Road collapse early warning method based on image recognition

Abstract

The invention relates to the technical field of road management, in particular to a road collapse early warning method based on image recognition, which comprises the following steps: acquiring road coordinates, inputting the road coordinates into aerial photography equipment, generating an aerial photography path by the aerial photography equipment based on the road coordinates, and executing road image acquisition in a navigation process, the collected road images are synchronously stored in the aerial photography equipment; road image source position environment parameters are collected, road image source position collapse risk factors are evaluated based on the environment parameters, and road images stored in the aerial photography equipment are sorted according to the collapse risk factors; according to the method, the road image is acquired through the aerial photography equipment, the road image quality is enhanced through graph optimization under the condition that the condition that the road image is blurred due to the fact that vehicles run on the road is ensured in combination with image optimization, and then the processed road image is used for carrying out collapse positioning, prediction and early warning on the road surface and the roadside slope of the road.

Description

Technical Field

[0001] The present invention relates to the technical field of highway management, and particularly relates to a highway landslide early warning method based on image recognition. Background Art

[0002] Highway collapse is a serious road disaster phenomenon. Affected by factors such as geological conditions, rainstorm scouring, and vehicle overloading, the highway pavement or slope structure is damaged, resulting in situations such as collapse and landslide. This will not only block traffic and affect people's travel, but may also endanger the lives of passing vehicles and pedestrians, causing greater losses to society and the economy.

[0003] The invention patent with the application number 201910317156.5 discloses a comprehensive dynamic landslide safety early warning method for tunnel construction based on multi-source data, including the following steps: S1, collecting data and constructing a dynamic landslide engineering database for tunnel construction; through construction personnel combined with surveying instruments for overall design and overall investigation, collecting basic information before tunnel construction: during the construction process, conducting section and cross-section investigations and real-time monitoring, collecting dynamic information during tunnel construction; S2, risk identification; according to the dynamic parameters of the dynamic landslide engineering database for tunnel construction, using multi-source data for construction risk identification, the multi-source data includes geological risk early warning, advanced prediction early warning, precursor information early warning, and monitoring information early warning: the data of the geological area risk early warning, advanced prediction early warning, and precursor information early warning are managed through section and cross-section information: the monitoring information pre-data is managed and released through cross-section information: S3, unit data analysis; respectively performing unit data analysis on the information of geological risk early warning, advanced prediction early warning, precursor information early warning, and monitoring information early warning, and judging their risk levels, where: geological risk early warning: when the risk is not processed or ignored, take the maximum value, and after the risk is processed, it should be re-evaluated, and take the value after re-evaluation and always participate in the dynamic monitoring and evaluation.

[0004] This application aims to solve the problem: "At present, there is no complete comprehensive dynamic landslide safety early warning system that can consider multi-source data. The existing systems are single and overly rely on monitoring and measurement, resulting in a blurred boundary between disaster special monitoring early warning and monitoring and measurement early warning work. In actual projects, after a disaster is discovered through monitoring and measurement, generally, monitoring and measurement continue to be used for disaster pre-alarm, and the disaster is not monitored and warned from the objective existence of the disaster body."

[0005] However, in the case of highway landslides, the landslides mainly come from the pavement or the side slope of the road, which pose a safety threat to the vehicles driving on the highway and block traffic, seriously affecting the road transport capacity. At present, there is no relatively predictive and intelligent technology for monitoring and early warning of highway collapse problems.

[0006] Therefore, a highway landslide early warning method based on image recognition is proposed. Summary of the Invention

[0007] In view of the above drawbacks of the prior art, the present invention provides a highway landslide warning method based on image recognition, which solves the technical problems proposed in the above background art.

[0008] To achieve the above objectives, the present invention is realized through the following technical solutions:

[0009] A highway landslide warning method based on image recognition, comprising:

[0010] Obtain road coordinates, input the road coordinates into an aerial photography device, the aerial photography device generates an aerial photography path based on the road coordinates, and performs road image acquisition during the navigation process. The acquired road images are stored synchronously in the aerial photography device;

[0011] The content of the road images includes the road surface and the roadside mountain body. After the road images are acquired and before the storage operation is performed in the aerial photography device, image optimization processing is synchronously performed;

[0012] The image optimization processing logic is expressed as:

[0013] Logic1: Determine whether there is a vehicle in the road image. If the determination result is no, jump to Logic2. If the determination result is yes, then: I compensated (x,y) = I(x - v x (x,y), y - v y (x,y));

