation for High Dynamic Range Compositing. It refers to merging multiple frames of images taken at different exposure parameters (short exposure, normal exposure, long exposure) of the same scene to create a single frame that simultaneously contains details in both bright and dark areas. Since the target surface has high reflectivity while the background may be dark, a single exposure is insufficient to capture both details; HDR compositing effectively solves this problem.

[0055] It should be noted that uneven illumination refers to the phenomenon where different areas of an image receive inconsistent lighting intensity. For example, some areas may be shaded, while others may be directly exposed to sunlight, or the direction of the light source may cause one side of the target to be brighter than the other. Uneven illumination can interfere with the accuracy of target edge extraction and centroid localization.

[0056] In this embodiment, after completing step A120, the high dynamic range industrial area scan camera outputs raw image data (which may be a sequence of images with different exposures, or a single-frame high dynamic range image directly output by the built-in HDR function). The edge AI image processing module first performs automatic white balance on the image data: it statistically analyzes the distribution of the red, green, and blue channels of the pixels in the image, estimates the color temperature of the current ambient light source, and then calculates the gain coefficient of each channel and applies it to the image, so that the areas in the target pattern that should be white appear neutral white. Subsequently, HDR synthesis is performed: if the camera outputs multiple frames of images with different exposures, a weighted mapping algorithm (such as exposure-based fusion) is used to extract and synthesize the appropriately exposed pixel areas in each frame into a single frame of wide dynamic range image; if the camera has already output a single frame of HDR image, only necessary tone mapping is performed for subsequent processing. After the above preprocessing, a frame of "preprocessed specific target image" with realistic color, balanced brightness, and clear details is obtained and sent to step B200.

[0057] Understandably, this step involves crucial quality enhancement before the image enters the AI ​​model. Eliminating or mitigating the destructive effects of complex outdoor lighting conditions (such as color temperature variations during sunrise and sunset, alternating shadows and sunlight, and local overexposure or underexposure of the target) on image feature stability significantly improves the pixel-level classification accuracy and sub-pixel localization stability of subsequent semantic segmentation, thereby indirectly ensuring the reliability of millimeter-level displacement monitoring.

[0058] Step B200: Based on a preset lightweight AI model, the preprocessed image of a specific target is processed to obtain displacement data corresponding to the specific target.

[0059] In this embodiment, the edge AI image processing module inputs the preprocessed image of a specific target into a preset lightweight AI model. The preset lightweight AI model processes the preprocessed image of the specific target to obtain the displacement data corresponding to each target.

[0060] Specifically, the step of processing the preprocessed image of a specific target based on a preset lightweight AI model to obtain the displacement data corresponding to the specific target further includes steps B210 to B230: Step B210: Use a preset lightweight AI model to perform pixel-level segmentation on the preprocessed specific target image to accurately extract the pixel region of the specific target. The pixel region is a binary region. The preset lightweight AI model is obtained after training on a large dataset of specific target images from multiple outdoor scenes. It should be noted that pixel-level segmentation refers to classifying each pixel in the input image, determining whether it belongs to the target object (target) or the background, and outputting a classification result image with the same resolution as the input image, where each pixel is assigned a category label.

[0061] Binarization refers to the pixel-level segmentation output where pixels belonging to the target are marked with the first value (e.g., 1, corresponding to white) and pixels belonging to the background are marked with the second value (e.g., 0, corresponding to black). The entire image thus presents a binary image with only black and white values, where the white connected regions are the pixel regions of the target.

[0062] A large dataset of field-specific target images in various scenarios refers to a collection of samples used to train the lightweight AI model. This dataset contains real or simulated enhanced images collected under different weather conditions (sunny, rainy, foggy), different lighting times (dawn, dusk, night), and different target states (partial occlusion, dirt, motion blur), and the precise pixel boundaries of the targets have been manually or automatically labeled.

[0063] In this embodiment, the edge AI image processing module inputs the preprocessed target image output from step B100 into a pre-loaded lightweight AI model. The model performs forward inference computation: the encoder (such as the MobileNet-V3 backbone) extracts multi-scale semantic features from the image; the decoder (such as the hollow spatial pyramid pooling module of DeepLabV3+) fuses features from different receptive fields and outputs a probability value for each pixel belonging to the target. Subsequently, the probability value is compared with a preset threshold (typically 0.5). Pixels greater than the threshold are marked as foreground (target), and pixels less than or equal to the threshold are marked as background, thereby generating a binary segmentation map. In this binary map, one or more connected white pixel regions represent the pixel regions of each target present in the image.

[0064] Understandably, this step achieves accurate separation of the target from the complex field background. Leveraging the powerful feature learning capabilities of deep learning models, it can accurately identify and segment the target area even under adverse conditions such as drastic lighting changes, similar background textures, and partial target occlusion or dust contamination. The segmentation result is in binary form, with a small data volume, facilitating rapid subsequent processing.

