Pavement quality detection method and device based on artificial intelligence

Abstract

The invention relates to the technical field of artificial intelligence, in particular to a pavement quality detection method and device based on artificial intelligence, and the method comprises the following steps: obtaining pavement images and load data, extracting crack boundary coordinates through edge detection, calculating a displacement difference to obtain crack distribution, calculating an extension rate in combination with time, and carrying out the weighted analysis of a direction difference. The method comprises the following steps: extracting crack boundary displacement difference, identifying defect evolution parameters, calculating stability and environmental sensitivity coefficients through cross coupling, judging degradation types in combination with load fluctuation, calculating defect grades through classification comparison threshold values of a support vector machine, and outputting a pavement overall quality result. Angle difference variation recognizes evolution direction difference, coupling extension rate and direction consistency depicts stability, environmental sensitivity and load fluctuation linkage calculation is introduced to present external action association, and grading is completed based on parameter and threshold difference to realize dynamic grading recognition of defect development rules.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method and apparatus for road surface quality detection based on artificial intelligence. Background Technology

[0002] The field of artificial intelligence technology involves research and applications related to simulating human intelligent activities using computers. Its core aspects include the construction and implementation of methods such as machine learning, deep learning, pattern recognition, and computer vision. Through the collection, feature extraction, and training of massive amounts of data, this field can complete the recognition and analysis of images, voice, text, and environmental states. It is widely used in scenarios such as transportation, medical care, industrial inspection, and intelligent manufacturing, forming a comprehensive technical system based on data-driven approaches and combining algorithm models with sensor acquisition methods.

[0003] Traditional road surface quality inspection methods refer to monitoring the road surface through manual inspection or using vehicles equipped with measuring devices. This involves manually observing road surface defects such as cracks and potholes, or using accelerometers and laser rangefinders installed on the inspection vehicle to collect data on road surface smoothness and texture depth, thereby determining the quality of the road surface.

[0004] Current detection methods rely on manual methods or vehicle-mounted devices to obtain data on cracks and road surface conditions. Manual observation has the drawbacks of limited scope and insufficient accuracy. Although vehicle-mounted sensors can collect indicators such as smoothness, the results are mostly isolated and static information, lacking a quantitative description of the crack development process in the time series. They cannot reveal the evolution trend of cracks with load fluctuations, resulting in the detection results only reflecting the surface condition at a certain point in time. In the long-term use scenario of the road surface, potential deterioration risks are easily overlooked, causing maintenance strategies to lag behind the actual change process, increasing road safety hazards and maintenance uncertainty. Summary of the Invention

[0005] To address the shortcomings of existing methods that rely on manual or vehicle-mounted devices to acquire crack and pavement condition data, which suffer from limited scope and accuracy, and the fact that vehicle-mounted sensors, while capable of collecting indicators such as smoothness, often provide isolated, static information lacking a quantitative description of crack development over time and failing to reveal the evolution trend of cracks with load fluctuations, resulting in detection results that only reflect the surface condition at a specific point in time. This can easily overlook potential deterioration risks in long-term pavement use scenarios, causing maintenance strategies to lag behind actual changes and increasing road safety hazards and maintenance uncertainty. Therefore, this invention provides an artificial intelligence-based pavement quality inspection method, comprising the following steps: To achieve the above objectives, the present invention adopts the following technical solution: a road surface quality detection method based on artificial intelligence, comprising the following steps: S1: Obtain surface image data of the road surface through image acquisition equipment, collect load data, process the road image data with edge detection algorithm, extract the set of crack boundary coordinate points, calculate the displacement difference of boundary coordinate points, and obtain the distribution state of road crack defects. S2: Calculate the unit time extension rate based on the displacement difference of coordinate points and sampling time in the distribution state of the road crack defects, perform time-dimensional weighted accumulation processing on the extension rate, analyze the degree of variation of the angle difference in the extension direction, and identify the characteristic parameters of the road defect evolution. S3: Call the time-weighted extension rate and directional consistency coefficient in the pavement defect evolution characteristic parameters for cross-coupling calculation, analyze the stability index of crack boundary evolution, calculate the environmental sensitivity coupling coefficient, perform dynamic correlation analysis based on load time-series fluctuation characteristics, and determine the pavement quality deterioration type characteristics; S4: Perform support vector machine classification processing on the characteristics of the road quality deterioration type, compare the stability index of crack boundary evolution with the preset stability threshold, calculate the difference between the dynamic correlation analysis result and the load fatigue benchmark value, and determine the quality defect level of the road area.

[0006] As a further aspect of the present invention, the distribution state of pavement crack defects includes crack morphology, crack range and crack spatial location; the pavement defect evolution characteristic parameters include the amplitude of extension rate change, directional angle variability and time accumulation characteristics; the crack boundary evolution stability index includes stability level, environmental sensitivity coefficient and load coupling factor; and the pavement area quality defect level includes mild defect, moderate defect and severe defect.

[0007] As a further aspect of the present invention, the specific steps of S1 are as follows: S101: Acquire surface image data of the road surface through image acquisition equipment, compare the gray-scale pixel gradient values ​​of the image data with the edge detection threshold, record the coordinates of pixels with gradient values ​​greater than the threshold, aggregate and sort the coordinates in sequence, and obtain the set of crack boundary coordinate points. S102: Based on the set of coordinate points of the crack boundary, collect load data, calculate the horizontal and vertical difference values ​​of adjacent coordinate points, match the difference values ​​with the load data index and integrate them into a sequence, compare the sequence index by index and record the difference values ​​to generate a crack boundary displacement difference sequence. S103: Call the crack boundary displacement difference sequence and combine it with the corresponding values ​​of the load data. Compare the displacement difference with the crack distribution benchmark value, perform clustering annotation on the difference index that deviates from the benchmark value, and aggregate it in the order of regions to obtain the distribution status of road crack defects.

[0008] As a further aspect of the present invention, the specific steps of S2 are as follows: S201: Based on the displacement difference of coordinate points and sampling time in the distribution state of road crack defects, the displacement difference is compared with the adjacent sampling time interval, and the rate parameters are arranged in time order to obtain the unit time extension rate sequence. S202: Based on the unit time extension rate sequence, the rate values ​​of adjacent time periods are weighted by the sampling interval, and the weighting factor is called to perform cumulative operation in continuous time periods to obtain a time dimension cumulative rate set; S203: Call the accumulated rate set of the time dimension, calculate the angle difference between the crack displacement difference direction vector and the reference direction vector, and statistically summarize the fluctuation amplitude of the angle difference to obtain the pavement defect evolution characteristic parameters.

[0009] As a further aspect of the present invention, the specific steps of S3 are as follows: S301: Based on the time-weighted extension rate and directional consistency coefficient in the pavement defect evolution characteristic parameters, perform a vector inner product operation on the two, and compare the difference between the slope of the inner product curve and the directional consistency coefficient to generate a crack boundary evolution stability index. S302: Call the crack boundary evolution stability index, calculate the index with the temperature sensitivity factor and humidity sensitivity factor one by one, compare the results with the sensitivity benchmark value and perform weight integration to obtain the environmental sensitivity coupling coefficient. S303: Based on the environmental sensitivity coupling coefficient, the amplitude range and period range of the load time-series fluctuation characteristics are mapped, and the dynamic offset amplitude corresponding to the interval is compared to obtain the characteristics of the road quality deterioration type.

