Landslide early warning device and method based on image recognition and computer vision

Abstract

The application discloses a landslide early warning device and method based on image recognition and computer vision, relates to the technical field of geological disaster monitoring, and comprises an anchor rod, an energy collection intelligent sensing component and a visual verification active protection component. The top of the anchor rod is respectively provided with the energy collection intelligent sensing component and the visual verification active protection component. The energy collection intelligent sensing component comprises a flexible tail wing, a blunt body, two powerful magnets, two electromagnetic induction coils, a piezoelectric cantilever beam and a counterweight magnet. The energy collection intelligent sensing component and the visual verification active protection component are in deep cooperation, realizing full-process intelligent operation of slope monitoring from energy autonomy to intelligent diagnosis. The application is different from a traditional extensive scheme which relies on external power supply, a single sensor and fixed angle monitoring, and makes the landslide early warning process more reliable, more accurate, more adaptive and more durable. The application completely breaks through the inherent contradiction in the traditional scheme that continuous power supply, reliable sensing and accurate early warning cannot be achieved simultaneously.

Description

Technical Field

[0001] This invention relates to the field of geological disaster monitoring technology, specifically to a landslide early warning device and method based on image recognition and computer vision. Background Technology

[0002] Image recognition and computer vision technology for landslide early warning is a key technology that uses visual imaging and intelligent algorithms to capture precursors of landslides such as subtle changes in the ground surface and crack propagation. It has been widely applied in fields such as geological disaster monitoring and slope engineering safety control. Its core logic is to use image sensors to collect visual information of the monitoring area, analyze pixel-level changes through computer vision algorithms, and combine this with physical sensor data to assess landslide risk.

[0003] In complex and demanding monitoring scenarios such as mountain slopes and open-pit mines, existing technologies often rely on a passive combination of externally powered sensor networks and fixed-viewpoint visual monitoring to address the end-to-end early warning requirements for deep deformation, surface micro-changes, and visual verification. Operators typically attempt to achieve a global assessment of slope stability by simply overlaying preset fixed thresholds and independent data sources. However, this traditional early warning mode has inherent systemic flaws. Under extreme conditions such as drastic temperature changes, torrential rain erosion, strong wind disturbances, dust cover, and continuous micro-vibrations in mountainous areas, multiple intertwined and irreconcilable technical contradictions arise. Firstly, at the energy supply level, a fundamental conflict exists between the high-power continuous monitoring requirements and limited energy supply capacity. The performance of solar panels or lithium batteries, which are relied upon by traditional solutions, is significantly constrained by weather and obstructions, making it difficult to match the high-power requirements of continuous visual monitoring. This energy imbalance directly leads to data acquisition interruptions and early warning failures. The current approach suffers from several drawbacks. Firstly, it fails to meet the millisecond-level response requirements for landslide precursors, and secondly, frequent manual maintenance limits large-scale deployment in remote mountainous areas. Thirdly, at the perception architecture level, it exposes a systemic contradiction between the need for precise panoramic perception and the static, rigid structure. On one hand, the fixed-view design has inherent blind spots, failing to achieve full slope coverage and easily missing key deformation precursors. On the other hand, exposed optical lenses are susceptible to dust contamination and rain erosion, severely degrading image quality and hindering pixel-level micro-change recognition accuracy. More importantly, the camera itself experiences reference drift due to slight slope displacement, and its errors, combined with signal interference from physical sensors, further amplify the overall perception uncertainty of the system. In existing solutions, physical sensors and the vision system operate independently, lacking an effective collaborative verification mechanism. This makes it impossible to verify the validity of visual recognition results through physical signals, and also difficult to use visual evidence to rule out false alarms from physical sensors. This data silo effect prevents the complementary nature of multi-source information from being fully utilized, severely restricting the fundamental improvement of early warning accuracy.

[0004] Therefore, we propose a landslide early warning device and method based on image recognition and computer vision to solve the problems mentioned above. Summary of the Invention

[0005] The purpose of this invention is to provide a landslide early warning device and method based on image recognition and computer vision, which realizes intelligent operation of the entire process of slope monitoring from energy autonomy to intelligent diagnosis. It is different from the traditional extensive solution that relies on external power supply, single sensor and fixed viewing angle monitoring, making the landslide early warning process more reliable, more accurate, more adaptive and more durable, and completely breaking through the inherent contradiction of continuous power supply, reliable perception and accurate early warning in the traditional solution.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a landslide early warning device based on image recognition and computer vision, comprising an anchor bolt, an energy harvesting intelligent sensing component, and a visual verification active protection component, wherein the top of the anchor bolt is respectively equipped with the energy harvesting intelligent sensing component and the visual verification active protection component; The energy harvesting intelligent sensing component includes a flexible tail fin, a blunt body, two powerful magnets, two electromagnetic induction coils, a piezoelectric cantilever beam, a counterweight magnet, and a fixed magnet. The flexible tail fin is used to trigger vortex-induced vibrations using mountain winds. The blunt body is used to stabilize the generation of the Karman vortex street and enhance the regularity and continuity of the flexible tail fin's oscillation. The two powerful magnets and two electromagnetic induction coils are used to generate induced currents to provide the main power for the system. The piezoelectric cantilever beam is used to sense micro-vibrations in the mountain to induce mechanical deformation and to convert mechanical energy into electrical energy using the piezoelectric effect. The counterweight magnet and the fixed magnet are used to keep the piezoelectric cantilever beam in a pre-bent state, and the counterweight magnet is used to amplify the deformation amplitude of the piezoelectric cantilever beam during vibration. The visual verification active protection component includes eight miniature electromagnets, eight protective covers, and eight driving magnets. The eight miniature electromagnets and eight driving magnets are used to generate a repulsive force when energized, which pushes the eight protective covers to move in a straight line.

[0007] Preferably, a first flange is fixedly welded to the top of the anchor rod, and the anchor rod is inserted below the landslide surface.

[0008] Preferably, the energy harvesting intelligent sensing component further includes a second flange, the bottom of which is bolted to the top of the first flange. An anti-slip washer is connected to the top of the second flange, and the anti-slip washer is used for sealing and moisture protection and buffering installation stress. A protective shell is connected to the top of the anti-slip washer, and a top cover is fixedly connected to the top of the protective shell. A bearing seat is bolted to the top of the top cover, and four mounting holes are opened on the top circumference of the bearing seat.

[0009] Preferably, a rotating shaft is inserted between the inner surfaces of the bearing housing, and a swing shaft is fixedly connected to the top of the rotating shaft. One end of the outer wall of the swing shaft is fixedly connected to one end of the outer wall of the flexible tail fin. The flexible tail fin adopts a dovetail structure and the bifurcated wing surface is completely symmetrically stressed. The flexible tail fin and the opposite side of the blunt body are fixedly connected. Two strong magnets are embedded and connected to both sides of the outer wall of the flexible tail fin. The bottom of the top cover is symmetrically connected to annular fixing sleeves, and the inner surface of a corresponding electromagnetic induction coil is sleeved and connected to the annular groove on the outer surface of each annular fixing sleeve.