[0065] Step B220: Perform morphological filtering to remove noise from the pixel region to obtain the noise-removed pixel region; It should be noted that morphological filtering refers to image processing operations based on mathematical morphology, mainly using structuring elements to perform erosion, dilation, opening, and closing operations on binary images. Erosion can eliminate isolated single-pixel noise points, dilation can fill tiny holes within a region, opening operations (erosion followed by dilation) are used to remove small, protruding noise points, and closing operations (dilation followed by erosion) are used to fill small cracks or holes within a region.

[0066] The pixel region after noise removal refers to the target pixel region after morphological filtering, where the edges are smoother, there are no holes inside, no isolated false detection pixels around, and the overall shape is closer to the geometric contour of the real target.

[0067] In this embodiment, the edge AI image processing module sequentially performs morphological operations on the binarized region image output in step B210. First, an opening operation is performed: a structuring element of appropriate size (such as a 3×3 or 5×5 square kernel) is selected, and an erosion operation is performed on the binary image to eliminate isolated noise points outside the target region caused by missegmentation. Then, a dilation operation is performed to restore the target region, which has shrunk due to erosion, to near its original size. Next, a closing operation is performed: dilation followed by erosion to fill small holes or cracks within the target region that may be caused by dirt, occlusion, or incomplete segmentation. After these operations, a noise-removed pixel region is obtained, where each connected component within this region corresponds to a complete target.

[0068] Understandably, this step improves the integrity and purity of the target region. It eliminates potential misclassification noise and region defects from the semantic segmentation model, providing high-quality input for subsequent sub-pixel localization.

[0069] Step B230: Based on the pixel region after noise removal, obtain the displacement data corresponding to the specific target.

[0070] In this embodiment, the edge AI image processing module processes each target pixel region after noise removal and finally outputs the displacement data corresponding to each target.

[0071] Specifically, the step of obtaining the displacement data corresponding to a specific target based on the pixel region after noise removal further includes steps B231 to B234: Step B231: Based on the specific target, calibrate the high dynamic range industrial area array camera to obtain the intrinsic and extrinsic parameters of the high dynamic range industrial area array camera. The intrinsic parameters include focal length and distortion coefficient, and the extrinsic parameters include spatial position and orientation relative to the monitoring area. The physical dimensions of the specific target are known. It should be noted that calibration refers to the process of establishing a mathematical relationship between the pixel coordinates of a camera image and the real-world spatial coordinates by photographing a calibration object of known geometric dimensions (in this case, a specific target). The calibration results are used to subsequently calculate the actual displacement from the pixel displacement.

[0072] It should be noted that intrinsic parameters refer to the set of inherent parameters inside the camera, mainly including focal length (the distance from the optical center of the camera lens to the imaging plane, in pixels) and distortion coefficients (parameters that describe the radial and tangential distortions produced when the lens produces images, used to correct image distortion).

[0073] It should be noted that external parameters refer to parameters that describe the position and orientation of the camera in the world coordinate system, including spatial position (the three-dimensional coordinates of the camera's optical center in the world coordinate system) and attitude (the direction of the camera's optical axis, horizontal rotation angle, pitch angle, etc.).

[0074] It should be noted that the spatial position and attitude relative to the monitoring area refers to the geometric relationship between the camera and the slope monitoring area, specifically including the three-dimensional coordinate offset of the camera relative to a fixed reference point (such as the center of the target deployment area) as well as the camera's pitch angle, horizontal angle, and roll angle.

[0075] It should be noted that the physical dimensions of a specific target refer to the known true length of the geometric features used for calibration on the target, such as the diameter of the concentric circles in the target pattern, the side length of the coded mark, etc., in millimeters.

[0076] In this embodiment, during system deployment or periodic maintenance, a specific target (or a dedicated calibration plate) is securely placed at a known location within the monitoring area. A high dynamic range industrial area array camera captures multiple images of the target from different angles. The edge AI image processing module or backend platform runs a calibration algorithm (Zhang Zhengyou calibration method): extracting the coordinates of the target's corner points or feature points in the images, establishing a correspondence with known physical dimensions, and obtaining intrinsic parameters such as the camera's focal length, principal point coordinates, and distortion coefficients through least-squares optimization. Simultaneously, using the target's known coordinates in space (e.g., obtained through total station measurement), the camera's spatial position and attitude (pitch angle, horizontal angle, roll angle) relative to the monitoring area are calculated as extrinsic parameters. The calibration results are stored in the non-volatile memory of the edge AI image processing module for real-time retrieval in subsequent steps.

[0077] Understandably, this step forms the mathematical basis for converting image pixel displacement into actual physical displacement. Its purpose is to accurately quantify the geometric mapping relationship between the camera and the monitored scene, eliminating the impact of lens distortion and perspective distortion on measurement accuracy.

