Pavement disease automatic detection method and system based on deep learning

Abstract

The invention relates to the technical field of pavement disease automatic detection, and particularly discloses a pavement disease automatic detection method and system based on deep learning, and the system comprises a municipal pavement data collection module, a pavement image quality analysis module, a pavement image re-collection module, a pavement disease evaluation feedback module and a database. Flight parameters (speed and height) are dynamically adjusted through an environmental quality index (EQI), the image acquisition quality is ensured, the detection precision in a complex environment is improved, a multi-disease coupling risk model is constructed, region weight and traffic volume dynamic weight are integrated, a disease risk level is output, and finally a disease development trend is predicted through a one-time exponential smoothing method. The system achieves the dual functions of current situation evaluation and trend early warning, predicts the disease evolution trend based on historical data, remarkably improves the scientificity and timeliness of municipal road risk management, reduces the maintenance cost, and prevents safety accidents.

Description

Technical Field

[0001] The present invention belongs to the technical field of automatic detection of road surface diseases, and relates to an automatic detection method and system for road surface diseases based on deep learning. Background Art

[0002] With the acceleration of the urbanization process, the maintenance demand for municipal roads is increasing day by day. At present, the detection of road surface diseases mainly relies on manual inspections or traditional image processing techniques. Manual inspections are inefficient, costly, and limited by the recognition accuracy of the human eye, making it difficult to detect minor diseases. Traditional image processing methods usually rely on fixed thresholds or simple feature extraction, and are easily interfered by environmental factors such as changes in lighting and weather conditions (such as rain and fog), resulting in high false detection and missed detection rates. Therefore, in order to achieve accurate disease risk assessment and trend prediction, it is necessary to develop an automatic detection method for road surface diseases based on deep learning.

[0003] For example, the patent with the Chinese patent publication number CN117671495A discloses a real-time automatic detection method and system for road surface diseases based on edge computing technology, including: receiving road surface original image data and an automatic recognition request, decoding and preprocessing the road surface original image data to generate road surface image data, performing deep neural network calculations on the road surface image data to obtain the detection type of the disease and the detection frame of the location of the disease, using the non-maximum suppression algorithm to retain the detection frame with the highest confidence, optimizing the data of the detection frame with the highest confidence based on the connectivity principle to determine the optimal detection frame position, generating response data from the disease type and location recognition results, and returning it to the road condition acquisition device, improving the edge-side data processing efficiency and realizing the real-time recognition and processing of road surface disease image data.

[0004] The following problems also exist in the above existing technologies: 1. Only relying on fixed road condition acquisition devices to receive original images, without considering the impact of environmental changes on image quality, unable to ensure image quality in harsh environments, resulting in a decrease in detection accuracy.

[0005] 2. Only relying on the output of the deep neural network detection frame, lacking quantitative analysis of the geometric parameters of the disease (such as crack width, depression depth), and at the same time, the detection results are not linked to traffic load and regional functions, with limited practicality.

[0006] 3. Only returning the disease type and location, without providing risk level assessment and prediction of the development trend of road surface diseases, not predicting the evolution law of diseases based on historical data, unable to timely feedback the development trend of diseases, and unable to support the implementation of priority maintenance decisions for municipal road surfaces. Summary of the Invention

[0007] In view of this, to solve the problems raised in the above background art, an automatic detection method and system for road surface diseases based on deep learning are proposed.

[0008] The object of the present invention can be achieved by the following technical solutions: In the first aspect of the present invention, an automatic road surface disease detection method based on deep learning is provided, including: S1. The unmanned aerial vehicle (UAV) collects natural environment data and road surface image data of the municipal road surface at position time series by carrying sensors and cameras.

[0009] S2. Coupling the natural environment data with historical illumination data to analyze the environmental quality index of the municipal road surface at the current moment, and triggering a flight parameter adjustment mechanism when the environmental quality index is within the set environmental quality index range.

[0010] S3. Adjusting the image re-acquisition flight parameters of the UAV based on the flight parameter adjustment mechanism, and using the re-acquired images with qualified image quality indicators as the final road surface images.

[0011] S4. Identifying the types of road surface diseases and the set of characteristic parameters corresponding to various road surface diseases from the collected road surface images, constructing a disease risk assessment index for the municipal road surface and an assessment index for the development trend of various road surface diseases, outputting the disease risk level and the influence trend level of various road surface diseases and corresponding feedback.