[0014] Logic2:

[0015] Logic3:

[0016] In the formula: I compensated (x,y) is the road image after motion compensation; v x (x,y), v y (x,y) represents the motion speed of the pixel point (x, y) in the x direction and the y direction; I deblurred (x,y) is the road image after deblurring; H * (u,v) is the Fourier transform of the blur kernel; H(u,v) is the conjugate complex number of H * (u,v); λ is the regularization parameter; F{I compensated (x,y) or I(x,y)} is the Fourier transform of the image I compensated (x,y) after motion compensation or the original image I(x,y); J(x,y) is the finally optimized road image; α is the parameter controlling the sharpening degree; For the deblurred image I deblurred (x,y) is the result of applying the Laplacian operator;

[0017] where the selection of F{I compensated (x,y) or I(x,y)} is based on the determination result in Logic1;

[0018] The expression formula of the blur kernel is:

[0019]

[0020] In the formula: h(x,y) is the blur kernel; L is the blur length; θ is the motion direction angle;

[0021] where the motion direction angle θ is determined based on the angle of the section where the vehicle is located on the road, the blur length L is customized with reference to the distance traveled by the vehicle during the exposure time, the unit of the distance traveled by the vehicle during the exposure time is pixels. When the determination result of Logic1 is no, in Logic2 transforms to I deblurred (x,y) = F{I compensated (x,y) or I(x,y)};

[0022] Collect the environmental parameters of the source location of the road image, evaluate the landslide risk factor of the source location of the road image based on the environmental parameters, and sort the road images stored in the aerial photography device according to the risk factor; retrieve the road images in sequence according to the sorting result of the road images. After each retrieval of a road image, analyze the road landslide risk with the retrieved road image; traverse the analysis results of the road landslide risks corresponding to each road image, and re - sort each road image based on the analysis results of the road landslide risks corresponding to each road image; mark the source location for each road image in the re - sorted road images, and feedback the re - sorted road images with source location marks to the user terminal.

[0023] Furthermore, the number of road coordinates input to the aerial photography device is customized by the user terminal. When performing the input operation, each continuously input road coordinate is adjacent and has an equal spacing on the road. The number of road coordinates input to the aerial photography device follows: the higher the accuracy requirement for highway landslide warning, the more road coordinates are input to the aerial photography device, and vice versa, the fewer road coordinates are input to the aerial photography device. The aerial photography path is generated by connecting adjacent road coordinates input to the aerial photography device;

[0024] where, after the aerial photography path is generated, the aerial photography device synchronously operates based on the aerial photography path and collects road images during the operation. The collection position of the road images is each road coordinate on the aerial photography path.

[0025] Further, the environmental parameters of the road image source location include: the number of contents included in the road image, rainfall, wind speed, vibration transmission information of historical vehicles traveling on the road, and soil looseness coefficient;

[0026] The calculation formula for the landslide risk factor is:

[0027]

[0028] In the formula: σ is the landslide risk factor of the road image source location; q is a constant; F is the rainfall; V is the wind speed; κ is the soil looseness coefficient; P MIN 、P MAX are the minimum and maximum amplitudes generated by historical vehicles traveling on the road;

[0029] Among them, the constant q is rounded to 1 or 1.1. When the number of contents included in the road image is 1, q takes 1; when the number of contents included in the road image is 2, q takes 1.1.

[0030] Further, the larger the landslide risk factor σ of the road image source location, the higher the landslide risk of the road source location; conversely, it indicates that the landslide risk of the road source location is lower;

[0031] The calculation target of the soil looseness coefficient κ is the soil of the road base and the soil of the roadside mountain. When the constant q takes 1.1, the soil looseness coefficient κ takes the average value of the soil looseness coefficients of the road base soil and the roadside mountain soil. The calculation formula for the soil looseness coefficient κ is:

[0032] κ = ρ 1 / ρ 2 ;

[0033] In the formula: ρ 1 is the density of the soil in its natural state; ρ 2 is the density of the soil in its loose state;

[0034] Among them, when sorting the road images, they are sorted based on the landslide risk factor of the road image source location, so that the road images with larger corresponding risk factors are arranged in the front.

[0035] Further, when the road image content includes both the road surface and the roadside mountain, the road landslide risk is analyzed separately based on the road surface and the roadside mountain, and the final analysis result takes the maximum value of the two analysis results.