[0078] Step B232: Based on the noise-removed pixel region and the preset centroid calculation algorithm, calculate the sub-pixel level coordinates of the centroid of the noise-removed pixel region in the image coordinate system. The preset centroid calculation algorithm includes gray-level weighted centroid method and ellipse fitting. It should be noted that the preset centroid calculation algorithm refers to the mathematical method selected in advance for calculating the geometric center position of the target area. In this embodiment, the algorithm includes two types: gray-scale weighted centroid method and ellipse fitting method.

[0079] It should be noted that the gray-scale weighted centroid method uses the gray-scale value of each pixel within a region as a weight to calculate the coordinates of the region's centroid. Compared to the ordinary centroid (which only considers pixel position), the gray-scale weighted centroid method is more robust to changes in illumination and can achieve sub-pixel accuracy.

[0080] It should be noted that the ellipse fitting method refers to performing least-squares ellipse fitting on the outer contour points of the target area, using the center of the fitted ellipse as the centroid coordinates. This method is particularly accurate when the target appears as an ellipse due to perspective projection.

[0081] It should be noted that subpixel-level coordinates refer to coordinate values ​​that are not limited to integer pixel positions, but can be accurate to the decimal part of a pixel (such as 0.1 pixels or 0.01 pixels). Subpixel positioning is key to achieving millimeter-level displacement monitoring.

[0082] It should be noted that the image coordinate system refers to a two-dimensional coordinate system with the top left corner of the image as the origin, the horizontal axis to the right as the U-axis, and the vertical axis downward as the V-axis, and the coordinate unit is pixels.

[0083] In this embodiment, the edge AI image processing module performs the following operations on each noise-removed pixel region output in step B220: First, it obtains the coordinates and corresponding grayscale values ​​of all pixels within the region. If the grayscale weighted centroid method is used, the centroid coordinates are calculated according to the formula. If the ellipse fitting method is used, it first extracts the set of pixels on the region boundary contour, fits the ellipse equation using the least squares method, and uses the center of the ellipse as the centroid coordinates. The calculated centroid coordinates are floating-point numbers, i.e., sub-pixel level coordinates. These coordinates, along with the timestamp of the current frame and the target identification, are temporarily stored for use in step B233.

[0084] Understandably, this step achieves precise quantization of the target position. It transforms the binarized region into accurate numerical coordinates, providing high-resolution input for displacement calculations.

[0085] Step B233: Based on the sub-pixel level coordinates of the centroid corresponding to the current frame in the image coordinate system and the sub-pixel level coordinates of the centroid corresponding to the previous frame in the image coordinate system, determine the two-dimensional displacement of the specific target. It should be noted that the current frame refers to the frame corresponding to the specific target image obtained in this acquisition and processing.

[0086] It should be noted that the previous frame refers to the image frame of the same target that was acquired and processed before the current frame. The initial reference frame can be the first frame acquired after the system starts or a user-specified reference frame.

[0087] It should be noted that two-dimensional displacement refers to the vector in the image coordinate system in which the centroid of the target moves from the position of the previous frame to the position of the current frame. It includes the pixel change in the horizontal direction (U axis) and the pixel change in the vertical direction (V axis), and the unit is pixels.

[0088] In this embodiment, the edge AI image processing module maintains a historical centroid coordinate record for each target. For the sub-pixel level coordinates of the target in the current frame, the module looks up the coordinates of the target in the previous frame (or the initial reference frame), calculates the difference, and this difference is the two-dimensional displacement of the target in the image coordinate system. If the system requires monitoring the cumulative displacement relative to the initial state, the initial reference frame coordinates are used; if it requires monitoring the incremental displacement between frames, the coordinates of the previous frame are used. The calculation results are temporarily stored in floating-point form for conversion in step B234.

[0089] Understandably, this step extracts the change in the target's position between the two frames and uses it as the direct input for subsequent spatial transformation. Its calculation is simple and extremely fast, making it suitable for high-frequency processing at the edge. By comparing the centroid coordinates of the same target, common-mode errors such as the camera's own minor vibrations are automatically eliminated. It supports the selection of absolute displacement (relative to the initial frame) or relative displacement (relative to the previous frame), flexibly meeting different monitoring needs.

[0090] Step B234: Based on the intrinsic parameters, the extrinsic parameters, the physical dimensions of the specific target, and the preset perspective projection geometry model, the two-dimensional displacement is processed to obtain the displacement data corresponding to the specific target. The displacement data includes three-dimensional displacement and displacement change trend. The three-dimensional displacement is the actual displacement of the specific target on the plane perpendicular to the optical axis.

[0091] It should be noted that the preset perspective projection geometry model refers to the mathematical relationship describing the projection of three-dimensional spatial points onto a two-dimensional image plane through a camera lens. It is usually based on a pinhole camera model and takes into account lens distortion. This model utilizes intrinsic parameters, extrinsic parameters, and known object point depth information to achieve the mapping from image coordinates to spatial coordinates.