[0010] Preferably, the inner surface of the protective shell is bolted with a fixing clamp and a plastic support, and the fixing clamp is placed above the plastic support. The inner surface wall of the fixing clamp is fixedly connected to one end of the outer wall of the piezoelectric cantilever beam, and the bottom of the piezoelectric cantilever beam is connected to the top of the counterweight magnet. The top of the plastic support is connected to the bottom of the fixing magnet, and the counterweight magnet and the fixing magnet are arranged symmetrically.

[0011] Preferably, the bottom of the inner wall of the protective shell is connected to a main circuit board, two supercapacitors, and a thin-film battery compartment. The top of the main circuit board integrates a MEMS inclinometer and a multi-functional accelerometer. The MEMS inclinometer is used to capture the tilt angle of the top of the anchor bolt in real time, and the multi-functional accelerometer is used to continuously monitor environmental micro-vibrations. The main circuit board also integrates a communication module, a positioning module, an energy management circuit, and a microcontroller. The two supercapacitors are used to store electrical energy. The thin-film battery compartment encapsulates a thin-film battery, which is used for emergency power supply in extreme windless and vibration-free environments.

[0012] Preferably, the visual verification active protection component further includes a visual chamber, four ball bearings, and a transmission gear. The bottom of the visual chamber is connected to the top of the top cover, and a sealed cover is fixedly connected to the top of the visual chamber. Monitoring windows are provided on all four sides of the outer wall of the visual chamber, and sliding grooves are symmetrically provided on all four sides of the outer wall of the visual chamber. The inner surface of each of the four sliding grooves is fixedly connected to the outer surface of two miniature electromagnets.

[0013] Preferably, each of the inner walls of the slide groove is connected to two rollers in a rolling manner, and a solid column is inserted between the inner surfaces of each roller. One end of the outer wall of every four solid columns is connected to one side of the outer wall of a corresponding protective cover. One side of the outer wall of each protective cover is embedded and connected to a corresponding driving magnet. The outer surfaces of the sixteen solid columns are fitted with fixing blocks. One side of the outer wall of each fixing block is elastically connected to a set of miniature springs, and each set of miniature springs is elastically connected to one end of the inner wall of a corresponding slide groove. Every two sets of miniature springs are used to provide a closing and restoring force for a corresponding protective cover.

[0014] Preferably, the outer walls of the visual cabin are symmetrically connected with connecting plates on all four sides. Each connecting plate and the opposite side of the protective cover are connected with a scraper, and each scraper is used to scrape and clean the outer surface of a corresponding protective cover. The outer surfaces of the four ball bearings are fixedly connected to the inner surfaces of the four mounting holes. The inner surface of each ball bearing is rotatably connected to a rotating rod. The outer surface of each rotating rod is fitted with a driven gear, and the four driven gears are meshed with the outer walls of the transmission gears. The inner surface of the transmission gear is rotatably connected to the outer surface of the rotating shaft. The top of each rotating rod is fitted with a support frame, and a miniature camera is bolted to one end of the outer wall of each support frame. Each miniature camera is directly facing the central axis of a corresponding monitoring window.

[0015] A landslide early warning method based on image recognition and computer vision includes the following steps: Step 1: Insert and fix the anchor bolts below the landslide surface. Connect the first flange and the second flange with bolts to complete the integrated installation of the energy harvesting intelligent sensing component and the visual verification active protection component. The anti-slip gasket achieves sealing and moisture protection as well as installation stress buffering. After the device is started, the hybrid energy harvesting module enters a continuous working state. Mountain winds trigger vortex-induced vibration of the flexible tail fin. The blunt body stabilizes the Karman vortex street, which enhances the regularity of the oscillation. This causes the powerful magnets on both sides of the flexible tail fin to pass over the electromagnetic induction coil outside the annular fixed sleeve, cutting the magnetic field lines to generate induced current (main energy). At the same time, the micro-vibration of the mountain causes the piezoelectric cantilever beam to deform. The repulsive force between the counterweight magnet and the fixed magnet puts the beam in a pre-bending sensitive state, amplifying the deformation amplitude and converting it into auxiliary electrical energy through the piezoelectric effect. After being optimized by the energy management circuit on the main circuit board, the electrical energy is stored in two supercapacitors. In extreme windless and vibration-free environments, the thin-film battery in the thin-film battery compartment provides emergency power to ensure the system's continuous energy autonomy. Step 2: The MEMS inclinometer on the main circuit board captures the tilt angle of the top of the anchor rod in real time, and the multi-purpose accelerometer continuously monitors the environmental micro-vibration signal. Both are in low-power operation mode. When the MEMS inclinometer detects a tilt angle in the anchor rod and the multi-functional accelerometer captures a micro-vibration signal, the microcontroller triggers a system wake-up command, and the energy management circuit switches to the working mode to power the visual verification active protection component. Step 3: The microcontroller sends pulse currents to eight miniature electromagnets. The eight miniature electromagnets and the driving magnet on the protective cover generate a repulsive force, pushing the protective cover to move in a straight line along the slide, quickly opening the monitoring windows on all four sides of the visual cabin. At the same time, the miniature springs are stretched to store energy. The swing of the flexible tail fin drives the transmission gear to rotate through the rotating shaft. The meshing transmission drives four passive gears and rotating rods to rotate synchronously. The miniature camera on the support frame rotates coaxially around the anchor rod to take 360° panoramic pictures of the ground outside the monitoring window and realize wind speed adaptive sampling. During the shooting process, the scraper on the connecting plate cleans the outer surface of the protective cover as the protective cover moves, ensuring image clarity. The image data is processed in real time by the image stitching control board, stitched into a complete panoramic image and compressed for storage. After filming is completed, the miniature electromagnet is de-energized, and the miniature spring releases its elastic force to pull the protective cover back to its original position and close, completing the secondary cleaning and sealing protection. Step 4: The main circuit board integrates multi-source data: macroscopic deformation data from the MEMS inclinometer, microscopic vibration data from the multiplexed accelerometer, and panoramic visual data from the miniature camera; The microcontroller performs cross-validation through a preset algorithm, and combines the expansion of surface cracks in the visual image to verify the validity of the physical sensor data. It eliminates environmental interference in the visual data through physical signals and eliminates false alarms caused by data silos. After determining the landslide risk level through data fusion, the communication module transmits the early warning information to the remote monitoring center, and the positioning module simultaneously sends the device deployment location information; The monitoring center outputs corresponding early warning commands based on the risk level to achieve early warning of landslides. At the same time, the system returns to a low-power sleep state, waiting for the next trigger.

