Artificial intelligence-based asphalt pavement multi-type disease automatic identification method

Abstract

The present application relates to the technical field of pavement disease type identification, in particular to an asphalt pavement multi-type disease automatic identification method based on artificial intelligence, the present application carries out infrared thermal imaging image collection on the target detection section, sorts the collected images according to the preset mark to construct a continuous image sequence, completes suspected disease area identification and extraction frame by frame, forms a suspected disease meta-image set, then carries out multi-dimensional feature analysis on the images in the set, obtains geometric feature parameters and temperature feature parameters, and then fuses the two types of feature parameters to calculate the disease state index, according to the disease state index, the safe disease meta-image set and the dangerous disease meta-image set are screened, the dangerous disease meta-image is diagnosed for disease type, and the matched diagnosis and treatment scheme is generated as needed, the disease grading control and maintenance scheme intelligent matching are realized, and the efficient and refined operation and detection requirements of the asphalt pavement are met.

Description

Technical Field

[0001] This invention relates to the field of pavement distress type identification technology, specifically to an automatic identification method for multiple types of asphalt pavement distress based on artificial intelligence. Background Technology

[0002] Asphalt pavement is a core component of highway transportation infrastructure. Its service condition directly determines traffic safety, driving comfort, and road lifespan. Accurate identification and efficient treatment of pavement defects are crucial for maintaining good pavement performance and reducing maintenance risks. Currently, traditional asphalt pavement defect identification technologies have multiple shortcomings and are difficult to adapt to the needs of modern road maintenance. The core pain points are concentrated in the following aspects: Firstly, most existing identification technologies rely on the grayscale difference of visible light images to identify apparent defects, which is easily affected by external light conditions and has extremely poor identification stability. In strong light environments, crack areas are prone to reflections that obscure the characteristics of defects. In low light or cloudy environments, the grayscale boundary between defects and normal road surfaces is blurred, which can easily lead to misjudgment and missed detection, and cannot guarantee the accuracy of identification. Secondly, traditional optical imaging can only detect surface defects on the road surface and cannot penetrate the surface layer to detect hidden defects such as loose base layers and internal cavities. These hidden defects will continue to damage the road's load-bearing structure and significantly shorten the road's lifespan, but they are difficult to detect in advance by conventional technologies. In addition, conventional detection mainly relies on manual inspection and fixed-point sampling, which is inefficient and has large subjective errors. It cannot achieve continuous data collection across long road sections, lacks adaptive judgment and defect classification logic, consumes a lot of computing power, cannot distinguish between minor flaws and high-risk defects, and cannot be combined with maintenance schedules to coordinate treatment plans. The overall level of intelligence and precision is seriously insufficient. There is an urgent need for a set of efficient and accurate intelligent recognition technologies to fill the gap in the industry.

[0003] To address the aforementioned shortcomings, a technical solution is provided. Summary of the Invention

[0004] To address the aforementioned shortcomings of existing technologies, this invention provides an artificial intelligence-based automatic identification method for multiple types of asphalt pavement defects. This method effectively solves the problems in existing technologies, such as pavement defect identification being easily affected by light interference, inability to detect hidden internal defects, low detection efficiency, and difficulty in coordinating maintenance schedules and treatment plans.

[0005] To achieve the above objectives, the present invention can be implemented through the following technical solutions: This invention provides an artificial intelligence-based method for automatic identification of multiple types of defects in asphalt pavement, comprising the following steps: A preset image acquisition time window is used to acquire infrared thermal images of the target detection section. The infrared thermal imaging images are sorted according to the preset labels to form an ordered sequence of infrared thermal imaging images. Then, the suspected disease areas are identified in each frame of the infrared thermal imaging image sequence in turn. The identified suspected disease areas are then extracted in turn to form a set of suspected disease meta-images. Multidimensional feature recognition and analysis were performed on each suspected disease element image in the suspected disease element image set to determine the geometric and temperature feature parameters of the corresponding suspected disease element images. Geometric feature parameters include actual physical area, aspect ratio, and circularity; Temperature characteristic parameters include temperature mean, temperature range, and temperature skewness; By combining geometric and temperature feature parameters, the disease status index of each suspected disease element image is determined. Based on the disease status index, the suspected disease element images are comprehensively evaluated, thereby dividing the suspected disease element images into a safe disease element image set and a dangerous disease element image set. Disease type diagnosis is performed on the meta-images in the dangerous disease meta-image set. Based on the diagnosis results, it is determined whether to generate a diagnosis and treatment plan. If the diagnosis and treatment plan is generated, the processing time of the diagnosis and treatment plan is determined.