[0010] As a further aspect of the present invention, the specific steps of S4 are as follows: S401: Based on the characteristics of the road quality deterioration type, construct a feature matrix and input it into the classification processing unit. Compare the crack morphology parameters according to the support vector boundary division, mark the region according to the category index, and generate crack classification results. S402: Call the crack classification results, retrieve the boundary change sequence and extract the discrete point stability index, compare the index with the preset stability threshold, record the index with insufficient difference and encode it to obtain the crack boundary stability set. S403: Based on the set of crack boundary stability, call the dynamic correlation analysis value corresponding to the set index, perform difference calculation with the load fatigue benchmark value, map the difference to the defect level classification table, and obtain the quality defect level of the pavement area.

[0011] As a further aspect of the present invention, the preset stability threshold is the lower limit of the allowable value of the stability index of discrete points at the crack boundary; The load fatigue benchmark value is the reference value of the fatigue load that the pavement structure can withstand under design conditions.

[0012] As a further aspect of the present invention, the method further includes step S5: S5: Based on the quality defect level of the road surface area, evaluate the condition of the inspected road surface, determine the area corresponding to the level as the corresponding quality state, and output the overall quality inspection result of the road surface; The overall road surface quality inspection results include grade assessment, regional condition, and comprehensive conclusions.

[0013] As a further aspect of the present invention, the specific steps of S5 are as follows: S501: Based on the quality defect level of the road surface area, the pixel gray value in the surface image matrix of the detection area is compared with the defect level judgment threshold. When the gray value is lower than the threshold, the corresponding area is marked as a low-quality block, and the marked block indexes are aggregated to obtain the defect area index set. S502: Call the defect area index set, extract structural integrity parameters for the corresponding road surface area data, compare them with the quality grading benchmark value, determine the area level based on the comparison result, and generate a road surface area quality level sequence. S503: Based on the road surface area quality grade sequence, classify and summarize the different grade areas, perform weighted calculations according to their proportions, and generate the overall road surface quality test results.

[0014] An artificial intelligence-based road surface quality detection device includes: The image acquisition module acquires surface image data and load data through image acquisition equipment, processes the surface image data using edge detection algorithm, extracts the set of crack boundary coordinate points, calculates the displacement difference of boundary coordinate points, obtains the distribution state of road crack defects, and transmits it to the crack extraction module. The crack extraction module calculates the unit time extension rate based on the displacement difference of coordinate points in the distribution state of the road crack defects and the sampling time. It performs time-dimensional weighted accumulation processing on the extension rate, analyzes the degree of variation of the angle difference in the extension direction, obtains the road defect evolution characteristic parameters, and transmits them to the evolution analysis module. The evolution analysis module calls the time-weighted extension rate and directional consistency coefficient in the pavement defect evolution characteristic parameters for cross-coupling calculation, analyzes the stability index of crack boundary evolution, calculates the environmental sensitivity coupling coefficient, performs dynamic correlation analysis based on load time-series fluctuation characteristics, determines the pavement quality deterioration type characteristics, and transmits them to the quality assessment module. The quality assessment module calls the pavement quality deterioration type features to perform support vector machine classification, compares the crack boundary evolution stability index with the preset stability threshold, calculates the difference between the dynamic correlation analysis results and the load fatigue benchmark value, determines the pavement area quality defect level, and transmits it to the state determination module. The status determination module receives the quality defect level of the road surface area, performs status grading assessment on the inspected road surface according to the damage level classification result, marks the area corresponding to the level identifier as the corresponding quality status category, and outputs the overall quality inspection result of the road surface.

[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, the expansion rate is calculated by extracting the displacement difference of the boundary coordinate points of road cracks and combining it with the time dimension, so that the expansion dynamics of cracks at different time periods can be quantified. The difference in the direction of crack evolution is identified by combining the degree of variation of the angle difference, so that the defect development law can be comprehensively described. The expansion rate and the consistency of direction are coupled for analysis to realize the dynamic characterization of crack stability. Furthermore, the environmental sensitivity coefficient and load fluctuation characteristics are introduced for linkage calculation, so that the correlation characteristics between external forces and crack expansion can be accurately presented. The grade classification is completed based on the difference of characteristic parameters and thresholds, so that the quality deterioration identification has dynamic and graded characteristics, thereby avoiding the distortion of results caused by single static monitoring. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2 This is a detailed schematic diagram of S1 of the present invention; Figure 3 This is a detailed schematic diagram of S2 of the present invention; Figure 4 This is a detailed schematic diagram of S3 of the present invention; Figure 5 This is a detailed schematic diagram of S4 of the present invention; Figure 6 This is a detailed schematic diagram of S5 of the present invention; Figure 7 This is a schematic diagram of the device module of the present invention. Detailed Implementation

[0018] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0019] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0020] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0021] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0022] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0023] Please see Figure 1 This invention provides an artificial intelligence-based road surface quality detection method, comprising the following steps: S1: Obtain surface image data of the road surface through image acquisition equipment, collect load data, process the road image data with edge detection algorithm, extract the set of crack boundary coordinate points, calculate the displacement difference of boundary coordinate points, and obtain the distribution state of road crack defects. S2: Calculate the unit time extension rate based on the displacement difference of coordinate points in the distribution state of pavement crack defects and the sampling time. Perform time-dimensional weighted accumulation processing on the extension rate, analyze the degree of variation of the angle difference in the extension direction, and identify the characteristic parameters of pavement defect evolution. S3: Call the time-weighted extension rate and directional consistency coefficient in the pavement defect evolution characteristic parameters for cross-coupling calculation, analyze the stability index of crack boundary evolution, calculate the environmental sensitivity coupling coefficient, perform dynamic correlation analysis based on load time-series fluctuation characteristics, and determine the pavement quality deterioration type characteristics; S4: Support vector machine classification is performed on the characteristics of pavement quality deterioration type. The stability index of crack boundary evolution is compared with the preset stability threshold. The difference between the dynamic correlation analysis results and the load fatigue benchmark value is calculated to determine the quality defect level of the pavement area. S5: Based on the quality defect level of the road surface area, the condition of the inspected road surface is evaluated, the area corresponding to the level is determined as the corresponding quality state, and the overall quality inspection result of the road surface is output. The distribution status of pavement cracks includes crack morphology, crack range, and crack spatial location. The pavement defect evolution characteristic parameters include the amplitude of extension rate change, directional angle variability, and time accumulation characteristics. The stability index of crack boundary evolution includes stability level, environmental sensitivity coefficient, and load coupling factor. The pavement area quality defect level includes mild, moderate, and severe defects. The overall pavement quality inspection results include grade assessment, regional status, and comprehensive conclusions.