[0016] Compared with the prior art, the beneficial effects of the present invention are: In this invention, the deep synergy between the energy harvesting intelligent sensing component and the visual verification active protection component enables fully intelligent operation of slope monitoring, from energy autonomy to intelligent diagnosis. This differs from traditional, rudimentary solutions that rely on external power supply, single sensors, and fixed-viewpoint monitoring. It makes the landslide early warning process more reliable, accurate, adaptive, and durable, completely overcoming the inherent contradiction in traditional solutions where continuous power supply, reliable sensing, and accurate early warning are mutually exclusive. Firstly, the energy harvesting and intelligent sensing components, through the synergistic innovation of wind and vibration complementary energy harvesting and multimodal cross-verification, construct an integrated system of energy autonomy and closed-loop sensing, fundamentally solving the problem. This solution overcomes the energy supply bottlenecks and data fragmentation issues inherent in traditional solutions. The hybrid energy harvesting module, composed of a powerful magnet, electromagnetic induction coil, and piezoelectric cantilever beam, enhances Karman vortex street stability through optimized blunt-body design of vortex-induced vibration and magnetoelectric power generation units. Combined with a gear transmission mechanism consisting of drive and driven gears, it achieves efficient conversion of wind energy into electrical energy, providing a stable primary energy source for the system. Furthermore, the counterweight magnet and fixed magnet, through a repulsive pre-tightening design using like-pole magnets, keep the piezoelectric cantilever beam in a sensitive state, significantly improving the efficiency of environmental vibration energy harvesting. Together, they form an all-weather energy supply mode, complemented by hybrid energy storage using supercapacitors and thin-film batteries. The system ensures stable operation even in extreme environments such as continuous rainy days. Through collaborative monitoring of MEMS inclinometers and multiplexed accelerometers, a dual sensing mechanism for macroscopic deformation and micro-vibration is constructed. Combined with the correlation with visual data, signal interference and the risk of missed precursors from single sensors are completely eliminated, achieving accurate capture of landslide precursors across the entire process. Simultaneously, the visual verification and active protection components, through precise coordination of gear transmission and intelligent protection design, achieve triple optimization of panoramic monitoring, low-power operation, and environmental adaptability, completely solving the blind spots of traditional visual monitoring. This is achieved through a transmission gear driven by the swinging of the flexible tail fin. With the passive gear meshing and linkage of the miniature camera, an innovative wind speed adaptive sampling mechanism is realized, which not only matches the landslide triggering conditions but also avoids ineffective energy consumption. The four miniature cameras facing the four directions rotate synchronously, and 360° panoramic coverage without blind spots is achieved through image stitching. The design of the miniature camera rotating coaxially around the anchor rod ensures that the visual benchmark and the physical benchmark are completely identical, completely solving the benchmark drift problem of fixed positions. Secondly, the protective cover is guided by the sliding groove and linked in the same direction with the electromagnetic drive. Combined with the self-cleaning function of the scraper, it builds an intelligent mechanism of on-demand opening, automatic cleaning, and sealing protection. After being triggered, it quickly opens for imaging, and the protective cover is cleaned when it is reset. Attached Figure Description

[0017] Figure 1 This is a three-dimensional view of the main structure of the landslide early warning device based on image recognition and computer vision according to the present invention; Figure 2 This is a three-dimensional view of the structure from below in the landslide early warning device based on image recognition and computer vision according to the present invention; Figure 3 This is a diagram showing the positional relationship between the energy harvesting intelligent sensing component and the visual verification active protection component in the landslide early warning device based on image recognition and computer vision according to the present invention. Figure 4 This is a schematic diagram of the installation location of the intelligent sensing component for energy harvesting in the landslide early warning device based on image recognition and computer vision according to the present invention. Figure 5 This is a schematic diagram showing the installation position of the anti-slip pad, protective shell, and top cover in the landslide early warning device based on image recognition and computer vision according to the present invention. Figure 6 This is a schematic diagram of the installation position structure of the rotating shaft and the swing shaft in the landslide early warning device based on image recognition and computer vision of the present invention; Figure 7 This is a schematic diagram of the installation position structure of the flexible tail fin, blunt body, and strong magnet in the landslide early warning device based on image recognition and computer vision according to the present invention. Figure 8 This is a schematic diagram showing the installation positions of the piezoelectric cantilever beam, counterweight magnet, and fixed magnet in the landslide early warning device based on image recognition and computer vision according to the present invention. Figure 9 for Figure 7 Enlarged 3D view of the structure at point A in the middle; Figure 10 This is a schematic diagram showing the installation positions of the main circuit board, MEMS inclinometer, and multiplexed accelerometer in the landslide early warning device based on image recognition and computer vision according to the present invention. Figure 11 This is a diagram showing the positional relationship of the visual verification active protection components in the landslide early warning device based on image recognition and computer vision, as used in this invention. Figure 12 This is a schematic diagram of the installation position structure of the monitoring window and protective cover in the landslide early warning device based on image recognition and computer vision according to the present invention. Figure 13 This is a schematic diagram of the installation position structure of the miniature camera and transmission gear in the landslide early warning device based on image recognition and computer vision according to the present invention. Figure 14 This is a schematic diagram of the installation position of the connecting plate and scraper in the landslide early warning device based on image recognition and computer vision according to the present invention; Figure 15 This is a schematic diagram of the installation position of the protective cover and rollers in the landslide early warning device based on image recognition and computer vision according to the present invention. Figure 16This is a schematic diagram of the installation position structure of the passive gear and the transmission gear in the landslide early warning device based on image recognition and computer vision according to the present invention. Figure 17 This is a schematic diagram of the installation positions of the rotating rod, passive gear, support frame, and miniature camera in the landslide early warning device based on image recognition and computer vision according to the present invention. Figure 18 for Figure 13 Enlarged 3D view of the structure at point B in the middle; Figure 19 for Figure 14 Enlarged 3D view of the structure at point C; Figure 20 for Figure 15 Enlarged 3D view of the structure at point D.

[0018] In the diagram: 100, Anchor bolt; 200, First flange; 300, Energy harvesting intelligent sensing component; 301, Second flange; 302, Anti-slip washer; 303, Protective housing; 304, Top cover; 305, Bearing seat; 306, Mounting hole; 307, Rotating shaft; 308, Swing shaft; 309, Flexible tail fin; 310, Blunt body; 311, Powerful magnet; 312, Annular fixing sleeve; 313, Electromagnetic induction coil; 314, Fixing clamp; 315, Piezoelectric cantilever beam; 316, Counterweight magnet; 317, Plastic support; 318, Fixing magnet; 319, Main circuit board; 320, MEM S-Clinometer; 321. Reusable Accelerometer; 322. Supercapacitor; 323. Thin-film Battery Compartment; 400. Vision Verification Active Protection Component; 401. Vision Cabin; 402. Sealed Cabin Cover; 403. Slide; 404. Miniature Electromagnet; 405. Monitoring Window; 406. Protective Cover; 407. Drive Magnet; 408. Solid Column; 409. Roller; 410. Fixing Block; 411. Miniature Spring; 412. Connecting Plate; 413. Scraper; 414. Ball Bearing; 415. Rotating Rod; 416. Passive Gear; 417. Support Frame; 418. Miniature Camera; 419. Transmission Gear. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] like Figures 1-2As shown, this embodiment discloses a landslide early warning device based on image recognition and computer vision, including an anchor bolt 100, an energy harvesting intelligent sensing component 300, and a visual verification active protection component 400. The top of the anchor bolt 100 is respectively equipped with the energy harvesting intelligent sensing component 300 and the visual verification active protection component 400. like Figures 7-9 As shown, the energy harvesting intelligent sensing component 300 includes a flexible tail fin 309, a blunt body 310, two powerful magnets 311, two electromagnetic induction coils 313, a piezoelectric cantilever beam 315, a counterweight magnet 316, and a fixed magnet 318. The flexible tail fin 309 is used to trigger vortex-induced vibrations using mountain winds. The blunt body 310 is used to stabilize the generation of the Karman vortex street and enhance the regularity and continuity of the flexible tail fin 309's oscillation. The two powerful magnets 311 and the two electromagnetic induction coils 313 are used to generate induced currents to provide the main power for the system. The piezoelectric cantilever beam 315 is used to sense micro-vibrations in the mountain to induce mechanical deformation and to convert mechanical energy into electrical energy using the piezoelectric effect. The counterweight magnet 316 and the fixed magnet 318 are used to keep the piezoelectric cantilever beam 315 in a pre-bent state, and the counterweight magnet 316 is used to amplify the deformation amplitude of the piezoelectric cantilever beam 315 during vibration. like Figure 18 as well as Figure 20 As shown, the visual verification active protection component 400 includes eight miniature electromagnets 404, eight protective covers 406, and eight driving magnets 407. The eight miniature electromagnets 404 and eight driving magnets 407 are used to generate a repulsive force when energized to push the eight protective covers 406 to move in a straight line.