[0092] It should be noted that three-dimensional displacement specifically refers to the actual spatial displacement after converting two-dimensional pixel displacement, and is limited to "the actual displacement perpendicular to the optical axis plane".

[0093] It should be noted that the displacement change trend refers to the smooth deformation pattern extracted after filtering the displacement time series (such as Kalman filtering), which is used to determine whether the slope is stable, deforming at a constant speed or accelerating, and to provide a basis for early warning.

[0094] In this embodiment, the edge AI image processing module calls the intrinsic parameters (focal length, distortion coefficient) and extrinsic parameters (camera position and attitude) obtained in step B231. First, the distortion coefficient is used to correct the distortion of the two-dimensional displacement, eliminating errors caused by lens deformation. Then, according to the perspective projection geometry model: given the physical dimensions of the target (e.g., the actual diameter D mm), its corresponding pixel diameter d in the image can be calculated from the target area. Through the proportional relationship, the pixel displacement is multiplied by this ratio to obtain the actual displacement component on the plane perpendicular to the optical axis. For scenes requiring complete three-dimensional displacement, if there is a second camera or gimbal rotation information, the depth displacement along the optical axis is estimated using the triangulation principle. Finally, the three-dimensional displacement vector is obtained. Subsequently, the displacement data of the target is updated in the time series, and optional Kalman filtering is used to smooth the sequence and extract the displacement change trend (e.g., velocity, acceleration). Finally, the displacement data corresponding to each target is output, including the three-dimensional displacement value and trend information.

[0095] Understandably, this step converts abstract pixel changes into the actual movement distance and direction of the slope surface. By utilizing a perspective projection geometric model and calibration parameters, the effects of perspective distortion and lens distortion are eliminated, ensuring the accuracy of the measurements. Proportional conversion is performed using the physical dimensions of the target, eliminating the need for additional distance measurement equipment and simplifying the system. It can output the actual displacement perpendicular to the optical axis plane, directly reflecting the horizontal and vertical sliding of the slope surface. Combined with the displacement change trend obtained through filtering, it provides rate and acceleration criteria in addition to instantaneous values ​​for subsequent early warning, improving the reliability of monitoring.

[0096] Specifically, the step of processing the two-dimensional displacement based on the intrinsic parameters, the extrinsic parameters, the physical dimensions of the specific target, and the preset perspective projection geometric model to obtain the displacement data corresponding to the specific target further includes steps 1 to 4: Step 1: Based on the intrinsic parameters, the extrinsic parameters, the physical dimensions of the specific target, and the preset perspective projection geometry model, calculate the two-dimensional displacement to obtain the three-dimensional displacement corresponding to the specific target; Step 2: Based on the three-dimensional displacement, construct the displacement time series data corresponding to the specific target; It should be noted that displacement time series data refers to a sequence of three-dimensional displacement observations of the same target at different sampling times, arranged in chronological order. Each data point includes a timestamp (year / month / day / hour / minute / second) and the corresponding three-dimensional displacement component or cumulative displacement value.

[0097] In this embodiment, the edge AI image processing module maintains a fixed-length queue (e.g., capable of storing the most recent 1000 data points) in memory or non-volatile storage for each identified target. After each step 1 completes the 3D displacement calculation for the current frame, the module reads the system clock to obtain the current timestamp and combines this timestamp with the 3D displacement value or cumulative displacement value (relative to the initial reference frame) to form a data record. This record is appended to the tail of the queue corresponding to that target. If the queue is full, the oldest data point is discarded according to the first-in, first-out principle. At this point, the displacement time series data of the target is constructed, ready for subsequent steps to perform filtering and trend analysis.

[0098] Step 3: Perform a filtering operation on the displacement time series data to smooth out random errors and obtain filtered displacement time series data; It should be noted that filtering refers to the process of using signal processing algorithms to suppress noise components in a time series and extract the true signal. Kalman filtering can be considered a preferred method.

[0099] It should be noted that random error refers to minute measurement deviations without a fixed pattern caused by factors such as image noise, illumination fluctuations, slight target jitter, and subpixel positioning dispersion. This type of error manifests as high-frequency random fluctuations in the displacement time series.

[0100] It should be noted that the filtered displacement time series data refers to the smoothed series data that has been processed by the filtering algorithm to remove most of the random errors and better reflect the true deformation trend of the slope.