[0012] In the second aspect of the present invention, an automatic road surface disease detection system based on deep learning is provided, including: a municipal road surface data acquisition module, which is used for the UAV to collect natural environment data and road surface image data of the municipal road surface at position time series by carrying sensors and cameras.

[0013] A road surface image quality analysis module, which is used for coupling the natural environment data with historical illumination data to analyze the environmental quality index of the municipal road surface at the current moment, and triggering a flight parameter adjustment mechanism when the environmental quality index is within the set environmental quality index range.

[0014] A road surface image re-acquisition module, which is used for adjusting the image re-acquisition flight parameters of the UAV based on the flight parameter adjustment mechanism, and using the re-acquired images with qualified image quality indicators as the final road surface images.

[0015] A road surface disease assessment and feedback module, which is used for identifying the types of road surface diseases and the set of characteristic parameters corresponding to various road surface diseases from the collected road surface images, constructing a disease risk assessment index for the municipal road surface and an assessment index for the development trend of various road surface diseases, outputting the disease risk level and the influence trend level of various road surface diseases and corresponding feedback.

[0016] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects: (1) By combining the sensors and cameras carried by the drone, the present invention synchronously collects natural environment data and road surface images, and dynamically adjusts the flight parameters (speed, height) through the environmental quality index to ensure the image acquisition quality and improve the detection accuracy in complex environments.

[0017] (2) By constructing a multi-disease coupling risk model, integrating the regional weight and traffic volume dynamic weight, the present invention outputs the disease risk level, breaks through the traditional single-disease assessment mode, realizes multi-dimensional risk quantification and dynamic adaptive optimization, can not only capture the synergistic effects of diseases such as cracks and depressions, but also adjust the regional weight in real time according to the road function level and surrounding environment, and dynamically correct the risk coefficient in combination with traffic flow, load, etc., to form an accurate risk portrait.

[0018] (3) By using the first-order exponential smoothing method to predict the development trend of diseases, the present invention realizes the dual functions of "current situation assessment + trend early warning", predicts the evolution trend of diseases based on historical data, significantly improves the scientificity and timeliness of municipal road risk management, reduces maintenance costs and prevents safety accidents. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0020] Figure 1 It is a schematic flowchart of the method steps of the present invention.

[0021] Figure 2 It is a schematic connection diagram of the system structure of the present invention.

[0022] Figure 3 It is a schematic flowchart for judging the trigger flight parameter adjustment mechanism of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0023] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0024] Please refer to Figure 1As shown in the figure, the first aspect of the present invention provides an automatic road surface disease detection method based on deep learning, including: S1. The drone collects natural environment data and road surface image data of the municipal road surface at the position time sequence by carrying sensors and cameras.

[0025] S2. Coupling the natural environment data with the historical light data to analyze the environmental quality index of the municipal road surface at the current moment, and triggering the flight parameter adjustment mechanism when the environmental quality index is within the set environmental quality index range.

[0026] In a specific embodiment of the present invention, the specific process of coupling the natural environment data with the historical light data to analyze the environmental quality index of the municipal road surface at the current moment is as follows: Coupling the light intensity, environmental humidity, and visibility at the current moment in the natural environment data with the maximum light intensity, environmental humidity critical threshold, and minimum visibility for image acquisition stored in the database at the historical light moment to obtain the environmental quality index of the municipal road surface at the current moment.

[0027] It should be noted that the process of obtaining the environmental quality index of the municipal road surface at the current moment is as follows: First step, taking the ratio of the current light intensity to the maximum light intensity at the historical light moment, squaring the ratio result, and then multiplying it by the weight coefficient of the light intensity. Second step, taking the ratio of the current environmental humidity to the humidity critical threshold, taking the ratio result as the power and substituting it into the exponential function with e as the base, and then multiplying the function result by the weight coefficient of the environmental humidity. Third step, taking the ratio of the minimum visibility for image acquisition to the current visibility, squaring the ratio result, and then multiplying it by the weight coefficient of the visibility. Fourth step, adding the calculation results of the above three steps to obtain the final environmental quality index.

[0028] It should also be noted that visibility directly determines the image clarity and recognizability, light intensity affects the image brightness and contrast, and high environmental humidity may cause road surface reflection, water accumulation or fog, indirectly affecting the image quality. Therefore, in the environmental quality assessment, the importance of visibility is greater than that of light intensity and environmental humidity. Therefore, in a specific embodiment of the present invention, the weight coefficients corresponding to light intensity, environmental humidity, and visibility for environmental quality assessment can be taken as 0.5, 0.3, and 0.2 respectively.