[0021] This embodiment primarily addresses the end-to-end early warning requirements for deep deformation, surface micro-changes, and visual verification in complex and demanding monitoring scenarios such as mountain slopes and open-pit mines. Existing technologies often rely on a passive combination of externally powered sensor networks and fixed-viewpoint visual monitoring. Operators typically attempt to achieve a global assessment of slope stability by simply overlaying preset fixed thresholds and independent data sources. However, this traditional early warning mode has inherent systemic flaws. Under extreme conditions such as drastic temperature changes, torrential rain erosion, strong wind disturbances, dust cover, and continuous micro-vibrations in mountainous areas, multiple intertwined and irreconcilable technical contradictions arise. Firstly, at the energy supply level, a fundamental conflict exists between the high-power continuous monitoring requirements and limited energy supply capacity. The performance of solar panels or lithium batteries, which are relied upon by traditional solutions, is significantly constrained by weather and obstructions, making it difficult to match the high-power requirements of continuous visual monitoring. This energy imbalance directly leads to problems during data acquisition. The delayed early warning response cannot meet the millisecond-level response requirements for landslide precursors, and frequent manual maintenance further limits large-scale deployment in remote mountainous areas. Secondly, at the perception architecture level, a systemic contradiction is exposed between the need for panoramic and accurate perception and the static rigid structure. On the one hand, the fixed-view design has inherent blind spots, failing to achieve full slope coverage and easily missing key deformation precursors. On the other hand, exposed optical lenses are susceptible to dust pollution and rain erosion, severely degrading image quality and limiting pixel-level micro-change recognition accuracy. More importantly, the camera itself will experience reference drift due to slight slope displacement, and its error, combined with signal interference from physical sensors, further amplifies the overall perception uncertainty of the system. In existing solutions, physical sensors and vision systems operate independently, lacking an effective collaborative verification mechanism. This makes it impossible to verify the validity of visual recognition results through physical signals, and also difficult to use visual evidence to rule out false alarms from physical sensors. This data silo effect prevents the complementarity of multi-source information from being utilized, severely restricting the fundamental improvement of early warning accuracy.

[0022] This embodiment addresses the problems of existing technologies by deeply collaborating the energy harvesting intelligent sensing component 300 and the visual verification active protection component 400. This enables fully intelligent operation of slope monitoring, from energy autonomy to intelligent diagnosis. Unlike traditional, crude solutions relying on external power supply, single sensors, and fixed-viewpoint monitoring, this makes the landslide early warning process more reliable, accurate, adaptive, and durable, completely overcoming the inherent contradiction in traditional solutions where continuous power supply, reliable sensing, and accurate early warning are mutually exclusive. Firstly, the energy harvesting intelligent sensing component 300, through the synergistic innovation of wind and vibration complementary energy harvesting and multimodal cross-verification, constructs an integrated system of energy autonomy and closed-loop sensing, fundamentally solving the problem... This solution overcomes the energy supply bottlenecks and data fragmentation issues inherent in traditional solutions. The hybrid energy harvesting module, composed of a powerful magnet 311, an electromagnetic induction coil 313, and a piezoelectric cantilever beam 315, enhances Karman vortex street stability through eddy-induced vibration and an optimized blunt body 310 design for the magnetoelectric power generation unit. Combined with a gear transmission mechanism consisting of a transmission gear 419 and a passive gear 416, it achieves efficient conversion of wind energy into electrical energy, providing a stable primary energy source for the system. Furthermore, the counterweight magnet 316 and the fixed magnet 318, through a repulsive pre-tightening design using like-pole magnets, keep the piezoelectric cantilever beam 315 in a sensitive state, significantly improving the efficiency of environmental vibration energy harvesting. Together, they form an all-weather energy supply mode, complemented by a supercapacitor 322 and a thin-film battery. The hybrid energy storage system ensures stable operation even in extreme environments such as continuous rainy days. Through the collaborative monitoring of the MEMS inclinometer 320 and the multiplexed accelerometer 321, a dual sensing mechanism for macroscopic deformation and micro-vibration is constructed. Combined with the homologous correlation with visual data, it can completely eliminate signal interference and the risk of missed precursors from a single sensor, achieving accurate capture of landslide precursors across the entire process. Simultaneously, the visual verification and visual verification active protection component 400, through precise coordination of gear transmission linkage and intelligent protection design, achieves triple optimization of panoramic monitoring, low-power operation, and environmental adaptability, completely solving the blind spots of traditional visual monitoring. The transmission gear 419, driven by the swing of the flexible tail fin 309, and the micro-vibration... The passive gear 416 of the miniature camera 418 meshes and links, innovatively realizing a wind speed adaptive sampling mechanism, which not only matches the landslide triggering conditions but also avoids ineffective energy consumption. The four miniature cameras 418 facing the four directions rotate synchronously, achieving 360° panoramic coverage without blind spots through image stitching. Furthermore, the design of the miniature camera 418 rotating coaxially around the anchor rod 100 ensures that the visual reference and the physical reference are completely identical, completely solving the reference drift problem of fixed positions. Secondly, the protective cover 406 is guided by the slide groove 403 and linked in the same direction with the electromagnetic drive. Combined with the self-cleaning function of the scraper 413, it constructs an intelligent mechanism of on-demand opening, automatic cleaning, and sealing protection. After triggering, it quickly opens for imaging, and the protective cover 406 is cleaned when it is reset.

[0023] according to Figure 1As shown, the top of the anchor bolt 100 is fixedly welded with a first flange 200, and the anchor bolt 100 is inserted below the landslide surface.