[0101] In this embodiment, the edge AI image processing module sequentially performs Kalman filtering on the displacement time series data of each target constructed in step 2. Kalman filtering is a recursive optimal estimation algorithm, the process of which includes: first, establishing the system's state equation (displacement, velocity, etc.) and observation equation (actual measured values). For the observation value at the current moment (i.e., the three-dimensional displacement measurement value in step 2), the filtering algorithm combines the state estimate and covariance matrix of the previous moment to calculate the Kalman gain, and then performs weighted correction on the current observation value, outputting the optimal state estimate (smoothed displacement value) at the current moment. At the same time, the covariance matrix is ​​updated, and the recursion proceeds to the next moment. Through repeated iterations, the algorithm gradually filters out high-frequency random noise, making the displacement sequence tend to be smooth. If the system's computing resources are limited, a simplified moving average filter or an exponentially weighted moving average can also be used as an alternative. Finally, the filtered displacement time series data is output.

[0102] Understandably, this step improves the signal-to-noise ratio and stability of the displacement data. It extracts the true deformation signal from the noise-contaminated raw measurements, avoiding false alarms caused by errors in a single measurement.

[0103] Step 4: Extract the displacement change trend based on the filtered displacement time series data.

[0104] It should be noted that the displacement change trend refers to the overall direction and dynamic characteristics of slope deformation over time, specifically including the deformation rate (the amount of displacement change per unit time, such as mm / day), acceleration (the rate of change of the deformation rate, such as mm / day²), and whether the deformation tends to converge (stability), be uniform, or accelerate (a precursor to instability). These trend parameters are the core basis for early warning judgment.

[0105] In this embodiment, the edge AI image processing module performs trend analysis on the filtered displacement sequence output in step 3. First, it selects several recent data points (e.g., the last 10 sampling points) to fit a straight line or a quadratic curve: if linear fitting is used, the slope obtained is the current deformation rate (mm / day); if quadratic fitting is used, the coefficient of the quadratic term can be used to estimate the acceleration. Second, it calculates the relationship between the cumulative displacement and the preset stability threshold: if the cumulative displacement continues to increase and the rate gradually decreases, it indicates that the deformation tends to converge; if the rate remains constant, it indicates uniform deformation; if the rate increases with time (positive acceleration), it indicates that it has entered the accelerated deformation stage, which is an important signal of instability precursor. The extracted trend parameters (rate, acceleration, stable / uniform / accelerated state) along with the filtered displacement data are output as the displacement change trend, which is sent by step A300 to the backend monitoring platform for early warning judgment.

[0106] Understandably, this step elevates the filtered numerical sequence to a physically meaningful description of deformation behavior. It transforms the raw displacement data into trend indicators directly usable for engineering decision-making, providing a quantitative basis for tiered early warning systems.

[0107] Based on the first and second embodiments of this application, in the third embodiment of this application, the content that is the same as or similar to that in embodiments one and two above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 3 , Figure 3 This is a flowchart illustrating the third embodiment of the high slope deformation monitoring method based on edge AI in this application.

[0108] In this embodiment, the high slope deformation monitoring method based on edge AI includes steps S10~S20: Step S10: In response to the high slope surface deformation monitoring command, receive the displacement data corresponding to a specific target collected in real time by the front-end acquisition unit, wherein the specific target is fixed at a key position on the slope surface, and the key position on the slope surface includes the potential slip surface and both sides of the crack. It should be noted that the high slope surface deformation monitoring command refers to the control signal issued by the backend monitoring platform or user through a remote control client (such as a web or app interface) to start or adjust the monitoring task. This command may include specific operational requirements such as starting monitoring, stopping monitoring, and changing the sampling frequency.

[0109] It should be noted that the front-end acquisition unit refers to the acquisition system deployed on the slope, specifically comprising a high-resolution image acquisition module (industrial area scan camera), an edge AI image processing module, and a data communication module. Its function is to autonomously complete image acquisition, target recognition, displacement calculation, and output displacement data.

[0110] It should be noted that the specific target refers to a specially designed optical target with a unique identification code pattern and retroreflective optical properties, which is deployed at key locations on the slope surface that need to be monitored. This target is a passive device and requires no power supply or signal cable.

[0111] It should be noted that the specific target employs a three-layer composite design: the bottom layer is a rigid substrate; the middle layer is a high-precision printed geometric pattern layer, the pattern consisting of high-contrast asymmetric coded graphics (such as the improved ArUco marker) and concentric rings for sub-pixel center positioning; the surface layer is an anti-reflective microprism film or a high-reflectivity diffuse reflection coating. This design ensures that the target has a unique, machine-recognizable identity (ID) and stable, high-brightness optical characteristics in the image, significantly improving recognition robustness and positioning accuracy under complex backgrounds, low illumination, or partially contaminated conditions.

[0112] It should be noted that the displacement data refers to the actual spatial displacement value calculated by the edge AI image processing module, including three-dimensional displacement and displacement change trend, and the data volume is extremely small.

[0113] It should be noted that key locations on the slope surface refer to the characteristic areas that control slope stability or are most likely to deform. Specifically, these include potential slip surfaces (i.e., the projection area of ​​the interface on the slope surface where shear sliding may occur) and the sides of cracks (i.e., the locations on either side of existing or potential cracks). Displacement changes at these locations best reflect the precursors to slope instability.