[0029] Among them, the light intensity, environmental humidity, and visibility at the current moment in the natural environment data are respectively detected by a photodiode, a humidity sensor, and a transmissive visibility sensor. The detection principles and processes of the above three are all prior arts and will not be elaborated here.

[0030] Please refer to Figure 3As shown, compare the environmental quality index of the municipal road surface at the current moment with the range of environmental quality indexes corresponding to the trigger flight parameter adjustment mechanism stored in the database. If the environmental quality index of the municipal road surface at the current moment is within the range of environmental quality indexes corresponding to the trigger flight parameter adjustment mechanism, it indicates that the quality of the collected road surface image reaches the trigger flight parameter adjustment mechanism; otherwise, it indicates that the quality of the collected road surface image does not reach the trigger flight parameter adjustment mechanism.

[0031] S3. Adjust the image re - acquisition flight parameters of the drone based on the flight parameter adjustment mechanism, and use the re - acquired images with qualified image quality indicators as the final road surface images.

[0032] In a specific embodiment of the present invention, the specific process of adjusting the image re - acquisition flight parameters of the drone is as follows: Perform coupling processing on the current flight speed in the current flight parameters of the drone and the minimum flight speed preset for the drone stored in the database to obtain the image re - acquisition flight speed. The current flight parameters of the drone are directly extracted from the drone control background.

[0033] It should be noted that, through the formula the image re - acquisition flight speed is obtained, where represents the environmental quality index, represents the minimum flight speed preset for the drone stored in the database, represents the current flight speed in the current flight parameters of the drone, and respectively represent the speed adjustment ratio coefficient and the environmental quality index threshold.

[0034] In a specific embodiment of the present invention, is set to 0.2, is set to 1.2. As the speed adjustment ratio coefficient, if the value is too large, the speed attenuation amplitude is too high, which may cause the flight attitude of the drone to get out of control; if the value is too small, it cannot effectively respond to the demand for speed reduction due to environmental improvement. Therefore, through a large number of debugging and verification, a ratio of 0.2 can not only appropriately reduce the flight speed to obtain clearer images when the environmental quality improves, but also avoid excessive speed fluctuations and ensure the flight stability of the drone. is the environmental quality index threshold. This threshold of 1.2 is obtained through the correlation analysis of historical environmental data and image quality, which can not only avoid frequent speed adjustment due to insignificant improvement of environmental quality, but also trigger the speed reduction mechanism in a timely manner when the environment is suitable to improve the image acquisition effect.

[0035] Perform comprehensive analysis on the current flight height in the current flight parameters of the drone and the maximum flight height preset for the drone stored in the database to obtain the image re - acquisition flight height.

[0036] It should be noted that the flight altitude for image re - acquisition is obtained through the formula , where represents the maximum preset flight altitude of the drone stored in the database, represents the current flight altitude in the current flight parameters of the drone, and respectively represent the altitude adjustment base coefficient and the adjustment factor of the environmental quality on the altitude.

[0037] In a specific embodiment of the present invention, is set to 0.8, is set to 0.1. 0.8 is the altitude adjustment base coefficient. If has too large a value, the altitude adjustment will overly rely on the base coefficient, weakening 's adjustment effect on the altitude. If the value is too small, it may lead to too large an altitude adjustment range, affecting the flight safety of the drone. After testing and verification, 0.8 can achieve a balance between the stable benchmark and dynamic adjustment, ensuring that the altitude adjustment meets the actual engineering requirements. 0.1 is the balance factor for adjusting 's influence on the altitude. If the value is too large, the altitude change amplitude is too drastic. If it is too small, the adjustment effect is not obvious. After testing, 0.1 can ensure that the altitude changes reasonably with : when the environmental quality is high, the altitude is reduced to obtain a clearer image; when the environmental quality is low, the altitude adjustment slows down to avoid abnormal altitude caused by environmental interference.

[0038] In the embodiment of the present invention, by combining the sensors and cameras carried by the drone, natural environment data and road surface images are synchronously collected, and the flight parameters (speed, altitude) are dynamically adjusted through the environmental quality index (EQI) to ensure the image acquisition quality and improve the detection accuracy in complex environments.