[0024] In this embodiment of the invention, the anchor rod 100 is made of HRB400E high-strength threaded steel with a rod diameter of 32mm. It is inserted below the landslide surface and forms a rigid embedment with the deep stable rock and soil, which can accurately transmit the minute displacement and deformation of the rock and soil, providing an absolutely stable physical reference for the entire monitoring system. The first flange 200 is made of Q235B carbon steel and is fixed to the top of the anchor rod 100 by submerged arc welding, which not only ensures the connection strength but also ensures the levelness of the first flange 200. This provides a precise reference for the coaxial installation of subsequent components, fundamentally avoiding the distortion of monitoring data caused by installation deviations and solving the problem of micro-change identification error caused by unstable reference in traditional devices.

[0025] according to Figure 5 As shown, the energy harvesting intelligent sensing component 300 also includes a second flange 301. The bottom of the second flange 301 is bolted to the top of the first flange 200. An anti-slip washer 302 is connected to the top of the second flange 301. The anti-slip washer 302 is used for sealing and moisture protection and buffering installation stress. A protective shell 303 is connected to the top of the anti-slip washer 302. A top cover 304 is fixedly connected to the top of the protective shell 303. A bearing seat 305 is bolted to the top of the top cover 304. Four mounting holes 306 are opened on the top circumference of the bearing seat 305.

[0026] In this embodiment of the invention, firstly, both the second flange 301 and the first flange 200 are made of Q235B carbon steel and are fastened together with M12 high-strength bolts, which ensures the reliability of the connection and facilitates disassembly and maintenance. The anti-slip washer 302 is made of fluororubber, whose elastic deformation characteristics can effectively fill the tiny gap between the two flanges, achieving IP67-level sealing and moisture protection. At the same time, it can buffer the installation stress caused by vibration in mountainous environments and prevent the components from loosening due to long-term vibration. Secondly, the protective shell 303 is made of 6061 aluminum alloy in one piece and is anodized to improve corrosion resistance. It is connected to the top cover 304 by threads and has a silicone sealing ring embedded, thus forming an effective sealed cavity to isolate the internal power generation and sensing components from rainwater and dust corrosion. Meanwhile, the bearing seat 305 is made of 45# steel and precision machined and fixed to the top cover 304 with M8 countersunk bolts. The four mounting holes 306 on its top are evenly distributed in a circle, laying the structural foundation for the stable rotation of the rotating rod 415.

[0027] according to Figure 7 as well as Figure 9As shown, a rotating shaft 307 is inserted between the inner surfaces of the bearing housing 305. A swing shaft 308 is fixedly connected to the top of the rotating shaft 307. One end of the outer wall of the swing shaft 308 is fixedly connected to one end of the outer wall of the flexible tail fin 309. The flexible tail fin 309 adopts a dovetail structure and the bifurcated wing surface is completely symmetrically stressed. The flexible tail fin 309 and the blunt body 310 are fixedly connected to opposite sides. Two strong magnets 311 are embedded and connected to both sides of the outer wall of the flexible tail fin 309. Annular fixing sleeves 312 are symmetrically connected to the bottom of the top cover 304. The inner surface of the annular groove on the outer surface of each annular fixing sleeve 312 is sleeved and connected to the inner surface of a corresponding electromagnetic induction coil 313.

[0028] In this embodiment of the invention, the rotating shaft 307 is made of 40Cr alloy structural steel, and the clearance between it and the deep groove ball bearing in the bearing housing 305 is controlled at 0.02-0.05mm, which ensures smooth rotation while avoiding radial runout. The oscillating shaft 308 and the flexible tail fin 309 are integrally molded using glass fiber reinforced epoxy resin, improving structural rigidity and preventing breakage at the connection point during wind-induced oscillation. The dovetail-shaped symmetrical structure of the flexible tail fin 309 ensures balanced force on both sides, stabilizing the vortex-induced vibration frequency within the high-efficiency power generation range, and works in conjunction with the blunt body 310 to stabilize the Karman vortex street. The flexible tail fin 309 effectively increases its swing amplitude, thereby significantly improving power generation efficiency. Secondly, the powerful magnet 311 is made of N52 neodymium iron boron and is embedded on both sides of the flexible tail fin 309 with epoxy adhesive. When the flexible tail fin 309 swings, it forms a stable relative motion with the electromagnetic induction coil 313 on the annular fixing sleeve 312. At the same time, the annular fixing sleeve 312 is made of ABS engineering plastic injection molding, and the size of the annular groove on the outer surface is precisely matched with the electromagnetic induction coil 313. The snap-fit ​​fixation prevents the electromagnetic induction coil 313 from shifting due to vibration, thus ensuring the stability of power generation.

[0029] according to Figure 8 As shown, the inner surface of the protective housing 303 is bolted with a fixing plate 314 and a plastic support 317, and the fixing plate 314 is placed above the plastic support 317. The inner wall of the fixing plate 314 is fixedly connected to one end of the outer wall of the piezoelectric cantilever beam 315, and the bottom of the piezoelectric cantilever beam 315 is connected to the top of the counterweight magnet 316. The top of the plastic support 317 is connected to the bottom of the fixing magnet 318, and the counterweight magnet 316 and the fixing magnet 318 are arranged symmetrically.

[0030] In this embodiment of the invention, firstly, the fixing clamp 314 is made of 304 stainless steel, and one end of the piezoelectric cantilever beam 315 is fastened with hexagonal bolts, which ensures the fixing strength and avoids the rigid connection from restricting the beam deformation; the plastic support 317 is made of ABS engineering plastic, processed by injection molding, which reduces the overall weight while providing good insulation, preventing the magnet from forming an electrical circuit with the metal parts; secondly, the piezoelectric cantilever beam 315 uses PZT-5H dual crystal wafers, and both the counterweight magnet 316 and the fixing magnet 318 are made of N52 neodymium iron boron. The magnetic repulsion generated by the like poles facing each other controls the pre-bending amount of the piezoelectric cantilever beam 315 to be controlled within 0.5-1mm, as in CN112254908B "Self-Powered Monitoring Method of Piezoelectric Sensor for Simulating Embankment Slope Collapse" (Wenzhou University, authorized in 2022): This patent is applied to slope monitoring scenarios, and its piezoelectric ruler (length 38) The pre-bending amount (mm) is set to 0.5-1mm, and self-powered by bending deformation, further verifying the practicality and rationality of this pre-bending amount in the field of slope monitoring. This pre-bending state allows the resonant frequency of the piezoelectric cantilever beam 315 to be precisely matched with the micro-vibration frequency (10-100Hz) of the mountain. According to the "Technical Specification for Geological Disaster Monitoring" (DZ / T0221-2021) and published literature (such as "Analysis of Early Micro-vibration Signal Characteristics of Landslides", Chinese Journal of Geotechnical Engineering, Vol. 41, 2019), the core range of micro-vibration frequency generated by micro-fractures and joint slippage of soil and rock in the early stage of landslides is 10-100Hz. Below 10Hz, it is mostly environmental vibration (such as wind vibration and vehicle disturbance), and above 100Hz, the signal is easily absorbed and attenuated by soil and rock. Thus, small vibrations can induce significant deformation, thereby ensuring that the piezoelectric effect can be fully converted into electrical energy to make up for the energy supply gap in windless environments.