[0006] Furthermore, the solution steps for identifying suspected diseased areas are as follows: For each frame of an infrared thermal imaging image sequence, a uniform grid is divided according to a preset step size. The geometric center of each grid is used as a sampling point, and detection points are set up to construct a set of detection points for each frame of the image. Simultaneously, the temperature values ​​of all detection points in the detection point set are extracted and the average value is calculated to obtain the global reference temperature of each frame of the image; For each frame of image, after obtaining the global reference temperature, the temperature deviation between the temperature value of each detection point and the global reference temperature is calculated. Then, taking each detection point as the center, several neighboring detection points within its neighborhood are selected to form a local detection area. The local standard deviation of the temperature values ​​of all detection points in the local detection area is calculated, and the local standard deviation is used as a dynamic correction factor to construct an adaptive dynamic threshold. If the absolute value of the temperature deviation at a certain detection point is greater than the adaptive dynamic threshold, the temperature change at that detection point is determined to be significantly different from that of the surrounding normal area, and the detection point is marked as an abnormal detection point. The DBSCAN clustering algorithm is used to perform spatial clustering of anomaly detection points, and areas of anomaly detection points that are continuously clustered in space are identified as suspected disease areas.

[0007] Furthermore, the steps for solving the actual physical area in the geometric characteristic parameters are as follows: The images of each suspected disease element are preprocessed. Then, the total number of pixels in each suspected disease element image is counted. Then, the calibration parameters of the image acquisition system are combined, namely the pixel resolution. The pixel resolution is defined as the actual physical length corresponding to a single pixel. Based on pixel resolution, the total number of pixels in each suspected lesion image is converted into the actual physical area. The conversion logic is as follows: each pixel corresponds to an area of... The actual physical area of ​​each suspected lesion pixel image is equal to the sum of the areas of the tiny physical units corresponding to all pixels.

[0008] Furthermore, the steps for solving the aspect ratio and roundness among the geometric feature parameters are as follows: Extract the boundary contour point set within each suspected lesion element image. Then, calculate the minimum bounding rectangle of the contour based on the equivalent algorithm. Its length corresponds to the maximum physical span of the suspected lesion element image in the principal axis direction, and its width corresponds to the minimum physical span of the suspected lesion element image in the vertical principal axis direction. Based on the length-to-width ratio of the minimum bounding rectangle of each suspected lesion element image, the aspect ratio of each suspected lesion element image is obtained. The circularity is calculated by extracting the perimeter of the boundary contour and the actual physical area of ​​each suspected lesion element image.

[0009] Furthermore, the steps for solving the temperature characteristic parameters—mean temperature, temperature range temperature, and temperature skewness—are as follows: Temperature sample set is formed by extracting the temperature values ​​corresponding to all pixels in each suspected disease element image, and temperature feature parameters are calculated based on the temperature sample set. In this process, all temperature values ​​in the temperature sample set are summed and then divided by the total number of pixels corresponding to the suspected disease element images to obtain the arithmetic mean, which is used as the temperature mean. Traverse the temperature sample set, find the maximum and minimum temperature values ​​respectively, calculate the difference between the two, and obtain the temperature range; Temperature skewness is obtained by the ratio of the third central moment to the cube of the standard deviation.

[0010] Furthermore, the steps for calculating the disease state index are as follows: The geometric and temperature feature parameters of each suspected disease element image are linearly normalized, and the disease status index of each suspected disease element image is calculated accordingly. The specific calculation formula is as follows:

[0011] in, , , , , and These are the corresponding weight coefficients. , , , , and These are the dimensionless results after linear normalization of the actual physical area, aspect ratio, roundness, mean temperature, temperature range, and temperature skewness, respectively.