[0039] In a specific embodiment of the present invention, the specific analysis process of taking the re - acquisition image with the image quality index meeting the standard as the final road surface image is as follows: obtain the pixel values of each pixel point from each re - acquisition image, calculate the average value of the pixel gradient values of each re - acquisition image, and use it as the clarity of each re - acquisition image.

[0040] It should be noted that the formula for calculating the average value of the pixel gradient values of each re - acquisition image is: , where represents the horizontal and vertical coordinates of the th pixel point in the re - acquisition image, represents the pixel value at the coordinates of the th pixel point, represents the second - order differential operator, represents the pixel point number, , represents the number of pixel points.

[0041] Based on the pixel brightness of each pixel point obtained from each multiple acquisition image, the brightness variance of each multiple acquisition image is obtained and used as the illumination uniformity of each multiple acquisition image.

[0042] It should be noted that the formula for calculating the brightness variance of each multiple acquisition image is: , where represents the pixel brightness at the th pixel point coordinate, represents .

[0043] Select the multiple acquisition image with the highest clarity and the largest illumination uniformity as the final road surface image.

[0044] S4. Identify the types of road surface diseases and the corresponding characteristic parameter sets of various road surface diseases from the collected road surface images, construct the disease risk assessment index and the development trend assessment index of various road surface diseases of the municipal road surface, output the disease risk level and the influence trend level of various road surface diseases and give corresponding feedback.

[0045] In a specific embodiment of the present invention, the specific method for identifying the types of road surface diseases from the collected road surface images is as follows: Based on the historical road surface disease atlas, construct the disease feature sets corresponding to various road surface diseases, preprocess the collected road surface images to obtain each image feature, compare each image feature with the disease feature sets corresponding to various road surface diseases, and if an image feature is located in the disease feature set corresponding to a certain type of road surface disease, then regard this type of road surface disease as the type of road surface disease of the municipal road surface, and thus obtain the types of road surface diseases of the municipal road surface.

[0046] In a specific embodiment of the present invention, the specific process of constructing the disease risk assessment index of the municipal road surface is as follows: According to the characteristic parameter sets corresponding to various road surface diseases, obtain the risk indexes corresponding to various road surface diseases, and substitute them into the composite disease superposition formula to obtain the multi-disease coupling risk index of the market road surface.

[0047] In a specific embodiment of the present invention, the types of road surface diseases include but are not limited to cracks, depressions and deep pits.

[0048] It should be noted that the specific process of obtaining the risk indexes corresponding to various road surface diseases according to the characteristic parameter sets corresponding to various road surface diseases is as follows: Extract the total extension length and maximum opening width corresponding to the crack disease, the projected area and maximum depth corresponding to the depression disease, and the volume corresponding to the deep pit disease from the characteristic parameter sets corresponding to various road surface diseases.

[0049] It should be noted that the acquisition methods for the total extension length and maximum opening width corresponding to the crack disease are as follows: Using image recognition technology to automatically calculate and statistically analyze the total crack length, and analyzing the pixel ratio through software to convert the actual maximum opening width. The acquisition methods for the projected area and maximum depth corresponding to the depression disease are as follows: Taking multi-angle images of the depression area by a drone, constructing a 3D model, and extracting the projected area and maximum depth of the depression disease from the 3D model. The acquisition method for the volume corresponding to the deep pit disease is as follows: Similarly, taking multi-angle images of the deep pit by a drone to generate a 3D model, and automatically calculating the accurate volume through software.

[0050] Perform dynamic difference calculations on the total extension length and maximum opening width corresponding to the crack disease respectively with the corresponding set values, and sum the results after normalizing each dynamic difference to obtain the crack risk index of the market road surface.

[0051] It should be noted that the dynamic difference calculation is as follows: Subtract the crack warning length set by the municipal road surface safety criterion from the total extension length to obtain the crack length difference, and subtract the safety width set by the municipal road surface safety criterion from the maximum opening width to obtain the crack width difference. The normalization process is as follows: Divide the crack length difference by the crack warning length to obtain the normalized crack length difference, and divide the crack width difference by the safety width to obtain the normalized crack width difference.

[0052] Take the ratio of the projected area and maximum depth corresponding to the depression disease respectively with the corresponding set values and sum them, and use the sum result as the power to substitute into the exponential function with base e to obtain the depression risk index of the market road surface.