[0031] according to Figure 10 As shown, the bottom of the inner wall of the protective shell 303 is connected to a main circuit board 319, two supercapacitors 322, and a thin-film battery compartment 323. The top of the main circuit board 319 integrates a MEMS inclinometer 320 and a multi-functional accelerometer 321. The MEMS inclinometer 320 is used to capture the tilt angle of the top of the anchor rod 100 in real time, and the multi-functional accelerometer 321 is used to continuously monitor environmental micro-vibrations. The main circuit board 319 also integrates a communication module, a positioning module, an energy management circuit, and a microcontroller. The two supercapacitors 322 are used to store electrical energy. The thin-film battery compartment 323 contains a thin-film battery, which is used for emergency power supply in extreme windless and vibration-free environments.

[0032] In this embodiment of the invention, the main circuit board 319 firstly uses FR-4 fiberglass board, and the spacing between the integrated MEMS inclinometer 320 and the multiplexed accelerometer 321 is controlled within 5mm to reduce signal crosstalk. The data sampling rate of both is synchronized at 100Hz to ensure consistent monitoring data timestamps. The communication module uses a low-power SX1278LoRa module. Secondly, the positioning module uses a GPS+BeiDou dual-mode positioning chip (UBLOXNEO-7M) to ensure that the early warning information can be accurately associated with the monitoring location. At the same time, the energy management circuit integrates a TIBQ25504MPPT chip, which can dynamically adjust the energy storage strategy according to the output changes of wind energy and vibration energy. Two 350F / 2.7V supercapacitors 322 are set in parallel to effectively support the continuous operation of the system in low-power mode. The thin-film battery compartment 323 encapsulates a 10mAh / 3.7V lithium polymer thin-film battery as an emergency power source to ensure that the core sensing and communication functions are not interrupted in extreme windless and vibration-free environments.

[0033] according to Figure 11 as well as Figure 18 As shown, the visual verification active protection component 400 also includes a visual cabin 401, four ball bearings 414, and a transmission gear 419. The bottom of the visual cabin 401 is connected to the top of the top cover 304. A sealed cabin cover 402 is fixedly connected to the top of the visual cabin 401. Monitoring windows 405 are provided on all four sides of the outer wall of the visual cabin 401. Slide grooves 403 are symmetrically provided on all four sides of the outer wall of the visual cabin 401. The inner surface walls of the four slide grooves 403 are fixedly connected to the outer surface walls of two miniature electromagnets 404.

[0034] In this embodiment of the invention, firstly, the visual chamber 401 is precision machined from 6061 aluminum alloy and connected to the top cover 304 by M8 bolts and fitted with a fluororubber sealing ring, which can prevent rainwater and dust from entering the interior and affecting gear transmission and camera operation. The sealed chamber cover 402 is made of PC transparent material and is fixed by buckles, which facilitates observation of the internal components and provides dust protection. Secondly, the monitoring windows 405 on the four sides of the visual chamber 401 are made of Φ25mm×1mm sapphire glass, which is precision polished on both sides, making it wear-resistant and impact-resistant, and can maintain image clarity for a long time. Furthermore, the inner wall of the slide groove 403 is precision ground to provide low-resistance guidance for the sliding of the protective cover 406. At the same time, the miniature electromagnets 404 in the four slide grooves 403 are symmetrically arranged, with two electromagnets driving one protective cover 406 in each group, ensuring uniform driving force and effectively preventing jamming when the protective cover 406 slides.

[0035] according to Figures 19-20As shown, two rollers 409 are rolledly connected between the inner surfaces of each slide groove 403. A solid column 408 is inserted between the inner surfaces of each roller 409. One end of the outer wall of every four solid columns 408 is connected to one side of the outer wall of a corresponding protective cover 406. One side of the outer wall of each protective cover 406 is embedded and connected to a corresponding driving magnet 407. A fixing block 410 is fitted on the outer surface of each of the sixteen solid columns 408. A set of miniature springs 411 is elastically connected to one side of the outer wall of each fixing block 410. Each set of miniature springs 411 is elastically connected to one end of the inner wall of a corresponding slide groove 403. Every two sets of miniature springs 411 are used to provide a closing and restoring force for a corresponding protective cover 406.

[0036] In this embodiment of the invention, the roller 409 is made of polyurethane and has a miniature bearing embedded inside, which can convert sliding friction into rolling friction, effectively reducing the coefficient of friction and thus significantly reducing the sliding resistance of the protective cover 406. The solid column 408 is made of 304 stainless steel round rod and is fastened to the protective cover 406 with an M3 thread to ensure a firm connection that is not easy to loosen. The fixing block 410 is made of ABS engineering plastic injection molding and is fixed to the solid column 408 with an interference fit. Meanwhile, the miniature spring 411 is made of piano wire and is arranged symmetrically in pairs to provide closing and restoring force for each protective cover 406. This allows for rapid restoring without causing wear on the edge of the protective cover 406 due to excessive restoring force. The precise control of the restoring force ensures that the protective cover 406 accurately covers the monitoring window 405 after each closure, thereby effectively achieving reliable sealing protection and preventing dust and rainwater from entering and contaminating the lens.

[0037] according to Figure 17 as well as Figure 19 As shown, the outer walls of the visual cabin 401 are symmetrically connected with connecting plates 412 on all four sides. Each connecting plate 412 and the opposite side of the protective cover 406 are connected with a scraper 413, and each scraper 413 is used to scrape and clean the outer surface of the corresponding protective cover 406. The outer surfaces of the four ball bearings 414 are fixedly connected to the inner surfaces of the four mounting holes 306. The inner surface of each ball bearing 414 is rotatably connected to a rotating rod 415. The outer surface of each rotating rod 415 is fitted with a driven gear 416, and the four driven gears 416 are meshed with the outer wall of the transmission gear 419. The inner surface of the transmission gear 419 is rotatably connected to the outer surface of the rotating shaft 307. The top of each rotating rod 415 is fitted with a support frame 417. A miniature camera 418 is bolted to one end of the outer wall of each support frame 417, and each miniature camera 418 is directly facing the central axis of a corresponding monitoring window 405.

[0038] In this embodiment of the invention, the connecting plate 412 is made of 304 stainless steel and is fixed to the outer wall of the visual cabin 401 by laser welding. The scraper 413 is made of transparent silicone rubber with a Shore hardness of 50° and has a wedge-shaped cross section. It can effectively scrape off dust and water droplets from the surface during each opening and closing of the protective cover 406, preventing contaminants from affecting the imaging quality. The ball bearing 414 is made of 625 deep groove ball bearing, while the transmission gear 419 and the driven gear 416 are both made of POM + 30% glass fiber reinforced material, which can ensure that the swing of the flexible tail fin 309 can be efficiently converted into the rotational power of the miniature camera 418. At the same time, the miniature camera 418 is fixed by the support frame 417, and the center of the lens is effectively aligned with the central axis of the monitoring window 405, which can avoid imaging distortion. Furthermore, the four cameras rotate coaxially around the anchor rod 100 to achieve 360° panoramic coverage without blind spots.