[0114] In this embodiment, the backend monitoring platform or user issues a "high slope surface deformation monitoring command" via a remote control client. The communication module in the front-end acquisition unit continuously listens for the command. Once the command is received, it begins to operate according to a preset workflow: the edge AI image processing module in the front-end acquisition unit has been pre-calibrated and deployed with a lightweight AI model. It controls the camera to acquire target images at a preset frequency (e.g., once per minute) and calculates displacement data in real time. The data communication module transmits this displacement data to the backend monitoring platform in real time via 4G / 5G, wired, or LoRa wireless methods. In this step, the backend platform passively receives this data and can simultaneously verify data integrity and timestamps.

[0115] Understandably, this step enables edge processing and lightweight transmission of monitoring data. Its purpose is to upload the displacement results (rather than the original image) calculated by the front-end acquisition unit to the back-end, thereby avoiding the remote transmission of large volumes of original image data.

[0116] Step S20: Based on the displacement data, the deformation of the high slope surface is monitored in real time to obtain the monitoring results of the high slope surface deformation.

[0117] It should be noted that the high slope surface deformation monitoring results refer to the comprehensive information output after processing by the back-end monitoring platform, including but not limited to: displacement-time curves of each specific target point, cumulative displacement, deformation rate, whether the warning threshold is exceeded, three-dimensional visualized deformation field, and graded warning information (SMS, platform pop-up, etc.).

[0118] In this embodiment, after receiving the displacement data from step S10, the backend monitoring platform first stores and archives the data. Then, the platform's built-in digital twin and early warning system maps the real-time displacement data of each target onto a pre-constructed three-dimensional geological model of the slope, generating a dynamic deformation field and displaying the cumulative displacement and deformation rate of each point in real time in the form of curves or cloud maps. When the displacement or deformation rate of any monitoring point exceeds the threshold, the corresponding level of alarm information (e.g., yellow warning, orange warning, red warning) is automatically triggered, and the management personnel are notified via SMS, platform pop-ups, or APP push notifications. Finally, all processed data and early warning records together constitute the "High Slope Surface Deformation Monitoring Results," which can be queried by users, exported as reports, or used as a basis for subsequent decision-making.

[0119] Understandably, this step transforms raw displacement values ​​into engineering-ready monitoring conclusions. It assigns physical meaning and risk assessment to discrete, numerical displacement data, enabling quantitative evaluation and proactive early warning of slope safety status. A digital twin model provides three-dimensional visualization of the deformation process, intuitively displaying the overall slope deformation trend. Multi-level threshold automatic hierarchical early warning avoids missed and false alarms, improving the timeliness and reliability of early warnings. All data is traceable and analyzable, providing long-term data support for slope stability assessment and remediation decisions. Furthermore, because the front end filters out a large amount of noise and invalid information, the back end processing burden is extremely light, allowing multiple front-end acquisition units to connect simultaneously, demonstrating good system scalability.

[0120] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the high slope deformation monitoring method based on edge AI in this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0121] This application also proposes a high slope surface deformation monitoring device based on edge AI, applied to a front-end acquisition unit, wherein the high slope surface deformation monitoring device based on edge AI includes: The acquisition module is used to acquire images of a specific target in real time based on a preset intelligent gimbal, and obtain the specific target image. An image processing module is used to process the image of the specific target based on a preset lightweight AI model to obtain displacement data corresponding to the specific target. The sending module is used to send the displacement data to the backend monitoring platform, so that the backend monitoring platform can monitor the deformation of the high slope surface in real time based on the displacement data and obtain the high slope surface deformation monitoring results.

[0122] The edge AI-based high slope deformation monitoring device provided in this application employs the edge AI-based high slope deformation monitoring method described in the above embodiments, and can solve the technical problem of edge AI-based high slope deformation monitoring. Compared with related technologies, the beneficial effects of the edge AI-based high slope deformation monitoring device provided in this application are the same as those of the edge AI-based high slope deformation monitoring method provided in the above embodiments, and other technical features in the edge AI-based high slope deformation monitoring device are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0123] This application provides a high slope deformation monitoring device based on edge AI. The high slope deformation monitoring device based on edge AI includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the high slope deformation monitoring method based on edge AI in the above embodiment 1.

[0124] The following is for reference. Figure 4 This document illustrates a structural schematic diagram of a high slope deformation monitoring device based on edge AI, suitable for implementing embodiments of this application. The edge AI-based high slope deformation monitoring device in this application embodiment may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), and vehicle terminals (e.g., vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. Figure 4 The edge AI-based high slope deformation monitoring device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0125] like Figure 4As shown, the edge AI-based high slope deformation monitoring device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the edge AI-based high slope deformation monitoring device. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the edge AI-based high slope deformation monitoring equipment to wirelessly or wiredly communicate with other devices to exchange data. Although the figure shows an edge AI-based high slope deformation monitoring equipment with various systems, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems can be implemented alternatively.