[0036] Further, the road landslide risk analysis includes road surface landslide risk analysis and roadside mountain landslide risk analysis. The analysis logic is expressed as:

[0037]

[0038] Where: R is the risk value of road collapse; K v is the proportion of vegetation pixels to slope pixels; K s is the proportion of slope area pixels; K c is the proportion of crack pixels to road pixels; G c is the average gray value of cracks; G r is the average gray value of the road; ω 1 、ω 2 、ω 3 are weights;

[0039] The larger the risk value R of the road collapse, the higher the probability of road collapse; conversely, the lower the probability of road collapse. The weights ω 1 、ω 2 、ω 3 are all positive numbers and their sum is 1.

[0040] Furthermore, the K v 、K s 、K c 、G c 、G r are calculated by the following formula:

[0041]

[0042] Where: n v is the number of vegetation pixels in the slope area; n s is the number of pixels in the slope area; N is the total number of pixels in the road image; n c is the number of crack pixels; n r is the number of pixels in the road area; g i is the gray value of the i-th crack pixel; g′ i is the gray value of the i-th road area pixel.

[0043] Furthermore, when reordering the road images, the road images with high road collapse risk are arranged in the front, and the road images with low road collapse risk are arranged in the back;

[0044] When feeding back the road images with source location marks and reordering to the client, any mobile computer device held by the client is used as the feedback target. The client reads the road images with source location marks and reordering sequentially on the mobile computer, further sets a collapse warning determination interval, and compares the collapse warning determination interval with the road collapse risk analysis results corresponding to each read road image. The source location of the road image that meets the collapse warning determination interval is used as the collapse warning target.

[0045] Adopting the technical solution provided by the present invention, compared with the known public technology, it has the following beneficial effects:

[0046] The present invention provides a highway landslide early warning method based on image recognition. During the execution of this method, road images are collected by an aerial photography device. Considering that there are driving vehicles on the road, which may cause the road images to be blurred, image optimization is combined to enhance the quality of the road images. Then, the processed road images are used to locate and predict landslides on the road surface and roadside slopes of the highway, providing more effective and reliable data references for highway management users, so as to carry out more targeted maintenance services for the highway, ensure the health and safety of the highway, and ensure the stable transportation capacity of the highway. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0048] Figure 1 It is a schematic flow chart of a highway landslide early warning method based on image recognition. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0049] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

[0050] The following further describes the present invention with reference to the embodiments.

[0051] Embodiment:

[0052] A highway landslide early warning method based on image recognition in this embodiment, as Figure 1 shown, includes:

[0053] Obtain road coordinates, input the road coordinates into an aerial photography device, and the aerial photography device generates an aerial photography path based on the road coordinates. During the navigation process, road image collection is performed, and the collected road images are stored synchronously in the aerial photography device;

[0054] The number of road coordinates input to the aerial photography device is user-defined. When performing the input operation, the continuously input road coordinates are adjacent on the road and have equal spacing. The number of road coordinates input to the aerial photography device follows the rule that the higher the accuracy requirement for highway landslide warning, the more road coordinates are input to the aerial photography device; conversely, the fewer road coordinates are input to the aerial photography device. The aerial photography path is connected adjacent to each other based on the road coordinates input to the aerial photography device to complete the generation operation;

[0055] Among them, after the aerial photography path is generated, the aerial photography device runs synchronously based on the aerial photography path, and collects road images during the running process. The collection position of the road images is each road coordinate on the aerial photography path;

[0056] The content of the road images includes the road surface and the roadside mountain. After the road images are collected, before performing the storage operation in the aerial photography device, image optimization processing is performed synchronously;

[0057] The image optimization processing logic is expressed as:

[0058] Logic1: Determine whether there is a vehicle in the road image. If the determination result is no, jump to Logic2. If the determination result is yes, then: I compensated (x,y) = I(x - v x (x,y), y - v y (x,y));

[0059] Logic2:

[0060] Logic3:

[0061] In the formula: I compensated (x,y) is the road image after motion compensation; v x (x,y), v y (x,y) represents the motion speed of the pixel point (x, y) in the x direction and the y direction; I deblurred (x,y) is the road image after deblurring; H * (u,v) is the Fourier transform of the blur kernel; H(u,v) is the conjugate complex number of H * (u,v); λ is the regularization parameter; F{I compensated (x,y) or I(x,y)} is the Fourier transform of the image I compensated (x,y) after motion compensation or the original image I(x,y); J(x,y) is the finally optimized road image; α is the parameter controlling the sharpening degree; is the result of applying the Laplace operator to the deblurred image I deblurred (x,y);