[0012] Furthermore, the steps for dividing the suspected disease pixel images into a safe disease pixel image set and a dangerous disease pixel image set are as follows: The disease status index of each suspected disease element image. Compared with the preset disease risk threshold Compare; When the disease status index ≥Preset disease risk threshold At that time, the disease status index will be... The corresponding suspected disease element images are recorded as dangerous disease element images, and a dangerous disease element image set is constructed from them; When the disease status index <Preset disease risk threshold> At that time, the disease status index will be... The corresponding suspected disease element images are recorded as safe disease element images, and a safe disease element image set is constructed from them.

[0013] Furthermore, when generating the diagnostic treatment plan, the following steps are performed: The disease-related area corresponding to the current matching diagnosis and treatment plan is recorded as the target disease area. The disease treatment operation progress and the preset overall operation end time corresponding to the target disease area are retrieved. The theoretical processing completion time of the disease repair operation is generated by combining the extracted processing time with the current system benchmark time. It is determined whether the processing completion time is less than the operation end time. If the determination result is less than, the empty interval time between the processing completion time and the operation end time is calculated and recorded as the remaining buffer time. The system collects the actual progress of the current maintenance task and the corresponding time consumed. Based on the progress matching and calculation logic, it obtains the work progress and the remaining incremental progress. The work progress and the remaining incremental progress are accumulated to obtain the overall progress. The overall progress is compared and verified with the preset standard completion progress threshold. If it is determined to meet the standard, it proves that after the implementation of the diagnostic and treatment plan, it will not delay the completion of the original basic maintenance tasks in the target disease area on schedule.

[0014] The technical solution provided by this invention has the following advantages compared with the known prior art: 1. This invention uses infrared thermal imaging technology to acquire road surface images, completely eliminating dependence on lighting conditions. It avoids the problems of strong light reflections masking defects and blurred grayscale boundaries in low light leading to misjudgments and missed detections. By constructing an adaptive dynamic threshold combining global reference temperature and local standard deviation, it filters abnormal detection points and uses the DBSCAN spatial clustering algorithm to accurately locate suspected defect areas, reducing interference from light fluctuations. At the same time, relying on the differences in heat conduction and heat radiation in different road surface areas, it penetrates the surface to identify hidden defects such as loose base layers and internal cavities. It breaks through the limitations of traditional optical imaging, provides early warning of hidden dangers, protects the load-bearing structure, and extends the life of the road surface. Thus, it achieves the dual goals of accurate identification of road surface defects and early prevention and control of deep-seated hidden dangers under complex lighting conditions.

[0015] 2. This invention integrates two core parameters—geometric and temperature characteristics—and constructs a quantitative disease status index through linear normalization and weighting coefficients. Based on a preset danger threshold, it achieves precise differentiation between safe and dangerous diseases. This technical solution overcomes the shortcomings of conventional techniques that cannot distinguish between minor surface defects and structurally high-risk diseases, enabling a quantitative assessment of disease severity. This makes disease judgment logic more scientific and rigorous, providing accurate and reliable data support for subsequent disease treatment.

[0016] 3. This invention accurately diagnoses the type of asphalt pavement defects based on the image of the defect and matches a customized treatment plan, thereby improving the intelligence, precision and scientific level of pavement operation and maintenance, reducing safety risks and operation and maintenance costs, and thus realizing the coordinated advancement of efficient treatment of asphalt pavement defects and intelligent operation and maintenance throughout the entire process. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the 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.