[0126] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0127] The edge AI-based high slope deformation monitoring device provided in this application, employing the edge AI-based high slope deformation monitoring method described in the above embodiments, can solve the technical problems of edge AI-based high slope deformation monitoring. Compared with related technologies, the beneficial effects of the edge AI-based high slope deformation monitoring device provided in this application are the same as those of the edge AI-based high slope deformation monitoring method provided in the above embodiments, and other technical features of this edge AI-based high slope deformation monitoring device are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0128] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0129] 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 scope of the technology 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.

[0130] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, which are used to execute the edge AI-based high slope deformation monitoring method in the above embodiments.

[0131] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0132] The aforementioned computer-readable storage medium may be included in an edge AI-based high slope deformation monitoring device; or it may exist independently and not be assembled into an edge AI-based high slope deformation monitoring device.

[0133] The aforementioned computer-readable storage medium carries one or more programs that, when executed by the edge AI-based high slope deformation monitoring device, cause the edge AI-based high slope deformation monitoring device to: Based on a pre-set intelligent gimbal, images of a specific target are acquired in real time to obtain specific target images; Based on a preset lightweight AI model, the image of the specific target is processed to obtain the displacement data corresponding to the specific target. The displacement data is sent to the backend monitoring platform so that the backend monitoring platform can monitor the deformation of the high slope surface in real time based on the displacement data and obtain the high slope surface deformation monitoring results.

[0134] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0135] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0136] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0137] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described edge AI-based high slope deformation monitoring method, thereby solving the technical problem of edge AI-based high slope deformation monitoring. Compared with related technologies, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the edge AI-based high slope deformation monitoring method provided in the above embodiments, and will not be repeated here.

[0138] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the edge AI-based high slope deformation monitoring method described above.

[0139] The computer program product provided in this application can solve the technical problem of monitoring high slope deformation based on edge AI. Compared with related technologies, the beneficial effects of the computer program product provided in this application are the same as those of the edge AI-based high slope deformation monitoring method provided in the above embodiments, and will not be repeated here.

[0140] The above description is only a part of the embodiments of this application and does not limit the scope of protection of this application. All equivalent structural transformations made under the technical concept of this application and using the content of this application specification and drawings, or direct / indirect applications in other related technical fields, are included in the scope of protection of this application.

Claims

1. A method for monitoring slope deformation of high slopes based on edge AI, characterized in that, The method for monitoring high slope deformation based on edge AI includes: The front-end acquisition unit is integrated into a protective enclosure and installed on a stable reference point. Based on a preset intelligent gimbal, a specific target image is acquired in real time to obtain a specific target image. The specific target image corresponding to the specific target image is a specially made optical target. The specially made optical target has a unique identification code pattern and retroreflective optical characteristics. The specially made optical target adopts a three-layer composite design. The bottom layer is a rigid substrate, the middle layer is a high-precision printed geometric pattern layer, and the pattern in the geometric pattern layer is composed of a high-contrast asymmetric coded graphics and concentric rings for sub-pixel center positioning. The surface layer is a retroreflective microprism film or a high-reflectivity diffuse reflection coating. Based on a preset lightweight AI model, the image of the specific target is processed to obtain the displacement data corresponding to the specific target. The displacement data is sent to the backend monitoring platform so that the backend monitoring platform can monitor the deformation of the high slope surface in real time based on the displacement data and obtain the high slope surface deformation monitoring results. The step of acquiring a specific target image in real time based on a preset intelligent gimbal to obtain the specific target image further includes: Obtain environmental information; Based on the environmental information, the preset active illumination system, and the preset intelligent gimbal, a high dynamic range industrial area array camera is used to acquire images of a specific target in real time, thereby obtaining the specific target image. The preset active illumination system consists of an infrared LED array of a specific wavelength. The preset active illumination system is triggered synchronously with the exposure of the high dynamic range industrial area array camera. The specific wavelengths include 850 nm and 940 nm. The infrared light corresponding to the specific wavelengths is invisible or weakly visible to the human eye. The infrared light corresponding to the specific wavelengths is matched with the retroreflective microprism film or high reflectivity coating on the target surface to improve the contrast of the target in the image.

2. The method for monitoring high slope surface deformation based on edge AI as described in claim 1, characterized in that, The step of processing the specific target image based on a preset lightweight AI model to obtain the displacement data corresponding to the specific target further includes: The specific target image is preprocessed to obtain a preprocessed specific target image. The preprocessing includes automatic white balance and HDR synthesis. The preprocessing is used to suppress the influence of uneven illumination on the specific target image. Based on a pre-defined lightweight AI model, the pre-processed image of a specific target is processed to obtain the displacement data corresponding to the specific target.