[0053] When the volume corresponding to the deep pit disease is less than or equal to the deep pit volume set by the municipal road surface safety criterion, take the ratio of the volume corresponding to the deep pit disease to the deep pit volume set by the municipal road surface safety criterion as the deep pit risk index of the market road surface. When the volume corresponding to the deep pit disease is greater than the deep pit volume set by the municipal road surface safety criterion, take the result of squaring the ratio of the volume corresponding to the deep pit disease to the deep pit volume set by the municipal road surface safety criterion as the deep pit risk index of the market road surface.

[0054] It should also be noted that the composite disease superposition formula is: , where respectively represent the crack risk index, depression risk index, and deep pit risk index of the market road surface, Represents the second-highest index. Among multiple disease indicators (CSI, DSI, PRI), the second-highest index refers to the second-highest value excluding the maximum value. By introducing the second-highest index, the formula not only focuses on the most significant disease risks but also takes into account secondary risks, avoiding evaluation biases caused by ignoring other diseases. At the same time, in the engineering evaluation model, 0.3 is a verified compromise parameter. If the coefficient is too high (such as above 0.5), the second-highest index has too much influence on the results and may mask the main risks. If it is too low (such as below 0.1), the original intention of "taking into account secondary risks" cannot be reflected. Through repeated debugging, 0.3 makes the model more in line with the actual harm degree of disease superposition in the trade-off between primary and secondary risks.

[0055] Extract the road area type to which the municipal road surface belongs from the database, and compare it with the area weights corresponding to each road area type to obtain the area weight corresponding to the municipal road surface.

[0056] In a specific embodiment of the present invention, the road area types include but are not limited to main roads, secondary roads, living area roads, and industrial area roads. The area weights corresponding to each road area type are scored by organizing traffic engineering experts and municipal maintenance personnel based on dimensions such as road function positioning, traffic flow carrying capacity, and importance of surrounding facilities, in combination with industry maintenance specifications, to form an initial weight framework and store it in the municipal road management system.

[0057] Extract the average daily traffic volume of the municipal road surface, and obtain the traffic volume dynamic weight corresponding to the municipal road surface by taking the ratio of the average daily traffic volume to the reference traffic volume of the municipal road surface.

[0058] It should be noted that the average daily traffic volume of the municipal road surface can be extracted from the Municipal Transportation Bureau.

[0059] Multiply the multi-disease coupling risk index, area weight, and traffic volume dynamic weight of the market road surface to obtain the disease risk assessment index of the municipal road surface.

[0060] In a specific embodiment of the present invention, the specific method for outputting the disease risk level is as follows: Compare the disease risk assessment index of the municipal road surface with the disease risk assessment index ranges corresponding to each disease risk level stored in the database. If the disease risk assessment index of the municipal road surface is within the disease risk assessment index range corresponding to a certain disease risk level, then take that disease risk level as the disease risk level of the municipal road surface.

[0061] In the embodiment of the present invention, by constructing a multi-disease coupling risk model, integrating the regional weight and the traffic volume dynamic weight, the disease risk level is output, breaking through the traditional single-disease evaluation mode, realizing multi-dimensional risk quantification and dynamic adaptive optimization, which can not only capture the synergistic effects of diseases such as cracks and depressions, but also adjust the regional weight in real time according to the road function level and the surrounding environment, and combine the traffic flow, load, etc. to dynamically correct the risk coefficient, forming an accurate risk portrait.

[0062] In a specific embodiment of the present invention, the specific process of constructing the evaluation index of the development trend of various pavement diseases on the municipal road surface is as follows: extract the pavement images corresponding to each historical collection from the pavement image data of the municipal road surface in the position time series, and analyze the risk indexes corresponding to various pavement diseases in each historical collection in the same way according to the risk index analysis method corresponding to various pavement diseases.

[0063] According to the predicted value and the actual value of the development trend evaluation index corresponding to the previous historical collection of various pavement diseases, construct the evaluation index of the development trend of various pavement diseases on the municipal road surface by the single exponential smoothing method.

[0064] It should be noted that constructing the evaluation index of the development trend of various pavement diseases on the municipal road surface by the single exponential smoothing method: , where, represents the evaluation index of the development trend of the th type of pavement disease on the municipal road surface, is the predicted value of the development trend evaluation index corresponding to the previous historical collection of the th type of pavement disease, is the actual value of the development trend evaluation index corresponding to the previous historical collection of the th type of pavement disease, is the smoothing coefficient, and the value range is , represents the serial number of the pavement disease type, .