[0024] Please see Figure 2 The specific steps of S1 are as follows: S101: Acquire surface image data of the road surface through image acquisition equipment, compare the gray-scale pixel gradient values ​​of the image data with the edge detection threshold, record the coordinates of pixels with gradient values ​​greater than the threshold, aggregate and sort the coordinates in sequence, and obtain the set of crack boundary coordinate points. In daytime inspections of asphalt pavements, the process begins with continuously capturing images of the road surface using a high-definition camera mounted on a vehicle. Each image has a pixel matrix resolution of 1920×1080. Within a single frame, the grayscale value of each pixel is extracted sequentially according to their arrangement, with the grayscale value ranged from 0 to 255. After row-by-row traversal, the brightness difference between adjacent pixels is calculated as the grayscale pixel gradient value. The horizontal gradient is obtained by subtracting the grayscale value of the pixel to the right from the current pixel's grayscale value, and the vertical gradient is obtained by subtracting the grayscale value of the pixel below the current pixel's grayscale value. These values ​​are recorded as absolute values. The larger of the horizontal and vertical gradient values ​​is taken as the final gradient value for that pixel. This gradient is then compared to a pre-set edge detection threshold T, which is calculated based on sampled images of a standard pavement without cracks before the inspection. The gradient values ​​of the pixels in the sample images are statistically analyzed, and the mean distribution is taken. with standard deviation In crack-free pavements, the gradient value of the crack edge response is higher than the mean. Therefore, in this embodiment, T = 11.8 + 2 × 4.5 = 20.8 is set, and the integer 21 is taken as the actual execution value. Each pixel is judged. If the pixel gradient value is greater than 21, the row and column coordinates of the current pixel are recorded. For example, in the test sample, the horizontal gradient of the (x=350, y=420) pixel is 30 and the vertical gradient is 18. Finally, the maximum value of 30 is taken, which is greater than the threshold of 21. Therefore, the coordinates of the pixel are recorded. The pixels in the image are processed in this way to obtain a set of coordinates that meet the conditions. Each coordinate is recorded in an ordered pair of (x, y) in the memory array. Then, the coordinates are aggregated and sorted in row priority according to the pixel scanning order so that the coordinates of the same crack are arranged continuously in the sequence. Finally, the obtained set of coordinates is recorded as the set of crack boundary coordinate points. For example, when 1972 pixels that meet the conditions are detected in the test road image, the pixels are arranged into a two-dimensional array in row and column order and used as input data for subsequent crack analysis.

[0025] Table 1: Crack Edge Detection Sampling Data Table

[0026] As shown in Table 1, during the detection process, each pixel undergoes operations such as calculating the horizontal and vertical gradients, taking the maximum value, comparing with the threshold, and classifying and recording, thereby forming a set of crack boundary coordinate points aggregated sequentially.

[0027] S102: Based on the set of coordinate points of the crack boundary, collect load data, calculate the horizontal and vertical difference values ​​of adjacent coordinate points, match the difference values ​​with the load data index and integrate them into a sequence, compare the sequence index by index and record the difference values ​​to generate the crack boundary displacement difference sequence. Based on the set of crack boundary coordinate points, this set is input to the synchronous recording unit of the road load sensing system. The system reads the load data sequence from embedded pressure sensors deployed within the road surface detection zone at the corresponding detection time. The load data unit is kN, and the sampling frequency is set to 100Hz. A one-to-one correspondence is established between the image acquisition timestamp and the load data recording timestamp, ensuring that each crack boundary point corresponds to a recently recorded load value. For example, in this embodiment, the first coordinate point (350, 420) corresponds to a load value of 42.6 kN, and the second coordinate point (352, 421) corresponds to a load value of 4... 1.8kN, and so on, matching the entire sequence; then, the horizontal difference Δx and vertical difference Δy are calculated between adjacent pixel coordinate pairs. The horizontal difference is the column coordinate of the next coordinate minus the column coordinate of the previous coordinate, and the vertical difference is the row coordinate of the next coordinate minus the row coordinate of the previous coordinate. The calculation results are integers, with positive values ​​representing positive displacement in the column or row direction, negative values ​​representing negative displacement, and absolute values ​​representing the pixel distance of displacement. For example, in the example above, the horizontal difference Δx between (350, 420) and (352, 421) is 352 - 350 = 2, and the vertical difference Δy is 421 - 4. 20=1, then using the index of the adjacent pixel pair as a reference, the load value difference of the corresponding index is ΔF=41.8-42.6=-0.8kN. This difference is paired with the horizontal and vertical difference values ​​and recorded to form a set of differential load association pairs. According to this execution method, the complete sequence is traversed to obtain a differential result set of length (n-1), where n is the total number of crack boundary coordinates. During the traversal, if the absolute value of the horizontal difference is greater than 3 pixels or the absolute value of the vertical difference is greater than 3 pixels, it is necessary to add a label when recording whether the load difference at this position exceeds 1.5 times the average load difference. In this embodiment... In the calculation, the average absolute value of the differential load is 0.65kN, and 1.5 times it is 0.975kN. Therefore, when the absolute value of a differential load is ≥0.975kN, it is marked as a "high difference point". For example, in the index sequence, the 58th pair of coordinate differences has Δx=4, Δy=2, and a load difference of 1.12kN, which is greater than 0.975kN. Therefore, it is marked as "high difference" in the record. Otherwise, it is marked as "normal difference". The index correspondence records of adjacent coordinate differences and load differences are completed in sequence, so that each record in the output crack boundary displacement difference sequence is accurately associated with the specific horizontal and vertical pixel difference and load change value.

[0028] S103: Call the crack boundary displacement difference sequence and combine it with the corresponding values ​​of the load data. Compare the displacement difference with the crack distribution benchmark value, perform clustering annotation on the difference index that deviates from the benchmark value, and aggregate it in the order of regions to obtain the distribution status of pavement crack defects. To retrieve the crack boundary displacement difference sequence, first, perform lateral difference analysis on each record. Longitudinal difference With the corresponding load difference Synchronous calls are made based on index positions, and the crack distribution benchmark value for that road section is imported from the road inspection database. The benchmark value is derived from statistical records of similar road surfaces under conditions of no significant damage. It is calculated by collecting 1000 sets of boundary difference sequences during low traffic flow periods, removing outliers, and then using the sum of the absolute value of the average load difference and the average transverse and longitudinal differences as the indicator. In this embodiment, the original statistical result is the absolute value of the average load difference. The sum of the average horizontal and vertical differences pixels, therefore the baseline value is set to During the traversal of the current detection sequence, the displacement difference index is calculated for each index item. for and will and To make a direct comparison, when When an index deviates from the baseline, it is recorded as an outlier, for example, in the differential data of index 58. , , Calculate index value If the value is greater than the baseline value of 3.58, it is marked as an outlier; using this method, each of the 1971 difference records in the sequence is compared to obtain an outlier index set, for example... Then, the clustering analysis process is invoked to group outliers according to their coordinate proximity in the original boundary sequence. The proximity criterion is that the boundary index difference between adjacent outliers is less than or equal to 5 and the Euclidean distance between their original pixel coordinates does not exceed 6 pixels. For example, index 122 corresponds to coordinates... Coordinates corresponding to index 125 The sequences differ by 3 and have an Euclidean distance of [value missing]. Pixels, meeting the proximity requirement, are clustered into the same region. The final clustering results are sorted by sequence position and stored sequentially in a data structure list, so that each cluster corresponds to a continuous anomaly segment. The output is a set of distribution states of pavement crack defects. For example, the current detection results form 4 continuous anomaly segments, with index ranges of... , , , .