[0039] In operation, after the energy harvesting intelligent sensing component 300 is activated, the flexible tail fin 309 triggers vortex-induced vibration under the influence of mountain winds. The blunt body 310 stabilizes the Karman vortex street to enhance the regularity and continuity of the oscillation, thereby causing the powerful magnets 311 on both sides to pass over the electromagnetic induction coils 313 outside the annular fixed sleeve 312, cutting magnetic field lines to generate main electrical energy. Simultaneously, micro-vibrations in the mountain cause deformation of the piezoelectric cantilever beam 315. The repulsive force between the counterweight magnet 316 and the fixed magnet 318 keeps the beam in a pre-bending sensitive state of 0.5-1mm, amplifying the deformation amplitude and converting it into auxiliary electrical energy through the piezoelectric effect. This electrical energy is optimized by the energy management circuit on the main circuit board 319 and stored in two parallel 350F / 2.7V supercapacitors 32. In section 2, under extreme windless and vibration-free environments, the thin-film battery in the thin-film battery compartment 323 provides emergency power, thereby ensuring the system's energy autonomy. Simultaneously, the MEMS inclinometer 320 and the multiplexed accelerometer 321 on the main circuit board 319 synchronously sample at 100Hz, capturing the tilt angle of the anchor bolt 100 and environmental micro-vibration signals in real time. When a tilt of ≥0.1° or a micro-vibration signal of ≥0.01g is detected, the microcontroller triggers system wake-up. At this time, the energy management circuit switches to the working mode of powering the visual verification active protection component 400. The microcontroller then sends pulsed currents to eight miniature electromagnets 404, generating a repulsive force with the drive magnets 407 on the protective cover 406, and pushing the protective cover 406 along the slide groove 40. 3. The device moves in a straight line, quickly opening the monitoring windows 405 on all four sides of the visual cabin 401. Simultaneously, the micro spring 411 is stretched and stores energy. At the same time, the swing of the flexible tail fin 309 drives the transmission gear 419 to rotate via the rotating shaft 307. Through meshing transmission, it drives the four passive gears 416 and the rotating rod 415 to rotate synchronously. The micro camera 418 on the support frame 417 rotates coaxially around the anchor rod 100, thereby achieving 360° panoramic shooting and wind speed adaptive sampling. During the shooting process, the scraper 413 on the connecting plate 412 cleans its outer surface as the protective cover 406 moves, effectively ensuring image clarity. These image data are processed in real time by the image stitching control board and then compressed and stored. When the shooting is completed, the micro electromagnet 404 is de-energized, and the micro spring 411 is released. The release force pulls the protective cover 406 to reset and close, thereby completing secondary cleaning and sealing protection. The main circuit board 319 integrates the macroscopic deformation data of MEMS inclinometer 320, the microscopic micro-vibration data of multiplexed accelerometer 321, and the panoramic visual data of miniature camera 418. Through cross-validation with preset algorithms, the validity of physical sensor data is verified in combination with the expansion of surface cracks. Environmental interference of visual data is eliminated by physical signals, eliminating false alarms caused by data silos. After the landslide risk level is determined by data fusion, the communication module can transmit the early warning information to the remote monitoring center. The positioning module simultaneously sends the deployment location information. At this time, the monitoring center outputs the corresponding early warning command, and the system returns to a low-power sleep state to wait for the next trigger.

[0040] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A landslide early warning device based on image recognition and computer vision, characterized in that: It includes an anchor bolt (100), an energy harvesting intelligent sensing component (300), and a visual verification active protection component (400). The top of the anchor bolt (100) is respectively equipped with the energy harvesting intelligent sensing component (300) and the visual verification active protection component (400). The energy harvesting intelligent sensing component (300) includes a flexible tail fin (309), a blunt body (310), two powerful magnets (311), two electromagnetic induction coils (313), a piezoelectric cantilever beam (315), a counterweight magnet (316), and a fixed magnet (318). The flexible tail fin (309) is used to trigger vortex-induced vibration by utilizing mountain wind flow. The blunt body (310) is used to stabilize the generation of the Karman vortex street and enhance the regularity and continuity of the oscillation of the flexible tail fin (309). The two powerful magnets (311) and the two electromagnetic induction coils (313) are used to generate induced current to provide main power to the system. The piezoelectric cantilever beam (315) is used to sense the micro-vibration of the mountain to induce mechanical deformation and to convert mechanical energy into electrical energy using the piezoelectric effect. The counterweight magnet (316) and the fixed magnet (318) are used to ensure that the piezoelectric cantilever beam (315) is in a pre-bent state, and the counterweight magnet (316) is used to amplify the deformation amplitude of the piezoelectric cantilever beam (315) when it vibrates. The visual verification active protection component (400) includes eight miniature electromagnets (404), eight protective covers (406), and eight driving magnets (407). The eight miniature electromagnets (404) and the eight driving magnets (407) are energized to generate a repulsive force to push the eight protective covers (406) to move in a straight line.

2. The landslide early warning device based on image recognition and computer vision according to claim 1, characterized in that: The top of the anchor rod (100) is fixedly welded with a first flange (200), and the anchor rod (100) is inserted below the landslide surface.

3. The landslide early warning device based on image recognition and computer vision according to claim 2, characterized in that: The energy harvesting intelligent sensing component (300) also includes a second flange (301), the bottom of which is bolted to the top of the first flange (200). The top of the second flange (301) is connected to an anti-slip washer (302), which is used for sealing and moisture protection and buffering installation stress. The top of the anti-slip washer (302) is connected to a protective shell (303), and the top of the protective shell (303) is fixedly connected to a top cover (304). The top of the top cover (304) is bolted to a bearing seat (305), and the bearing seat (305) has four mounting holes (306) on its top circumference.

4. The landslide early warning device based on image recognition and computer vision according to claim 3, characterized in that: A rotating shaft (307) is inserted between the inner surfaces of the bearing housing (305). A swing shaft (308) is fixedly connected to the top of the rotating shaft (307). One end of the outer wall of the swing shaft (308) is fixedly connected to one end of the outer wall of the flexible tail fin (309). The flexible tail fin (309) adopts a dovetail structure and the bifurcated wing surface is completely symmetrically stressed. The flexible tail fin (309) and the blunt body (310) are fixedly connected to each other on opposite sides. The outer sides of the flexible tail fin (309) are embedded and connected to two strong magnets (311). The bottom of the top cover (304) is symmetrically connected to annular fixing sleeves (312). The annular groove on the outer surface of each annular fixing sleeve (312) is sleeved and connected to the inner surface of a corresponding electromagnetic induction coil (313).

5. The landslide early warning device based on image recognition and computer vision according to claim 4, characterized in that: The inner surface of the protective shell (303) is bolted with a fixing clamp (314) and a plastic support (317), and the fixing clamp (314) is placed above the plastic support (317). The inner wall of the fixing clamp (314) is fixedly connected to one end of the outer wall of the piezoelectric cantilever beam (315), and the bottom of the piezoelectric cantilever beam (315) is connected to the top of the counterweight magnet (316). The top of the plastic support (317) is connected to the bottom of the fixing magnet (318), and the counterweight magnet (316) and the fixing magnet (318) are arranged symmetrically.