[0065] In a specific embodiment of the present invention, the and can be illustrated by the data in Table 1 below. Here, we assume that the risk index data corresponding to three consecutive historical collections of a certain type of pavement disease on the municipal road surface are shown in Table 1 below:

[0066] Table 1 Risk index data

[0067]

[0068] In the single exponential smoothing method, we need to set the smoothing coefficient , here we assume , then according to the data in the table, the actual value of the risk index corresponding to the third historical collection of this type of road surface disease is 8, and the predicted value of the risk index corresponding to the third historical collection of this type of road surface disease is , then the predicted value of the risk index corresponding to the fourth historical collection of this type of road surface disease is: .

[0069] It should be noted that The smaller it is, the stronger the "memory" of historical data, and the greater the influence of past values on the prediction result. The larger it is, the more it focuses on the latest data. 0.3 is a commonly used compromise value, which can not only retain a certain historical trend but also respond moderately to new data. Therefore, the smoothing coefficient is taken as 0.3 here.

[0070] It should also be noted that the single exponential smoothing method constructs an evaluation index for the development trend of diseases, which has the characteristics of simple calculation and easy implementation. By integrating the formula of historical actual values and predicted values, it can efficiently process time series data, not only filter out random noise to restore the true evolution trend of diseases, but also dynamically update the evaluation results based on newly collected data to support the prediction of disease development. At the same time, it has a low demand for historical data volume, and trends can be mined even with limited data, providing a practical and dynamic analysis tool for the evaluation and maintenance decision-making of municipal road surface diseases.

[0071] In a specific embodiment of the present invention, the specific manner of outputting the influence trend levels of various types of road surface diseases is as follows: The specific manner of outputting the influence trend levels of various types of road surface diseases is as follows: Respectively extract the intervals of the road surface disease development trend evaluation indexes corresponding to the low risk level, medium risk level, and high risk level from the database. If the road surface disease development trend evaluation index of a certain type of road surface disease is within the low risk level interval, then the influence trend level of this type of road surface disease is the low risk level. If the road surface disease development trend evaluation index of a certain type of road surface disease is within the medium risk level interval, then the influence trend level of this type of road surface disease is the medium risk level. When the road surface disease development trend evaluation index of a certain type of road surface disease is within the high risk level interval, then the influence trend level of this type of road surface disease is the high risk level.

[0072] In a specific embodiment of the present invention, the intervals of the road surface disease development trend evaluation indexes corresponding to the low risk level, medium risk level, and high risk level are respectively , and , as the boundary between low risk and medium risk, indicates that when the disease impact trend index is less than 0.4, the disease is in a relatively stable state without significant signs of deterioration. When it exceeds 0.4, it begins to enter the slow development stage and needs to be included in the monitoring scope. This threshold is determined by analyzing the turning point from "stable to developing" in historical disease data, matching the initial evolution characteristics of the disease. 0.7 is the boundary between medium risk and high risk. When it exceeds 0.7, it indicates that the disease deterioration speed accelerates and urgent treatment is required. This value is based on engineering experience, corresponding to the critical state of the disease from "slow development" to "rapid deterioration", ensuring that the risk warning fits the actual hazard escalation node. 1 is the upper limit of the index, representing the extreme state of the disease impact trend, clarifying the boundary of the risk level and ensuring the closure of the grading logic.

[0073] In the embodiment of the present invention, the development trend of the disease is predicted by the single exponential smoothing method, realizing the dual functions of "current situation assessment + trend warning". Based on historical data, the evolution trend of the disease is predicted, significantly improving the scientificity and timeliness of municipal road risk management, reducing maintenance costs and preventing safety accidents.

[0074] Refer to Figure 2 As shown, the second aspect of the present invention provides a pavement disease automatic detection system based on deep learning, including: a municipal road surface data acquisition module, a road surface image quality analysis module, a road surface image re-acquisition module, and a road surface disease assessment and feedback module.

[0075] It should be noted that the present invention also includes a database for storing the maximum light intensity in historical light moments, the critical threshold of environmental humidity, and the minimum visibility for image acquisition, storing the environmental quality index range corresponding to triggering the flight parameter adjustment mechanism, storing the minimum flight speed and maximum flight height preset by the unmanned aerial vehicle, storing the road area type to which the municipal road surface belongs, and storing the pavement disease development trend assessment index intervals corresponding to the low risk level, medium risk level, and high risk level respectively.