[0029] Please see Figure 3 The specific steps of S2 are as follows: S201: Based on the displacement difference of coordinate points and sampling time in the distribution state of road crack defects, the displacement difference is compared with the adjacent sampling time interval, and the rate parameters are arranged in time order to obtain the unit time extension rate sequence. Based on the distribution of road surface crack defects, the displacement difference of coordinate points in each abnormal segment is selected, and the corresponding sampling time is extracted according to the time record order of its generation. The sampling time comes from the image frame recording timestamp, in seconds, with a resolution of 0.01 seconds. In this embodiment, the inspection vehicle travels at an average speed of 60 km / h, and the image sampling frequency is 50 Hz. Therefore, the average time interval of consecutive image frames is... =0.02, for each crack boundary coordinate point, record the displacement difference of the current point in index order. With the next point The absolute value of the difference is taken as the spatial displacement difference between the two points, for example, in an abnormal segment. Within, the displacement difference of index 58 is The displacement difference of pixel index 59 is Pixels, the displacement difference between the two is Pixel; then divide the displacement difference by the sampling time interval of 0.02 seconds to obtain the instantaneous stretching rate, specifically... The process continues by processing adjacent index points within the same anomalous segment, obtaining the extension rate value for each group of adjacent points within that anomalous segment. For rate parameters in pixels per second, conversion to the actual length unit is required during recording. In this embodiment, the ground resolution corresponding to the image resolution is 1 pixel = 2.5 mm. Therefore, the aforementioned rate value of 50 pixels per second is converted to... This processing method traverses the complete index of the four abnormal segments, for example, in the abnormal segment... In the image, the displacement difference at index 122 is 8 pixels, and at index 123 it is 10 pixels, with a difference of 2 pixels between them, corresponding to a certain rate. pixels per second, converted to The rate values ​​within each abnormal segment are arranged sequentially by index and then merged into the global unit time extended rate sequence. The length of the resulting rate sequence is the sum of the number of adjacent point pairs in all abnormal segments. A total of 47 valid extended rate records were obtained in this detection. The records are sorted in ascending order of time to form the final unit time extended rate sequence for use in subsequent steps.

[0030] S202: Based on the unit time extended rate sequence, the rate values ​​of adjacent time periods are weighted by the sampling interval, and the weighting factor is called to perform cumulative operation in continuous time periods to obtain the time dimension cumulative rate set; Based on the 47 time-stretched rate sequences arranged chronologically, each rate value within that sequence is called first. and their corresponding sampling intervals The sampling interval is fixed at 0.02s based on the image frame rate of the inspection vehicle. However, due to equipment buffering delay, the interval may vary between some adjacent abnormal segments. For example, the sampling interval between two sets of data in the actual detection record is 0.026s. Therefore, the actual interval value needs to be read for each set as a parameter for weight calculation. The weighting factor is set based on the standardized ratio of the sampling interval, and the standard sampling interval is taken as the baseline value. Weighting factor Set as Divide by For example, when the sampling interval is 0.02s, the corresponding weighting factor is... When the sampling interval is 0.026s, the corresponding weighting factor is: Then, the rates of adjacent time periods are multiplied by the weighting factor to obtain the weighted rate value. For example, the rate of the 5th group in the sequence is If the sampling interval is 0.02s, then the weighted rate value is s, the speed of the 6th group is The sampling interval is 0.026s, corresponding to a weighting rate of... Then, the weighted rate values ​​are continuously accumulated according to the time index. The accumulation method is to add each weighted rate value to the previous accumulated value sequentially from the beginning of the sequence. For example, when the first weighted rate is 142.5 mm / s, the accumulated value is 142.5. When the second weighted rate is 180.0 mm / s, the accumulated value is updated to... This sequential operation continues until all 47 sets of rate data have been processed. At the end of the calculation, the total cumulative value of this detection task is 8756.38 mm / s. The cumulative value of each step is stored in an array corresponding to the time index to form a time-dimensional cumulative rate set. The final cumulative rate set corresponds one-to-one with the abnormal segment index in the index dimension, serving as the data input source for subsequent direction vector comparison and evolution feature extraction.

[0031] S203: Call the time dimension cumulative rate set, calculate the angle difference between the crack displacement difference direction vector and the reference direction vector, and statistically summarize the fluctuation amplitude of the angle difference to obtain the pavement defect evolution characteristic parameters. The crack displacement difference direction vector, which matches the time-dimensional cumulative rate set with each time index, is obtained by transverse difference of the coordinates of adjacent crack boundary points. Longitudinal difference The direction is calculated using a ground coordinate system, with due east as the reference direction. The reference line is measured counterclockwise. The reference direction vector is extracted from the original inspection records and is the average angle of the main crack propagation direction during the stable period of this road section. In this embodiment, the reference direction is set as follows: For each direction vector currently detected, calculate its angular difference with the reference direction. The angular difference is processed as an absolute value, and the value falls within... For example, in the abnormal segment The corresponding direction vector, the coordinate difference of index 58 is , The direction angle can be obtained through Convert to The difference from the reference direction is The direction angle of index 59 is The difference is This process is repeated for all indices. After obtaining the full angular difference sequence, the fluctuation range is statistically analyzed. The fluctuation range is defined as the absolute value of the difference between the angular differences of adjacent time indices. For example, the fluctuation range between index 58 and index 59 is... When the fluctuation amplitude is ≥15.0°, it is marked as "severe fluctuation". In this embodiment, this threshold is based on the angular change distribution statistically analyzed from the original monitoring data, taking the mean of 9.8° plus 1.5 times the standard deviation (σ=3.47). We take 15.0° as the actual value; finally, we associate the angle difference and its fluctuation mark with the time accumulation rate set to form a set of road defect evolution characteristic parameters, and store it as structured data for subsequent use.

[0032] Table 2: Statistical Table of Differences Between Crack Direction Vector and Reference Direction

[0033] As shown in Table 2, the statistical results will , The angle is converted into a directional angle and compared with the reference direction to obtain the angle difference. The fluctuation type is then determined by the change in adjacent differences, thereby obtaining the crack evolution characteristic parameters in the time dimension.

[0034] Please see Figure 4 The specific steps of S3 are as follows: S301: Based on the time-weighted extension rate and directional consistency coefficient in the pavement defect evolution characteristic parameters, a vector inner product operation is performed on the two, and the difference between the slope of the inner product curve and the directional consistency coefficient is compared to generate a crack boundary evolution stability index. Based on the pavement defect evolution characteristic parameter set, the time-weighted propagation rate and directional consistency coefficient corresponding to each time index are matched item by item. In this embodiment, the time-weighted propagation rate data is derived from the time-dimensional cumulative rate set after time standardization. For example, at time index 58, the standardized rate value is 172.8 mm / s. The directional consistency coefficient represents the degree of consistency between the current crack propagation direction and the previous reference direction, and its value range is... Where 1 represents complete agreement and 0 represents complete opposition, the directional consistency coefficient for index 58 in this detection is 0.83. For each matching value, the vector inner product operation is performed by directly multiplying the standardized rate value by the directional consistency coefficient and recording the result. For example, the operation for index 58 is 172.8 × 0.83 = 143.424. This result is regarded as the inner product value at the corresponding time point. The inner product curve is formed by traversing all time index sequences in sequence. Then, slope analysis is performed on the curve. The slope is calculated by subtracting adjacent inner product values ​​and dividing by the corresponding time interval. For example, the inner product values ​​of index 58 and index 59 are 143.424 and 156.462 respectively, with a time interval of 0.02 seconds, and the slope value is... The slope value represents the rate of change of the inner product curve during that time period; simultaneously, the directional consistency coefficient offset difference of the corresponding index is called, which is the directional consistency coefficient of the current time index minus the baseline directional consistency coefficient. In this embodiment, the baseline directional consistency coefficient is set to the original average value of the road segment, 0.85. Therefore, the offset difference of index 58 is... The offset difference at index 59 is The slope value and offset difference of each time index are compared. When the absolute value of the slope is greater than 500 units / second and the absolute value of the offset difference is greater than 0.03, the evolution state of the time index is considered unstable. For example, the slope value of index 59 is 654.1, which is greater than 500 and the offset difference is 0.04, which is greater than 0.03. Therefore, it is marked as "unstable". Otherwise, it is marked as "stable". Finally, the stability judgment results of the entire time index sequence are summarized to generate a set of crack boundary evolution stability indexes.