6. The landslide early warning device based on image recognition and computer vision according to claim 5, characterized in that: The bottom of the inner wall of the protective shell (303) is connected to a main circuit board (319), two supercapacitors (322), and a thin-film battery compartment (323). The top of the main circuit board (319) is integrated with a MEMS inclinometer (320) and a multi-functional accelerometer (321). The MEMS inclinometer (320) is used to capture the tilt angle of the top of the anchor rod (100) in real time. The multi-functional accelerometer (321) is used to continuously monitor environmental micro-vibrations. The main circuit board (319) also integrates a communication module, a positioning module, an energy management circuit, and a microcontroller. The two supercapacitors (322) are used to store electrical energy. The thin-film battery compartment (323) is encapsulated with a thin-film battery, which is used for emergency power supply in extreme windless and vibration-free environments.

7. The landslide early warning device based on image recognition and computer vision according to claim 3, characterized in that: The visual verification active protection component (400) also includes a visual chamber (401), four ball bearings (414), and a transmission gear (419). The bottom of the visual chamber (401) is connected to the top of the top cover (304). A sealed cover (402) is fixedly connected to the top of the visual chamber (401). Monitoring windows (405) are provided on all four sides of the outer wall of the visual chamber (401). Slide grooves (403) are symmetrically provided on all four sides of the outer wall of the visual chamber (401). The inner surface of each of the four slide grooves (403) is fixedly connected to the outer surface of two miniature electromagnets (404).

8. The landslide early warning device based on image recognition and computer vision according to claim 7, characterized in that: Two rollers (409) are rolled between the inner walls of each groove (403). A solid column (408) is inserted between the inner surfaces of each roller (409). One end of the outer wall of every four solid columns (408) is connected to one side of the outer wall of a corresponding protective cover (406). One side of the outer wall of each protective cover (406) is embedded with a corresponding driving magnet (407). A fixing block (410) is fitted on the outer surface of each of the sixteen solid columns (408). A set of miniature springs (411) is elastically connected to one side of the outer wall of each fixing block (410). Each set of miniature springs (411) is elastically connected to one end of the inner wall of a corresponding groove (403). Every two sets of miniature springs (411) are used to provide a closing and restoring force for a corresponding protective cover (406).

9. The landslide early warning device based on image recognition and computer vision according to claim 8, characterized in that: The outer walls of the visual chamber (401) are symmetrically connected with connecting plates (412) on all four sides. Each connecting plate (412) and the protective cover (406) are connected to a scraper (413) on opposite sides. Each scraper (413) is used to scrape and clean the outer surface of a corresponding protective cover (406). The outer surfaces of the four ball bearings (414) are fixedly connected to the inner surfaces of the four mounting holes (306). The inner surface of each ball bearing (414) is rotatably connected to a rotating rod (415). Each rotating rod... The outer surface of each of the four driven gears (416) is fitted with a passive gear (416), and the four passive gears (416) are meshed with the outer wall of the transmission gear (419). The inner surface of the transmission gear (419) is rotatably connected to the outer surface of the rotating shaft (307). The top of each of the rotating rods (415) is fitted with a support frame (417). A miniature camera (418) is bolted to one end of the outer wall of each support frame (417), and each miniature camera (418) is directly opposite the central axis of a corresponding monitoring window (405).

10. A landslide early warning method based on image recognition and computer vision, using the landslide early warning device based on image recognition and computer vision according to any one of claims 1-9, comprising the following steps: S1: Insert and fix the anchor rod (100) below the landslide surface, and complete the integrated installation of the energy harvesting intelligent sensing component (300) and the visual verification active protection component (400) by bolting the first flange (200) and the second flange (301). The anti-slip washer (302) achieves sealing and moisture protection and installation stress buffering. After the device is started, the hybrid energy harvesting module enters a continuous working state. The mountain wind triggers the vortex-induced vibration of the flexible tail fin (309). The blunt body (310) stabilizes the Karman vortex street and strengthens the regularity of the oscillation. This causes the powerful magnets (311) on both sides of the flexible tail fin (309) to pass over the electromagnetic induction coil (313) outside the annular fixed sleeve (312), cutting the magnetic field lines to generate induced current (main energy). At the same time, the micro-vibration of the mountain causes the piezoelectric cantilever beam (315) to deform. The repulsive force between the counterweight magnet (316) and the fixed magnet (318) puts the beam in a pre-bending sensitive state, amplifying the deformation amplitude and converting it into auxiliary electrical energy through the piezoelectric effect. After being optimized by the energy management circuit on the main circuit board (319), the electrical energy is stored in two supercapacitors (322). In extreme windless and vibration-free environments, the thin film battery in the thin film battery compartment (323) provides emergency power to ensure the system’s continuous energy autonomy. S2: The MEMS inclinometer (320) on the main circuit board (319) captures the tilt angle of the top of the anchor rod (100) in real time, and the multi-purpose accelerometer (321) continuously monitors the environmental micro-vibration signal. Both are in low power operation mode. When the MEMS inclinometer (320) detects a tilt angle of the anchor (100) and the multiplex accelerometer (321) captures a micro-vibration signal, the microcontroller triggers a system wake-up command, and the energy management circuit switches to the working mode to power the visual verification active protection component (400). S3: The microcontroller sends pulse current to eight micro electromagnets (404). The eight micro electromagnets (404) and the driving magnets (407) on the protective cover (406) generate a repulsive force with the same pole, pushing the protective cover (406) to move in a straight line along the slide (403), quickly opening the monitoring windows (405) on all four sides of the vision cabin (401). At the same time, the micro spring (411) is stretched to store energy. The swing of the flexible tail fin (309) drives the transmission gear (419) to rotate through the rotating shaft (307), and through meshing transmission drives the four passive gears (416) and the rotating rod (415) to rotate synchronously. The miniature camera (418) on the support frame (417) rotates coaxially around the anchor rod (100) to take 360° panoramic pictures of the ground outside the monitoring window (405) and realize wind speed adaptive sampling. During the shooting process, the scraper (413) on the connecting plate (412) moves with the protective cover (406) to clean the outer surface of the protective cover (406) to ensure image clarity. The image data is processed in real time by the image stitching control board, stitched into a complete panoramic image and compressed for storage. After the shooting is completed, the miniature electromagnet (404) is de-energized, and the miniature spring (411) releases its elastic force to pull the protective cover (406) back to its original position and close, completing the secondary cleaning and sealing protection; S4: The main circuit board (319) integrates multi-source data: macroscopic deformation data from the MEMS inclinometer (320), microscopic vibration data from the multiplexed accelerometer (321), and panoramic visual data from the miniature camera (418); The microcontroller performs cross-validation through a preset algorithm, and combines the expansion of surface cracks in the visual image to verify the validity of the physical sensor data. It eliminates environmental interference in the visual data through physical signals and eliminates false alarms caused by data silos. After determining the landslide risk level through data fusion, the communication module transmits the early warning information to the remote monitoring center, and the positioning module simultaneously sends the device deployment location information; The monitoring center outputs corresponding early warning commands based on the risk level to achieve early warning of landslides. At the same time, the system returns to a low-power sleep state, waiting for the next trigger.

Images