[0062] Among them, the selection of F{I compensated (x,y) or I(x,y)} is based on the determination result in Logic1;

[0063] The expression formula of the fuzzy kernel is:

[0064]

[0065] In the formula: h(x,y) is the fuzzy kernel; L is the fuzzy length; θ is the motion direction angle;

[0066] Among them, the motion direction angle θ is determined based on the angle of the section where the vehicle is located on the road, and the fuzzy length L is customized with reference to the distance traveled by the vehicle during the exposure time. The unit of the distance traveled by the vehicle during the exposure time is pixels. When the determination result of Logic1 is no, in Logic2 It is transformed into I deblurred (x,y) = F{I compensated (x,y) or I(x,y)};

[0067] Through the above logical formula calculation, the road image is carefully optimized.

[0068] Collect the environmental parameters of the source location of the road image, evaluate the landslide risk factor of the source location of the road image based on the environmental parameters, and sort the road images stored in the drone according to the risk factor;

[0069] The environmental parameters of the source location of the road image include: the number of contents included in the road image, rainfall, wind speed, vibration transfer information of historical vehicles traveling on the road, and soil looseness coefficient;

[0070]

[0071] In the formula: σ is the landslide risk factor of the source location of the road image; q is a constant; F is the rainfall; V is the wind speed; κ is the soil looseness coefficient; P MIN 、P MAX are the minimum and maximum amplitudes generated by historical vehicles traveling on the road;

[0072] Through the above logical formula, the landslide risk factor of the source location of the road image is further calculated, providing logical support for the further execution of the method in this embodiment;

[0073] It should be noted that the risk factor here represents the degree of danger of the corresponding section of the road image in the event of a collapse accident.

[0074] Among them, the constant q is rounded to 1 or 1.1. When the number of contents included in the road image is 1, q is taken as 1, and when the number of contents included in the road image is 2, q is taken as 1.1;

[0075] The greater the landslide risk factor σ of the road image source location, the higher the landslide risk of the road source location; conversely, it indicates that the landslide risk of the road source location is lower.

[0076] The calculation target of the soil looseness coefficient κ is the soil of the road base and the soil of the roadside mountain. When the constant q is taken as 1.1, the soil looseness coefficient κ takes the average value of the soil looseness coefficients of the road base soil and the roadside mountain soil. The calculation formula of the soil looseness coefficient κ is:

[0077] κ = ρ 1 / ρ 2 ;

[0078] In the formula: ρ 1 is the density of the soil in the natural state; ρ 2 is the density of the soil in the loose state;

[0079] Among them, when sorting the road images, they are sorted based on the landslide risk factor of the road image source location, so that the road images with larger corresponding risk factors are arranged in the front;

[0080] The road images are retrieved in sequence according to the sorting result of the road images. After each road image is retrieved, the road collapse risk is analyzed based on the retrieved road image;

[0081] The road landslide risk analysis includes the road surface landslide risk analysis and the roadside mountain landslide risk analysis. The analysis logic is expressed as:

[0082]

[0083] In the formula: R is the road landslide risk value; K v is the proportion of vegetation pixels to slope pixels; K s is the proportion of slope area pixels; K c is the proportion of crack pixels to road pixels; G c is the average gray value of cracks; G r is the average gray value of the road; ω 1 , ω 2 , ω 3 are weights;

[0084] The greater the road landslide risk value R, the higher the road landslide probability; conversely, it indicates that the road landslide probability is lower. The weights ω 1 , ω 2 , ω 3 are all positive numbers and their sum is 1;

[0085] K v , K s , K c , G c , G r are calculated by the following formula:

[0086]

[0087] Where: n v is the number of vegetation pixels in the slope area; n s is the number of pixels in the slope area; N is the total number of pixels in the road image; n c is the number of crack pixels; n r is the number of pixels in the road area; g i is the gray value of the i-th crack pixel; g′ i is the gray value of the i-th road area pixel;

[0088] Through the above logical formula, the risk of road collapse is further quantified and determined finally for the positions on the road that really need early warning;

[0089] Traverse the analysis results of the road collapse risk corresponding to each road image, and re - sort each road image based on the analysis results of the road collapse risk corresponding to each road image;

[0090] Mark the source position of each road image in the re - sorted road images, and feedback the re - sorted road images with source position marks to the user terminal;

[0091] When re - sorting the road images, arrange the road images with high road collapse risk in the front and the road images with low road collapse risk in the back;

[0092] When feedbacking the road images with source position marks and re - sorting to the user terminal, use any mobile computer device held by the user terminal as the feedback target. The user terminal reads the road images with source position marks and re - sorting in sequence on the mobile computer, further sets the collapse early warning determination interval, compares the collapse early warning determination interval with the analysis results of the road collapse risk corresponding to each read road image, and takes the source position of the road image that meets the collapse early warning determination interval as the collapse early warning target.