[0035] Table 3: Example Table of Crack Evolution Stability Calculation

[0036] As shown in Table 3, the weighted extension rate corresponding to the time index and the directional consistency coefficient are multiplied to form an inner product value. Then, the difference between adjacent inner product values ​​is calculated and divided by the time interval to obtain the slope of the curve. The slope is then compared with the offset difference of the directional consistency coefficient to determine the stability of the time period, thus forming a set of stability indicators for crack boundary evolution.

[0037] S302: Call the crack boundary evolution stability index, calculate the index with temperature and humidity sensitive factors one by one, compare the results with the sensitivity benchmark value and perform weight integration to obtain the environmental sensitivity coupling coefficient. The stability index of crack boundary evolution and the stability judgment value of each record, along with the corresponding numerical index (i.e., the difference between the vector inner product value, the slope of the inner product curve, and the directional consistency coefficient), are called. Temperature-sensitive factors and humidity-sensitive factors are matched group by group according to the time index. In this embodiment, the temperature-sensitive factor... The data is collected in real time by meteorological sensors mounted on the testing vehicle, in units of... This represents the proportion of the crack change rate to the response of a unit temperature difference; humidity sensitivity factor. Similarly, the range of values ​​used to represent the effect of unit humidity change on crack evolution is as follows: At index 58, the measured value of the temperature sensitivity factor was 0.021, and the measured value of the humidity sensitivity factor was 0.34. For each time index, the numerical index of crack boundary evolution stability was directly multiplied by the temperature sensitivity factor to obtain the temperature effect. For example, the inner product value of index 58 was 143.424, multiplied by... The result is 3.0119. Using the same method, the numerical index of crack boundary evolution stability is multiplied by the humidity sensitivity factor to obtain the humidity effect. Index 58 corresponds to... The obtained temperature and humidity effects were compared with the sensitivity benchmark value, which was set as the average temperature effect monitored in this road section over the past year. The humidity sensitivity benchmark is set as the average humidity effect. When the temperature difference When the temperature exceeds the baseline value by 10% (threshold 0.265), it is recorded as "temperature deviation". The difference in humidity effect is also recorded. A deviation exceeding 8% of the humidity baseline value (threshold 3.664) is considered normal, therefore index 58 is marked as "temperature deviation, humidity normal". After completing the deviation determination for each item, the temperature effect and humidity effect of each record are integrated according to the set weight coefficients. In this embodiment, the temperature weight coefficient is 0.45 and the humidity weight coefficient is 0.55 (the weight coefficients are determined by referring to the relative contribution ratio of temperature and humidity changes to crack changes in the seasonal climate statistics of the region, ensuring that the sum of the two is 1). The integrated value is temperature effect × 0.45 + humidity effect × 0.55. For example, index 58 yields 3.0119 × 0.45 + 48.7642 × 0.55 = 1.3553 + 26.8203 = 28.1756, which is the environmental sensitivity coupling coefficient corresponding to this index. The same calculation is performed on all indices of the time series to form a set of environmental sensitivity coupling coefficients, which are used for subsequent mapping to load time series fluctuation characteristics.

[0038] S303: Based on the environmental sensitivity coupling coefficient, the amplitude range and period range of the load time-series fluctuation characteristics are mapped, and the dynamic offset amplitude corresponding to the interval is compared to obtain the characteristics of the road quality deterioration type. Based on the environmental sensitivity coupling coefficient, the load time-series fluctuation characteristic data corresponding one-to-one with its time index is called as input. The load time-series fluctuation characteristics are recorded by pressure sensors deployed within the road monitoring section at a sampling frequency of 100Hz, and undergo point-by-point filtering before analysis. In this embodiment, the fluctuation characteristics of each time index are characterized by two parameters: amplitude and period. The amplitude unit is kN, and the period unit is s. In order to perform mapping, the division criteria of amplitude interval and period range are first determined. The amplitude is taken into three intervals based on the statistical results of the road section under normal traffic conditions: the low amplitude interval is... The mid-range is greater than 5 and less than or equal to 12. The high range is greater than 12 and less than or equal to 25. The period range is divided into short-period intervals based on the original monitoring data. The medium-period range is greater than 1.2 and less than or equal to 2.0 s, and the long-period range is greater than 2.0 and less than or equal to 3.5 s. For each time index, the environmental sensitivity coupling coefficient is matched with the amplitude range and period range of the corresponding fluctuation characteristics. For example, the environmental sensitivity coupling coefficient of index 58 is 28.1756, its measured fluctuation amplitude is 8.6 kN (belonging to the medium-amplitude range), and its measured period is 1.35 s (within the medium-period range). In the mapping rules, the combination of medium amplitude and medium period sets the dynamic offset amplitude benchmark value. The value is 15. A high dynamic offset is defined as the difference between the actual environmental sensitivity coupling coefficient and the baseline value that exceeds 20% of the baseline value. For example, the difference at index 58 is... This value is greater than Therefore, it is recorded as high dynamic offset; similarly, the offset results are calculated for all indices, and the road quality deterioration type is classified according to the offset level mapped by the amplitude range and period range combined with the magnitude of the environmental sensitivity coupling coefficient. In this embodiment, the rule is set as follows: if the offset level is high and the coupling coefficient is ≥25, it is determined as "accelerated deterioration type", if the offset level is medium and the coupling coefficient is greater than or equal to 15 and less than 25, it is determined as "stable deterioration type", and if the offset level is low or the coupling coefficient is <15, it is determined as "slow deterioration type". Therefore, the calculation result of index 58 is high offset and coupling coefficient 28.1756≥25, and the type is determined as "accelerated deterioration type". Repeat the above operation to organize the deterioration type of the time index into a road quality deterioration type feature set as the output.