[0076] The municipal road surface data acquisition module is connected to the road surface image quality analysis module, both the road surface image re-acquisition module and the road surface disease assessment and feedback module are connected to the road surface image quality analysis module, the road surface image re-acquisition module is connected to the road surface disease assessment and feedback module, and the road surface image quality analysis module, the road surface image re-acquisition module, and the road surface disease assessment and feedback module are all connected to the database.

[0077] The municipal road surface data acquisition module is used for the unmanned aerial vehicle to collect the natural environment data and road surface image data of the municipal road surface at position time series by carrying sensors and cameras.

[0078] The road surface image quality analysis module is used to couple and analyze natural environment data and historical lighting data to obtain the environmental quality index of the municipal road surface at the current moment. When the environmental quality index is within the set environmental quality index range, a flight parameter adjustment mechanism is triggered.

[0079] The road surface image re-acquisition module is used to adjust the image re-acquisition flight parameters of the unmanned aerial vehicle based on the flight parameter adjustment mechanism, and use the re-acquired images with qualified image quality indicators as the final road surface images.

[0080] The road surface disease assessment and feedback module is used to identify the types of road surface diseases and the corresponding characteristic parameter sets of various road surface diseases from the collected road surface images, construct the disease risk assessment index and the development trend assessment index of various road surface diseases of the municipal road surface, and output the disease risk level and the influence trend level of various road surface diseases and provide corresponding feedback.

[0081] The above content is only an example and illustration of the concept of the present invention. Those skilled in the art of the present technology can make various modifications or supplements or use similar methods to replace the specific embodiments described, as long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, they should all fall within the protection scope of the present invention.

Claims

1. An automatic pavement disease detection method based on deep learning, characterized in that, Including: S1. The drone collects natural environment data and road surface image data of the municipal road surface at position time series by carrying sensors and cameras; S2. Coupling the natural environment data with historical light data to analyze the environmental quality index of the municipal road surface at the current moment. When the environmental quality index is within the set environmental quality index range, trigger the flight parameter adjustment mechanism; S3. Adjust the image re-acquisition flight parameters of the drone based on the flight parameter adjustment mechanism, and use the re-acquired images with qualified image quality indicators as the final road surface images; S4. Identify the types of road surface diseases and the set of characteristic parameters corresponding to various road surface diseases from the collected road surface images, construct the disease risk assessment index of the municipal road surface and the development trend assessment index of various road surface diseases, and output the disease risk level and the influence trend level of various road surface diseases and give corresponding feedback.

2. The automatic pavement disease detection method based on deep learning according to claim 1, wherein: The specific process of analyzing the environmental quality index of the municipal road surface at the current moment is as follows: Couple the light intensity, environmental humidity and visibility at the current moment in the natural environment data with the maximum light intensity, environmental humidity critical threshold and minimum visibility for image acquisition in the historical light moments stored in the database to obtain the environmental quality index of the municipal road surface at the current moment; Compare the environmental quality index of the municipal road surface at the current moment with the environmental quality index range corresponding to the trigger flight parameter adjustment mechanism stored in the database. If the environmental quality index of the municipal road surface at the current moment is within the environmental quality index range corresponding to the trigger flight parameter adjustment mechanism, it indicates that the quality of the collected road surface image reaches the trigger flight parameter adjustment mechanism. Otherwise, it indicates that the quality of the collected road surface image does not reach the trigger flight parameter adjustment mechanism.

3. The automatic pavement disease detection method based on deep learning according to claim 2, characterized in that: The specific process of adjusting the image re-acquisition flight parameters of the drone is as follows: Couple the current flight speed in the current flight parameters of the drone with the minimum flight speed preset for the drone stored in the database to obtain the image re-acquisition flight speed; Comprehensively analyze the current flight height in the current flight parameters of the drone and the maximum flight height preset for the drone stored in the database to obtain the image re-acquisition flight height.

4. The automatic pavement disease detection method based on deep learning according to claim 3, characterized in that: The specific analysis process of using the re-acquired images with qualified image quality indicators as the final road surface images is as follows: According to the pixel values of each pixel point obtained from each re-acquired image, calculate the average value of the pixel gradient values of each re-acquired image and use it as the clarity of each re-acquired image; According to the pixel brightness of each pixel point obtained from each re-acquired image, obtain the brightness variance of each re-acquired image and use it as the light uniformity of each re-acquired image; Select the re-acquired image with the highest clarity and the largest light uniformity as the final road surface image.