[0093] In this embodiment, through the method in the above - mentioned embodiment, a comprehensive and intelligent collapse risk early warning is brought to the road surface and the roadside slope of the highway, which is more efficient, lower in cost than the existing methods, and effectively supports the daily operation and maintenance of the highway.

[0094] Such as Figure 1 shown, when the content of the road image includes both the road surface and the roadside mountain body at the same time, the road collapse risk is analyzed based on the road surface and the roadside mountain body respectively, and the final analysis result takes the maximum value of the two analysis results.

[0095] Through the above settings, it further provides the necessary execution logic support for the execution of the steps of the method in the above embodiments, ensuring a stable output of the positions on the road that require early warning in the execution of the method steps in the above embodiments.

[0096] In summary, in the execution process of the method in the above embodiments, road images are collected by an aerial photography device. Combining image optimization, in the case where the road images are blurred due to the presence of moving vehicles on the road, the quality of the road images can be enhanced through graphic optimization. Furthermore, the processed road images are used to predict and give early warnings about landslides on the road surface and roadside slopes of the highway, providing more effective and reliable data references for highway management end-users, so as to carry out more targeted maintenance services for the highway, ensure the health and safety of the highway, and ensure the stable transportation capacity of the highway.

[0097] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A highway landslide early warning method based on image recognition, characterized in that: include: Obtaining road coordinates, inputting the road coordinates into the aerial photography device, the aerial photography device generates an aerial photography path based on the road coordinates, performing road image acquisition during the navigation process, and the acquired road images are synchronously stored in the aerial photography device; Collecting environmental parameters of the source location of the road image, evaluating the landslide risk factor of the source location of the road image based on the environmental parameters, and sorting the road images stored in the aerial photography equipment according to the landslide risk factor; Retrieving road images in sequence according to the road image sorting result, and analyzing the road collapse risk with the retrieved road images after each road image is retrieved; Traversing the road collapse risk analysis results corresponding to each road image, and reordering each road image based on the road collapse risk analysis results corresponding to each road image; The source position of each road image in the reordered road images is marked, and the reordered road images with the source position mark are fed back to the user end.

2. A highway landslide early warning method based on image recognition according to claim 1, characterized in that: The number of road coordinates input to the aerial photography device is customized by the user. When performing the input operation, each continuously input road coordinate is adjacent to the road and has an equal spacing. The number of road coordinates input to the aerial photography device follows the following rules: the higher the requirement for the highway collapse warning accuracy, the more road coordinates are input to the aerial photography device. Conversely, the fewer road coordinates are input to the aerial photography device. The aerial photography path is connected to each other based on the adjacent road coordinates input to the aerial photography device to complete the generation operation. After the aerial photography path is generated, the aerial photography equipment runs synchronously based on the aerial photography path, and collects road images during the operation. The collection position of the road image is each road coordinate on the aerial photography path.

3. A highway landslide early warning method based on image recognition according to claim 1, characterized in that: The road image content includes the road surface and the mountain on the road side. After the road image is acquired and before the storage operation is performed in the aerial photography device, the image optimization processing is performed synchronously; The image optimization processing logic is expressed as: Logic1: Determine whether there is a vehicle in the road image. If the result is no, jump to Logic2. If the result is yes, then: I compensated (x,y)=I(xv x (x,y),yv y (x,y)); Logic2: Logic3: Where: I compensated (x, y) is the road image after motion compensation; v x (x,y),v y (x, y) represents the movement speed of the pixel point (x, y) in the x direction and the y direction; deblurred (x, y) is the road image after deblurring; H * (u,v) is the Fourier transform of the blur kernel; H(u,v) is H * The conjugate complex number of (u,v); λ is the regularization parameter; F{I compensated (x,y)orI(x,y)} is the motion compensated image I compensated (x, y) or the Fourier transform of the original image I(x, y); J(x, y) is the road image after the final optimization processing; α is the parameter that controls the degree of sharpening; is the deblurred image I deblurred (x,y) is the result of applying the Laplacian operator; Among them, F{I compensated The selection of (x,y)orI(x,y)} is based on the decision result in Logic1.