[0039] Please see Figure 5 The specific steps of S4 are as follows: S401: Based on the characteristics of pavement quality deterioration type, a feature matrix is ​​constructed and input into the classification processing unit. The crack morphology parameters are compared according to the boundary division of support vectors. The region is marked according to the category index to generate crack classification results. Based on the characteristics of pavement quality deterioration types, these types are read sequentially according to time index and spatial distribution. The type code of each feature entry is mapped to a numerical parameter; for example, slow deterioration is coded as 1, stable deterioration as 2, and accelerated deterioration as 3. Here, index 58 is of the accelerated deterioration type, so its type parameter is 3. The time index, type parameter, corresponding environmental sensitivity coupling coefficient, and load temporal fluctuation amplitude are combined into a single record. Each record is then summarized to form each row of the feature matrix, with the matrix columns representing [index, type parameter, coupling coefficient, fluctuation amplitude, fluctuation period]. The crack length change rate is calculated by dividing the crack length difference between adjacent sampling times by the time interval. For example, if the crack length records at index 58 and index 59 are 1243 mm and 1261 mm respectively, with a time difference of 0.02 s, then the change rate is (1261-1243) / 0.02 = 900 mm / s. After constructing the complete feature matrix, it is input into the classification processing unit. The classification processing unit has pre-trained support vector partitioning boundary parameters, which are stored in the form of optimal boundary vectors between multiple categories in a two-dimensional to multi-dimensional feature space. For example, when projecting the "coupling coefficient" and "crack length change rate" of this matrix in two dimensions, the separating hyperplane equation defined by the support vector boundary parameters will divide the sample points coded as 1, 2, and 3 according to the boundary vector w = [0.52, -0.31] and the bias value b = 2.1. When performing the alignment calculation, the inner product operation is performed on the feature vector and the boundary vector of each row of samples, and the bias b is added. The sign of the result is judged to determine which side of the boundary it is located on. For example, the two-dimensional vector with index 58 takes "coupling coefficient = 28.1756" and "crack length change rate = 900", and the inner product with w is 28. 1756×0.52+900×(-0.31)=14.6513-279.0=-264.3487, adding an offset of 2.1 gives -262.2487, which is less than 0. Therefore, it belongs to the negative side region of the dividing plane. This judgment is compared with the original accelerated degradation type category parameters. If the classification result is consistent with the original type, the original label is kept. If they are different, the classification result label is updated. The final classification label is bound to the original time index, and the resulting output data is a set of spatial regions with category index labels. This set can be directly stored and used as crack classification results.

[0040] Table 4: Sample Data Table of Road Crack Feature Matrix

[0041] As shown in Table 4, each row of data constitutes a record of the feature matrix. After the matrix is ​​input into the classification processing unit, it can be compared and the category confirmed based on the boundary division of support vectors, and finally the corresponding crack classification result is generated.

[0042] S402: Call the crack classification results, retrieve the boundary change sequence and extract the discrete point stability index, compare the index with the preset stability threshold, record the index with insufficient difference and encode it, and obtain the crack boundary stability set. The crack classification results are retrieved sequentially by time index, and the corresponding crack boundary change sequence is retrieved. This sequence consists of the two-dimensional planar coordinate positions of crack boundary points in consecutive sampling frames, and is recorded in frame order. In this embodiment, each of the three records from index 58 to index 60 contains coordinate data of at least 50 boundary points. For each boundary point, the difference between the coordinates of the current frame and the coordinates of the previous frame is first calculated to obtain the magnitude of the point displacement vector in pixels. Then, it is multiplied by the ground resolution coefficient (1 pixel = 2.5 mm in this example) to complete the physical conversion of the length. The difference between the maximum and minimum converted lengths of the points in the same frame is taken as the boundary dispersion amplitude of that frame. For example, the dispersion amplitude of index 58 is 14.50 mm, obtained by subtracting the minimum point displacement of 4.25 mm from the maximum point displacement of 18.75 mm. Then, this amplitude value is divided by the total number of points in the same frame to obtain the average dispersion. In this embodiment, the average dispersion of index 58 is... mm / point; Substitute the average dispersion of each time index into the stability index calculation step. The stability index is defined as 1 minus the ratio of this average dispersion to the original stable period average dispersion. The original stable period average dispersion is 0.24 mm / point, obtained from 500 samples of the same road segment during the normal phase. For example, the stability index of index 58 is... After calculation, the stability index of each index is subtracted from the preset stability threshold. The stability threshold is set to 0.15 based on the pre-deterioration warning line in the original monitoring. When the difference is less than or equal to zero, the stability is considered insufficient. For example, the difference for index 58 is... If the value is less than zero, the corresponding index is marked as unstable. The time indices are calculated and judged in turn, and the index numbers of unstable indices are recorded in the set in order. For example, in this embodiment, if the difference between index 58 and index 60 is less than zero, it is recorded in the set, and if the difference between index 59 is greater than 0.15, it is not recorded. Finally, the indices in the set are encoded. The encoding rule adopts the prefix "US" to represent UnstableState, followed by the three-digit index number. For example, index 58 is encoded as US058 and index 60 is encoded as US060. The sequence of the encoding results is the output crack boundary stability set.

[0043] S403: Based on the crack boundary stability set, call the dynamic correlation analysis value corresponding to the set index, perform difference calculation with the load fatigue benchmark value, map the difference to the defect level classification table, and obtain the quality defect level of the pavement area. Based on the crack boundary stability set, the dynamic correlation analysis values ​​corresponding to the coded indices within the set are sequentially called. In this embodiment, these values ​​are calculated from the correlation coefficient between the load time-series fluctuation data and the crack boundary change rate, and their range is [value range missing]. Furthermore, it is dimensionless; for example, US058 and US060 in the set correspond to dynamic correlation analysis values ​​of 0.72 and 0.65, respectively. The load fatigue benchmark value is invoked, which is determined by the structural mechanical fatigue test results of similar road sections within their normal service life. In this embodiment, it is set to 0.60. When calculating the difference, the load fatigue benchmark value is subtracted from the dynamic correlation analysis value for each index to obtain the offset. For example, the offset for US058 is... The offset of US060 is The offsets are then mapped to intervals based on their absolute values, according to a defect level classification table. The levels are divided into four categories: Level I (offset greater than or equal to 0 and less than 0.05), Level II (offset greater than or equal to 0.05 and less than 0.1), Level III (offset greater than or equal to 0.1 and less than 0.2), and Level IV (offset ≥ 0.20). For example, US058's offset of 0.12 corresponds to Level III, and US060's offset of 0.05 is at the lower limit of Level II, therefore it is Level II. After mapping all indices within the set, the indices are aligned with their corresponding levels, for example, US058 → Level III, US060 → Level II. Paired records are then rearranged in ascending order by time index, resulting in structured data representing the road surface area quality defect levels. This set retains both spatial index numbers and clear defect level identifiers, facilitating comparative analysis of quality levels across different time periods or spatial locations.