5. The automatic pavement disease detection method based on deep learning according to claim 1, characterized in that: The specific method for identifying the types of road surface diseases from the collected road surface images is as follows: Based on the historical road surface disease atlas, a set of disease characteristics corresponding to various road surface diseases is constructed. The collected road surface images are preprocessed to obtain various image features, and these image features are compared with the set of disease characteristics corresponding to various road surface diseases. If an image feature is within the set of disease characteristics corresponding to a certain type of road surface disease, then this type of road surface disease is regarded as the type of road surface disease of the municipal road surface, and thus the types of various road surface diseases of the municipal road surface are obtained.

6. The automatic pavement disease detection method based on deep learning according to claim 1, characterized in that: The specific process of constructing the disease risk assessment index of the municipal road surface is as follows: According to the set of characteristic parameters corresponding to various road surface diseases, the risk index corresponding to various road surface diseases is obtained, and it is brought into the compound disease superposition formula to obtain the multi-disease coupling risk index of the market road surface; Extract the road area type to which the municipal road surface belongs from the database, and compare it with the area weights corresponding to each road area type to obtain the area weight corresponding to the municipal road surface; Extract the average daily traffic volume of the municipal road surface, and obtain the traffic volume dynamic weight corresponding to the municipal road surface by taking the ratio of it to the reference traffic volume of the municipal road surface; Multiply the multi-disease coupling risk index, area weight, and traffic volume dynamic weight of the market road surface to obtain the disease risk assessment index of the municipal road surface.

7. The automatic pavement disease detection method based on deep learning according to claim 6, characterized in that: The specific method for outputting the disease risk level is as follows: Compare the disease risk assessment index of the municipal road surface with the range of disease risk assessment indexes corresponding to each disease risk level stored in the database. If the disease risk assessment index of the municipal road surface is within the range of the disease risk assessment index corresponding to a certain disease risk level, then this disease risk level is regarded as the disease risk level of the municipal road surface.

8. The automatic pavement disease detection method based on deep learning according to claim 6, wherein: The specific process of constructing the development trend assessment index of various road surface diseases of the municipal road surface is as follows: Extract the road surface images corresponding to each historical collection from the road surface image data of the municipal road surface in the position time series, and analyze the risk indexes corresponding to various road surface diseases in each historical collection in the same way as the risk index analysis method corresponding to various road surface diseases; According to the predicted value and actual value of the development trend assessment index corresponding to the previous historical collection of various road surface diseases, construct the development trend assessment index of various road surface diseases of the municipal road surface by the single exponential smoothing method.

9. The automatic pavement disease detection method based on deep learning according to claim 8, characterized in that: The specific method for outputting the influence trend level of various road surface diseases is as follows: Extract the intervals of the development trend assessment indexes corresponding to the low risk level, medium risk level, and high risk level respectively from the database. If the development trend assessment index of a certain type of road surface disease is within the low risk level interval, then the influence trend level of this type of road surface disease is the low risk level. If the development trend assessment index of a certain type of road surface disease is within the medium risk level interval, then the influence trend level of this type of road surface disease is the medium risk level. When the development trend assessment index of a certain type of road surface disease is within the high risk level interval, then the influence trend level of this type of road surface disease is the high risk level.

10. An automatic pavement disease detection system based on deep learning, characterized in that, Including: The municipal road surface data collection module is used for the drone to collect the natural environment data and road surface image data of the municipal road surface in the position time series by carrying sensors and cameras; The road surface image quality analysis module is used to couple and analyze the natural environment data and historical lighting data to obtain the environmental quality index of the municipal road surface at the current moment. When the environmental quality index is within the set range of the environmental quality index, it triggers the flight parameter adjustment mechanism; The road surface image re-acquisition module is used to adjust the image re-acquisition flight parameters of the drone based on the flight parameter adjustment mechanism, and use the re-acquired images with qualified image quality indicators as the final road surface images; The road surface disease assessment and feedback module is used to identify the types of road surface diseases and the corresponding characteristic parameter sets of various road surface diseases from the collected road surface images, construct the disease risk assessment index and the development trend assessment index of various road surface diseases of the municipal road surface, and output the disease risk level and the influence trend level of various road surface diseases and give corresponding feedback.

Images