4. A highway landslide early warning method based on image recognition according to claim 3, characterized in that: The expression formula of the fuzzy kernel is: Where: h(x,y) is the blur kernel; L is the blur length; θ is the motion direction angle; The moving direction angle θ is determined based on the angle of the road section where the vehicle is located. The blur length L is customized based on the distance the vehicle moves during the exposure time. The distance the vehicle moves during the exposure time is in pixels. When the result of Logic1 is negative, Logic2 Transform to I deblurred (x,y)=F{I compensated (x,y)orI(x,y)}.

5. The highway collapse early warning method based on image recognition according to claim 1 is characterized in that: The environmental parameters of the road image source location include: the amount of content contained in the road image, rainfall, wind speed, vibration transmission information of historical vehicles traveling on the road, and soil looseness coefficient; The calculation formula of landslide risk factor is: Where: σ is the landslide risk factor at the source location of the road image; q is a constant; F is rainfall; V is wind speed; κ is soil looseness coefficient; P MIN , P MAX The minimum and maximum amplitudes generated by vehicles traveling on the road in history; The constant q is rounded to 1 or 1.

1. When the number of contents contained in the road image is 1, q is 1; when the number of contents contained in the road image is 2, q is 1.

1.

6. A highway landslide early warning method based on image recognition according to claim 5, characterized in that: The larger the landslide risk factor σ of the road image source position is, the higher the landslide risk of the road source position is; conversely, the smaller the landslide risk of the road source position is, the lower the landslide risk of the road source position is. The calculation targets of the soil loose coefficient κ are the road base soil and the roadside mountain soil. When the constant q is 1.1, the soil loose coefficient κ is the average of the loose coefficients of the road base soil and the roadside mountain soil. The calculation formula of the soil loose coefficient κ is: κ=ρ1 / ρ2; Where: ρ1 is the density of soil in its natural state; ρ2 is the density of soil in its loose state; When the road images are sorted, they are sorted based on the landslide risk factors of the road image source locations, so that road images with larger corresponding risk factors are arranged in front.

7. The highway landslide early warning method based on image recognition according to claim 1 is characterized in that: When the road image content includes both the road surface and the roadside mountain, the road collapse risk is analyzed based on the road surface and the roadside mountain respectively, and the final analysis result takes the maximum value of the two analysis results.

8. The highway landslide early warning method based on image recognition according to claim 1 is characterized in that: The road collapse risk analysis includes road collapse risk analysis and roadside mountain collapse risk analysis. The analysis logic is expressed as follows: Where: R is the road collapse risk value; K v K is the ratio of vegetation pixels to slope pixels; s is the pixel ratio of the slope area; K c is the ratio of crack pixels to road pixels; G c is the average gray value of the crack; G r is the average gray value of the road; ω1, ω2, ω3 are weights; The larger the road collapse risk value R is, the higher the probability of road collapse is. Conversely, the smaller the road collapse risk is, the lower the probability of road collapse is. The weights ω1, ω2, and ω3 are all positive numbers and their sum is 1.

9. A highway landslide early warning method based on image recognition according to claim 8, characterized in that: The K v , K s , K c , G c , G r The calculation is performed by the following formula: Where: n v is the number of vegetation pixels in the slope area; n s is the number of pixels in the slope area; N is the total number of pixels in the road image; n c is the number of crack pixels; n r is the number of pixels in the road area; g i is the gray value of the i-th crack pixel; g′ i is the gray value of the pixel in the i-th road area.

10. The highway landslide early warning method based on image recognition according to claim 1, characterized in that: When the road images are reordered, the road images with a high risk of road collapse are arranged in front, and the road images with a low risk of road collapse are arranged in the back; When feeding back the road images with source location marks and reordering to the user end, any mobile computer device held by the user end is used as the feedback target. The user end reads the road images with source location marks and reordering in sequence on the mobile computer, further sets the landslide warning judgment interval, compares the landslide risk analysis results of the road images read each time with the landslide warning judgment interval, and uses the source position of the road image that meets the landslide warning judgment interval as the landslide warning target.

Images