[0044] Please see Figure 6 The specific steps of S5 are as follows: S501: Based on the quality defect level of the road surface area, the pixel gray value in the surface image matrix of the detection area is compared with the defect level judgment threshold. When the gray value is lower than the threshold, the corresponding area is marked as a low-quality block, and the marked block indexes are aggregated to obtain the defect area index set. Based on the road surface defect level, the surface image matrix file corresponding to each area is read in the order of the area index. This matrix is ​​acquired by the road inspection imaging device at a resolution of 0.5 mm / pixel and stored in grayscale. The pixel grayscale value range is set from 0 to 255, where 0 represents the darkest and the lowest reflectivity of the road surface, and 255 represents the highest brightness and the highest reflectivity of the surface. The grayscale judgment threshold corresponding to the defect level of the area is called. In this embodiment, the grayscale judgment threshold is set according to the level classification rules. The threshold for level IV defects is set to 135, level III to 150, level II to 165, and level I to 180. This value is derived from the median of the grayscale distribution of different defect categories in the original image minus a standard deviation range. For example, the median of level III is statistically 158.4, and the standard deviation is 8.1, so the threshold is approximately 150. Then, the pixel values ​​in the image matrix are traversed and compared with the set corresponding defect level threshold one by one. When the pixel grayscale value is less than the corresponding level threshold, the threshold is determined. When the threshold of the quality level is reached, the coordinate region where the pixel is located is immediately marked as a low-quality pixel area. Region clustering operation is performed on consecutive low-quality pixels to form low-quality block labels. In this embodiment, the gray-scale matrix element number of the region corresponding to index RG102 is 256×256, with a total of 65536 pixels. Among them, the number of pixels below the threshold is 17432. After 8-neighbor clustering, 15 low-quality blocks are formed. Then, the region identifier of each low-quality block is merged with the road surface region index to form a low-quality block index. Each block index uses the prefix "BQ" to represent BlockofQuality-Low and is appended with a four-digit sequence number. For example, the third low-quality block of RG102 is coded as BQRG102-0003. The low-quality block indexes of this region are summarized into a list to form the low-quality block index set of this region. The low-quality block index sets of the detected regions are merged to obtain the defect region index set as the output result.

[0045] Table 5: Pixel Statistics of Defective Areas and Sample Results of Low-Quality Block Detection

[0046] As shown in Table 5, the defect level of each detection area corresponds to a grayscale threshold. After pixel-by-pixel comparison, the number of pixels below the threshold is counted and block clustering is performed. Finally, the number of low-quality blocks is obtained and encoded to form a defect area index set.

[0047] S502: Call the defect area index set, extract structural integrity parameters for the corresponding pavement area data, compare them with the quality grading benchmark value, determine the area level based on the comparison result, and generate a pavement area quality level sequence. The defect area index set is invoked, and the original detection data file of the pavement area corresponding to each defect area index is invoked sequentially. This file contains acceleration response sequences, deflection value records, and measured surface layer thickness data. In this embodiment, the structural integrity parameter is measured by the ratio of deflection value to surface layer thickness, and the unit of this ratio is 1. The smaller the value, the stronger the structure. The specific calculation steps are to first obtain the average deflection value and average surface thickness of the area. For example, the measured average deflection value of area RG102 is 1.85mm, and the average thickness is 52.4mm. Then the structural integrity parameter is... Then, this parameter is compared with the quality grading benchmark value. In this embodiment, the benchmark value set is calculated based on 200 regional data accumulated during the normal service phase of the road section, and is divided into four levels: Level I benchmark value range is greater than or equal to 0 and less than 0.025, Level II range is greater than or equal to 0.025 and less than 0.035, Level III range is greater than or equal to 0.035 and less than 0.045, and Level IV range is ≥0.045. If the parameter value is within a certain range, the level corresponding to that range is directly used as the current regional level, for example, RG102. Areas located at the lower limit of the Level III range are retained as Level III according to the rules because they are close to the lower boundary; areas in the defect area index set are processed sequentially, for example, RG108 has a deflection value of 1.42mm and a thickness of 58.3mm. Located in the Level I interval, it is therefore classified as Level I. The index of each region is paired with the classification level and recorded to form a road surface area quality level sequence, and the output is guaranteed to be sorted by index.

[0048] S503: Based on the road surface area quality grade sequence, different grade areas are classified and summarized, and weighted calculations are performed according to their proportions to generate the overall road surface quality test results; Based on the road surface area quality grade sequence, all areas are classified into four grade groups: I, II, III, and IV. The number and proportion of areas in each grade group are then calculated. For example, in a sequence of 40 areas, 18 are grade I, accounting for [percentage missing]. There are 12 Level II cases, accounting for 0.30%; 7 Level III cases, accounting for 0.175%; and 3 Level IV cases, accounting for 0.075%. The percentage of each level is multiplied by its corresponding weighting coefficient and summed. In this embodiment, the weighting coefficients are set as follows: Level I 1.00, Level II 0.75, Level III 0.50, and Level IV 0.25. These coefficients are set with reference to the relative quality index of different levels in structural safety assessment. The overall quality score is obtained through calculation. For example, in this case, the overall score = The scores are mapped to a preset overall quality level range table. In this embodiment, ≥0.85 is defined as "Excellent", ≥0.7 and <0.85 is defined as "Good", ≥0.5 and <0.7 is defined as "Average", and <0.50 is defined as "Poor". The score 0.78125 falls within the range of ≥0.7 and <0.85, so the overall road surface quality test result is determined to be "Good". This result is then bound to the corresponding date and road section information for output, forming the final overall road surface quality test result data.

[0049] Please see Figure 7 An artificial intelligence-based road surface quality detection device includes: The image acquisition module acquires surface image data and load data through image acquisition equipment, processes the surface image data using edge detection algorithm, extracts the set of crack boundary coordinate points, calculates the displacement difference of boundary coordinate points, obtains the distribution state of road crack defects, and transmits it to the crack extraction module. The crack extraction module calculates the unit time extension rate based on the displacement difference of coordinate points in the distribution state of pavement crack defects and the sampling time. It performs time-weighted accumulation processing on the extension rate, analyzes the degree of variation of the angle difference in the extension direction, obtains the pavement defect evolution characteristic parameters, and transmits them to the evolution analysis module. The evolution analysis module calls the time-weighted extension rate and directional consistency coefficient in the pavement defect evolution characteristic parameters for cross-coupling calculation, analyzes the stability index of crack boundary evolution, calculates the environmental sensitivity coupling coefficient, performs dynamic correlation analysis based on load time-series fluctuation characteristics, determines the pavement quality deterioration type characteristics, and transmits them to the quality assessment module. The quality assessment module calls the characteristics of pavement quality deterioration type for support vector machine classification, compares the stability index of crack boundary evolution with the preset stability threshold, calculates the difference between the dynamic correlation analysis results and the load fatigue benchmark value, determines the quality defect level of the pavement area, and transmits it to the state determination module. The status determination module receives the quality defect level of the road surface area, performs status grading assessment on the inspected road surface based on the damage level classification results, marks the area corresponding to the level identifier as the corresponding quality status category, and outputs the overall road surface quality inspection results.

[0050] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A road surface quality detection method based on artificial intelligence, characterized in that, Includes the following steps: S1: Obtain surface image data of the road surface through image acquisition equipment, collect load data, process the road image data with edge detection algorithm, extract the set of crack boundary coordinate points, calculate the displacement difference of boundary coordinate points, and obtain the distribution state of road crack defects. S2: Calculate the unit time extension rate based on the displacement difference of coordinate points and sampling time in the distribution state of the road crack defects, perform time-dimensional weighted accumulation processing on the extension rate, analyze the degree of variation of the angle difference in the extension direction, and identify the characteristic parameters of the road defect evolution. S3: Call the time-weighted extension rate and directional consistency coefficient in the pavement defect evolution characteristic parameters for cross-coupling calculation, analyze the stability index of crack boundary evolution, calculate the environmental sensitivity coupling coefficient, perform dynamic correlation analysis based on load time-series fluctuation characteristics, and determine the pavement quality deterioration type characteristics; S4: Perform support vector machine classification processing on the characteristics of the road quality deterioration type, compare the stability index of crack boundary evolution with the preset stability threshold, calculate the difference between the dynamic correlation analysis result and the load fatigue benchmark value, and determine the quality defect level of the road area.