3. The method for monitoring high slope surface deformation based on edge AI as described in claim 2, characterized in that, The step of processing the preprocessed image of a specific target based on a preset lightweight AI model to obtain displacement data corresponding to the specific target further includes: A preset lightweight AI model is used to perform pixel-level segmentation on the preprocessed specific target image to accurately extract the pixel region of the specific target. The pixel region is a binary region. The preset lightweight AI model is obtained after training on a large dataset of specific target images in multiple outdoor scenarios. Perform morphological filtering on the pixel region to remove noise, and obtain the pixel region after noise removal; Based on the pixel regions after noise removal, displacement data corresponding to a specific target is obtained.

4. The method for monitoring high slope surface deformation based on edge AI as described in claim 3, characterized in that, The step of obtaining the displacement data corresponding to a specific target based on the noise-removed pixel region further includes: Based on the specific target, the high dynamic range industrial area array camera is calibrated to obtain the intrinsic and extrinsic parameters of the high dynamic range industrial area array camera. The intrinsic parameters include focal length and distortion coefficient, and the extrinsic parameters include spatial position and orientation relative to the monitoring area. The physical dimensions of the specific target are known. Based on the noise-removed pixel region and the preset centroid calculation algorithm, the sub-pixel level coordinates of the centroid of the noise-removed pixel region in the image coordinate system are calculated. The preset centroid calculation algorithm includes gray-level weighted centroid method and ellipse fitting. Based on the sub-pixel level coordinates of the centroid in the image coordinate system corresponding to the current frame and the sub-pixel level coordinates of the centroid in the image coordinate system corresponding to the previous frame, the two-dimensional displacement of the specific target is determined. Based on the intrinsic parameters, the extrinsic parameters, the physical dimensions of the specific target, and the preset perspective projection geometry model, the two-dimensional displacement is processed to obtain the displacement data corresponding to the specific target. The displacement data includes three-dimensional displacement and displacement change trend. The three-dimensional displacement is the actual displacement of the specific target on the plane perpendicular to the optical axis.

5. The method for monitoring high slope surface deformation based on edge AI as described in claim 4, characterized in that, The step of processing the two-dimensional displacement based on the intrinsic parameters, the extrinsic parameters, the physical dimensions of the specific target, and a preset perspective projection geometric model to obtain the displacement data corresponding to the specific target further includes: Based on the intrinsic parameters, the extrinsic parameters, the physical dimensions of the specific target, and the preset perspective projection geometric model, the two-dimensional displacement is calculated to obtain the three-dimensional displacement corresponding to the specific target; Based on the three-dimensional displacement, the displacement time series data corresponding to the specific target is constructed; A filtering operation is performed on the displacement time series data to smooth out random errors, resulting in filtered displacement time series data. Based on the filtered displacement time series data, the displacement change trend is extracted.

6. A high slope deformation monitoring device based on edge AI, characterized in that, The edge AI-based high slope deformation monitoring device includes: The acquisition module is used to acquire images of a specific target in real time based on a preset intelligent gimbal, and the specific target corresponding to the specific target image is a special optical target. The special optical target has a unique identification code pattern and retroreflective optical characteristics. The special optical target adopts a three-layer composite design. The bottom layer is a rigid substrate, the middle layer is a high-precision printed geometric pattern layer, the pattern in the geometric pattern layer is composed of high-contrast asymmetric coding graphics and concentric rings for sub-pixel center positioning, and the surface layer is a retroreflective microprism film or a high-reflectivity diffuse reflection coating. An image processing module is used to process the image of the specific target based on a preset lightweight AI model to obtain displacement data corresponding to the specific target. The sending module is used to send the displacement data to the backend monitoring platform, so that the backend monitoring platform can monitor the deformation of the high slope surface in real time based on the displacement data and obtain the high slope surface deformation monitoring results. The acquisition module is also used to achieve: Obtain environmental information; Based on the environmental information, the preset active illumination system, and the preset intelligent gimbal, a high dynamic range industrial area array camera is used to acquire images of a specific target in real time, thereby obtaining the specific target image. The preset active illumination system consists of an infrared LED array of a specific wavelength. The preset active illumination system is triggered synchronously with the exposure of the high dynamic range industrial area array camera. The specific wavelengths include 850 nm and 940 nm. The infrared light corresponding to the specific wavelengths is invisible or weakly visible to the human eye. The infrared light corresponding to the specific wavelengths is matched with the retroreflective microprism film or high reflectivity coating on the target surface to improve the contrast of the target in the image.

7. A high slope surface deformation monitoring device based on edge AI, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the edge AI-based high slope deformation monitoring method as described in any one of claims 1 to 5.

8. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the high slope deformation monitoring method based on edge AI as described in any one of claims 1 to 5.

9. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps of the edge AI-based high slope deformation monitoring method as described in any one of claims 1 to 5.

Images