[0018] Figure 1 This is a flowchart illustrating the overall process of the present invention. Detailed Implementation

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

[0020] like Figure 1 As shown, the artificial intelligence-based automatic identification method for multiple types of asphalt pavement defects includes the following steps: Step 1: Set a preset image acquisition time window and acquire infrared thermal images of the target detection section. The specific acquisition process is as follows: A mobile robot equipped with an infrared thermal imager is used to perform road surface inspection. The infrared thermal imager has a resolution of 640×50×512 pixels, a temperature measurement range of -20℃ to 150℃, and a temperature measurement accuracy of ±0.5℃. The mobile robot is controlled to automatically travel to the target inspection section according to a preset path and perform a full-area scan of the road surface to collect infrared thermal images of the road surface. The mobile robot's travel speed is set to a constant speed of 0.5m / s. The infrared thermal imager continuously collects images of the road surface at a sampling frequency of 30 frames / second, and the effective coverage width of the road surface is not less than 3 meters, which can be adjusted according to the lens angle. The collected infrared thermal images are uploaded to a cloud image analysis terminal in real time via a 5G wireless communication module.

[0021] Step Two: After receiving the infrared thermal imaging images, the cloud-based image analysis terminal indexes the images according to their acquisition time and corresponding location information. Each frame of the infrared thermal imaging image is then marked with a preset identifier, which includes the acquisition time and location information. The images are then sorted according to the preset identifiers to form an ordered sequence. Suspected disease areas are then identified sequentially in each frame of the sequence. The specific identification process is as follows: For each frame of an infrared thermal imaging image sequence, a uniform grid is divided according to a preset step size (e.g., 10×10 pixels). The geometric center of each grid is used as a sampling point, and detection points are set up to construct a set of detection points for each frame of the image. Simultaneously, the temperature values ​​of all detection points in the detection point set are extracted and the average value is calculated to obtain the global reference temperature of each frame of the image; To overcome the limitations of a single threshold, this scheme adopts adaptive local deviation analysis to achieve accurate determination of temperature anomalies. Specifically, for each frame of image, after obtaining the global reference temperature, the temperature deviation between the temperature value of each detection point and the global reference temperature is calculated. Then, taking each detection point as the center, several neighboring detection points within its neighborhood are selected to form a local detection area. The local standard deviation of the temperature values ​​of all detection points in the local detection area is calculated, and the local standard deviation is used as a dynamic correction factor to construct an adaptive dynamic threshold. If the absolute value of the temperature deviation at a certain detection point is greater than the adaptive dynamic threshold, the temperature change at that detection point is determined to be significantly different from that of the surrounding normal area, and the detection point is marked as an abnormal detection point; otherwise, it is determined to be a normal detection point. The DBSCAN clustering algorithm is used to perform spatial clustering of anomaly detection points, and spatially clustered areas of anomaly detection points are identified as suspected disease areas. It should be noted that the DBSCAN clustering algorithm is used to perform spatial clustering of anomaly detection points as follows: the image coordinates corresponding to each anomaly detection point are used as the clustering basis, the neighborhood radius and the minimum number of points are preset as clustering parameters, all anomaly detection points are traversed, and anomaly detection points located in the same neighborhood and meeting the minimum number of points condition are grouped into the same cluster. Isolated single-point noise is removed, and the image area corresponding to the continuous cluster is determined as the suspected disease area. All identified suspected disease areas are extracted sequentially to form a suspected disease metadata set, which is then transmitted to the suspected disease metadata analysis terminal.