2. The road surface quality detection method based on artificial intelligence according to claim 1, characterized in that, The distribution status of pavement crack defects includes crack morphology, crack range, and crack spatial location. The pavement defect evolution characteristic parameters include the amplitude of extension rate change, directional angle variability, and time accumulation characteristics. The crack boundary evolution stability index includes stability level, environmental sensitivity coefficient, and load coupling factor. The pavement area quality defect level includes mild defects, moderate defects, and severe defects.

3. The road surface quality detection method based on artificial intelligence according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: Acquire surface image data of the road surface through image acquisition equipment, compare the gray-scale pixel gradient values ​​of the image data with the edge detection threshold, record the coordinates of pixels with gradient values ​​greater than the threshold, aggregate and sort the coordinates in sequence, and obtain the set of crack boundary coordinate points. S102: Based on the set of coordinate points of the crack boundary, collect load data, calculate the horizontal and vertical difference values ​​of adjacent coordinate points, match the difference values ​​with the load data index and integrate them into a sequence, compare the sequence index by index and record the difference values ​​to generate a crack boundary displacement difference sequence. S103: Call the crack boundary displacement difference sequence and combine it with the corresponding values ​​of the load data. Compare the displacement difference with the crack distribution benchmark value, perform clustering annotation on the difference index that deviates from the benchmark value, and aggregate it in the order of regions to obtain the distribution status of road crack defects.

4. The road surface quality detection method based on artificial intelligence according to claim 3, characterized in that, The specific steps of S2 are as follows: S201: Based on the displacement difference of coordinate points and sampling time in the distribution state of road crack defects, the displacement difference is compared with the adjacent sampling time interval, and the rate parameters are arranged in time order to obtain the unit time extension rate sequence. S202: Based on the unit time extension rate sequence, the rate values ​​of adjacent time periods are weighted by the sampling interval, and the weighting factor is called to perform cumulative operation in continuous time periods to obtain a time dimension cumulative rate set; S203: Call the accumulated rate set of the time dimension, calculate the angle difference between the crack displacement difference direction vector and the reference direction vector, and statistically summarize the fluctuation amplitude of the angle difference to obtain the pavement defect evolution characteristic parameters.

5. The road surface quality detection method based on artificial intelligence according to claim 4, characterized in that, The specific steps for S3 are as follows: S301: Based on the time-weighted extension rate and directional consistency coefficient in the pavement defect evolution characteristic parameters, perform a vector inner product operation on the two, and compare the difference between the slope of the inner product curve and the directional consistency coefficient to generate a crack boundary evolution stability index. S302: Call the crack boundary evolution stability index, calculate the index with the temperature sensitivity factor and humidity sensitivity factor one by one, compare the results with the sensitivity benchmark value and perform weight integration to obtain the environmental sensitivity coupling coefficient. S303: Based on the environmental sensitivity coupling coefficient, the amplitude range and period range of the load time-series fluctuation characteristics are mapped, and the dynamic offset amplitude corresponding to the interval is compared to obtain the characteristics of the road quality deterioration type.

6. The road surface quality detection method based on artificial intelligence according to claim 5, characterized in that, The specific steps of S4 are as follows: S401: Based on the characteristics of the road quality deterioration type, construct a feature matrix and input it into the classification processing unit. Compare the crack morphology parameters according to the support vector boundary division, mark the region according to the category index, and generate crack classification results. S402: Call the crack classification results, retrieve the boundary change sequence and extract the discrete point stability index, compare the index with the preset stability threshold, record the index with insufficient difference and encode it to obtain the crack boundary stability set. S403: Based on the set of crack boundary stability, call the dynamic correlation analysis value corresponding to the set index, perform difference calculation with the load fatigue benchmark value, map the difference to the defect level classification table, and obtain the quality defect level of the pavement area.

7. The road surface quality detection method based on artificial intelligence according to claim 6, characterized in that, The preset stability threshold is the lower limit of the allowable value of the stability index of discrete points at the crack boundary. The load fatigue benchmark value is the reference value of the fatigue load that the pavement structure can withstand under design conditions.

8. The road surface quality detection method based on artificial intelligence according to claim 1, characterized in that, The method also includes step S5: S5: Based on the quality defect level of the road surface area, evaluate the condition of the inspected road surface, determine the area corresponding to the level as the corresponding quality state, and output the overall quality inspection result of the road surface; The overall road surface quality inspection results include grade assessment, regional condition, and comprehensive conclusions.

9. The road surface quality detection method based on artificial intelligence according to claim 8, characterized in that, The specific steps of S5 are as follows: S501: Based on the quality defect level of the road surface area, the pixel gray value in the surface image matrix of the detection area is compared with the defect level judgment threshold. When the gray value is lower than the threshold, the corresponding area is marked as a low-quality block, and the marked block indexes are aggregated to obtain the defect area index set. S502: Call the defect area index set, extract structural integrity parameters for the corresponding road surface area data, compare them with the quality grading benchmark value, determine the area level based on the comparison result, and generate a road surface area quality level sequence. S503: Based on the road surface area quality grade sequence, classify and summarize the different grade areas, perform weighted calculations according to their proportions, and generate the overall road surface quality test results.

10. A road surface quality detection device based on artificial intelligence, characterized in that, The apparatus is used to implement the artificial intelligence-based road surface quality detection method according to any one of claims 1-9, and the apparatus comprises: The image acquisition module acquires surface image data and load data through image acquisition equipment, processes the surface image data using edge detection algorithm, extracts the set of crack boundary coordinate points, calculates the displacement difference of boundary coordinate points, obtains the distribution state of road crack defects, and transmits it to the crack extraction module. The crack extraction module calculates the unit time extension rate based on the displacement difference of coordinate points in the distribution state of the road crack defects and the sampling time. It performs time-dimensional weighted accumulation processing on the extension rate, analyzes the degree of variation of the angle difference in the extension direction, obtains the road defect evolution characteristic parameters, and transmits them to the evolution analysis module. The evolution analysis module calls the time-weighted extension rate and directional consistency coefficient in the pavement defect evolution characteristic parameters for cross-coupling calculation, analyzes the stability index of crack boundary evolution, calculates the environmental sensitivity coupling coefficient, performs dynamic correlation analysis based on load time-series fluctuation characteristics, determines the pavement quality deterioration type characteristics, and transmits them to the quality assessment module. The quality assessment module calls the pavement quality deterioration type features to perform support vector machine classification, compares the crack boundary evolution stability index with the preset stability threshold, calculates the difference between the dynamic correlation analysis results and the load fatigue benchmark value, determines the pavement area quality defect level, and transmits it to the state determination module. The status determination module receives the quality defect level of the road surface area, performs status grading assessment on the inspected road surface according to the damage level classification result, marks the area corresponding to the level identifier as the corresponding quality status category, and outputs the overall quality inspection result of the road surface.

Images