[0022] Step 3: After receiving the set of suspected disease element images, the suspected disease element image analysis terminal performs multi-dimensional feature recognition and analysis on each suspected disease element image in the set to determine the geometric feature parameters and temperature feature parameters of the corresponding suspected disease element image. The specific analysis process is as follows: The images of each suspected disease element are preprocessed as follows: grayscale conversion, binarization segmentation, and morphological denoising optimization are performed on each suspected disease element image in sequence to remove irrelevant interference information. The image is decomposed into independent and identifiable pixel units by enlarging the display of the labeled pixel grid, and finally, a suspected disease element image that can clearly display each pixel is generated. The total number of pixels in each suspected disease element image is counted and denoted as . Furthermore, considering the calibration parameters of the image acquisition system, namely pixel resolution, which is defined as the actual physical length corresponding to a single pixel, denoted as... (Unit: meters / pixel) The pixel resolution calibration method is as follows: Under the same shooting distance and lens parameters, acquire an image of a calibration object with known geometric dimensions. The pixel resolution can be obtained by calculating the ratio of the number of pixels occupied by the calibration object in the image to its actual geometric dimensions. The specific value; Based on pixel resolution, the total number of pixels in each suspected disease element image is... Converting to actual physical area, the conversion logic is as follows: each pixel corresponds to an area of... Therefore, the actual physical area of ​​each suspected lesion pixel image is equal to the sum of the areas of all the tiny physical units corresponding to all pixels. The conversion formula is as follows: ,in, The actual physical area of ​​each suspected disease element image; To quantify the morphological extension characteristics of each suspected disease element image, this scheme uses the minimum bounding rectangle method to fit the geometric contour of each suspected disease element image. The minimum bounding rectangle is the rectangle that can completely contain the suspected disease element image and has the smallest area. Its direction is consistent with the principal axis direction of the suspected disease element image, which can truly reflect the maximum and minimum span of the suspected disease element image on the two-dimensional plane. The specific implementation steps are as follows: Extract the boundary contour point set within each suspected lesion pixel image. Then, calculate the minimum bounding rectangle of the contour based on an equivalent algorithm. This rectangle has a clear geometric meaning, and its length (denoted as ) The longest side of the rectangle corresponds to the maximum physical span of the suspected lesion pixel image along the principal axis, and its width (denoted as...) (i.e., the shortest side of the rectangle) corresponds to the minimum physical span of the suspected lesion image in the direction of the vertical principal axis; Based on the length of the minimum bounding rectangle of each suspected lesion element image and width The aspect ratio of each suspected lesion pixel image is calculated according to the following formula. :

[0023] The aspect ratio The larger the value, the more elongated the suspected disease element image is; the closer the value is to 1, the closer the suspected disease element image is to a circle. The perimeter of the boundary contour and the actual physical area of ​​each suspected lesion element image are extracted and calculated according to the formula: To obtain the roundness ,in, The perimeter of the boundary profile. Pick The range of values ​​for this roundness is: The closer the roundness is to 1, the closer the outline of the suspected disease element image is to a circle; the closer the roundness is to 0, the more irregular and elongated the outline of the suspected disease element image is. Geometric feature parameters are composed of actual physical area, aspect ratio, and circularity. A temperature sample set is formed by extracting the temperature values ​​corresponding to all pixels in each suspected disease element image. Temperature feature parameters are calculated based on the temperature sample set, including temperature mean, temperature range, and temperature skewness. The average temperature is used to reflect the overall temperature level of the suspected disease pixel images, that is, the average thermal state of the suspected disease pixel images. Its calculation process is as follows: sum all temperature values ​​in the temperature sample set, then divide by the total number of pixels corresponding to the suspected disease pixel images to obtain the arithmetic mean, which is used as the average temperature. ; Temperature range is used to reflect the absolute amplitude of temperature fluctuations within a suspected disease pixel image, i.e., the severity of temperature changes within the suspected disease pixel image. The calculation process is as follows: traverse the temperature sample set, find the maximum and minimum temperature values, calculate the difference between the two, and obtain the temperature range. ; Temperature skewness measures the degree and direction of asymmetry in temperature distribution, that is, whether the distribution of a set of temperature samples around its mean is biased towards higher or lower temperatures. It is calculated as follows: the temperature skewness is obtained based on the ratio of the third central moment to the cube of the standard deviation. The specific calculation formula is as follows: , in, The third central moment is used to measure the asymmetry of the distribution. The standard deviation of the temperature sample set is used to characterize the dispersion of the temperature distribution. The pixel number, For the first Temperature value of each pixel; Step 4: Combining the geometric and temperature characteristic parameters of each suspected disease element image, the disease status index of each suspected disease element image is determined. Based on the disease status index, a comprehensive evaluation of the suspected disease element images is conducted, thereby dividing the suspected disease element images into a safe disease element image set and a dangerous disease element image set. The specific implementation process is as follows: The geometric and temperature feature parameters of each suspected disease element image are linearly normalized, and the disease status index of each suspected disease element image is calculated accordingly. The specific calculation formula is as follows:

[0024] in, , , , , and These are the corresponding weight coefficients. , , , , and These are the linearly normalized dimensionless results of actual physical area, aspect ratio, roundness, mean temperature, temperature range, and temperature skewness, respectively, representing the disease state index. The larger the value, the higher the degree of damage and the more significant the thermal field anomaly in the suspected disease area corresponding to the current suspected disease image. The disease status index of each suspected disease element image. Compared with the preset disease risk threshold Compare; When the disease status index ≥Preset disease risk threshold At that time, the disease status index will be... The corresponding suspected disease element images are recorded as dangerous disease element images, and a dangerous disease element image set is constructed from them; When the disease status index <Preset disease risk threshold> At that time, the disease status index will be... The corresponding suspected disease element images are recorded as safe disease element images, and a safe disease element image set is constructed from them.

[0025] Step 5: Perform disease type diagnosis on the meta-images in the dangerous disease meta-image set. Based on the diagnosis results, determine whether to generate a diagnosis and treatment plan. If the diagnosis and treatment plan is generated, determine the processing time of the diagnosis and treatment plan. The specific implementation process is as follows: A set of images of hazardous road surface defects is retrieved, and a refined intelligent diagnosis of defect types is conducted on this set through a background intelligent analysis terminal. It should be noted that the background intelligent analysis terminal is an integrated control center for unified analysis and identification of basic road surface defect images. It has a built-in pre-trained intelligent diagnostic model for defects. The current diagnostic model is built on a deep learning machine learning model. Specifically, it uses a massive amount of historical defect image samples collected from the basic road surface (including but not limited to surface image grayscale parameters, texture feature parameters, contour edge parameters, and color gradient parameters under different lighting, angles, and scenes). The data collected from internal scanning imaging, including tomographic data, density distribution data, and defect coordinate data; synchronous sensor-collected state parameters such as matrix humidity, matrix stress, matrix deformation, and material aging coefficient; and corresponding standard disease labeling results (disease category records: crack disease, spalling disease, corrosion disease, void disease, settlement disease, etc.; disease level records: mild disease, moderate disease, severe disease, critical disease; disease origin archives, historical repair records, disease evolution cycle data, etc.) are input into deep learning machine learning models (including but not limited to convolutional neural networks CNN, deep residual networks ResNet, YOLO). In feature recognition networks, support vector machines (SVM), gradient boosting classification models, etc., the model features are iteratively trained and parameters are optimized and calibrated based on massive labeled samples. This enables the trained diagnostic model to extract multi-dimensional feature information from dangerous disease meta-images in real time, automatically and accurately determine the specific disease category, severity level, and distribution range corresponding to the meta-image, and output standardized disease diagnosis results. Then, based on the disease diagnosis results, the model can retrieve a targeted disease repair and treatment knowledge base to quickly generate a precise diagnosis and treatment plan adapted to the on-site working conditions, thereby achieving rapid identification, classification, and emergency targeted treatment of high-risk diseases in basic pavements. In the above process, a comprehensive and precise disease diagnosis operation is initiated only for the meta-images of dangerous diseases. This eliminates the need to systematically inspect and analyze all collected matrix images, effectively avoiding the computational and resource costs associated with invalid image data recognition and redundant feature extraction. The comprehensive feature data of the collected meta-images of dangerous diseases is then input into the diagnostic model, outputting standardized disease diagnosis results. If the diagnosis determines that the meta-image has a clear high-risk disease hazard, a specific diagnostic and treatment plan corresponding to the current disease is simultaneously determined, and the baseline treatment time required for the diagnostic and treatment plan is accurately extracted and marked as the processing time. Based on this, the following logical steps are executed: The disease-related area corresponding to the current matched diagnosis and treatment plan is recorded as the target disease area. The basic maintenance task corresponding to the target disease area is retrieved, and the current real-time disease treatment progress and the preset overall operation end time of the maintenance task are determined. Combined with the extracted processing time... The system calculates and extrapolates the theoretical completion time for disease repair operations based on the current baseline time. Since routine inspections, monitoring, and maintenance of the target disease area must be suspended during the implementation of the diagnostic treatment plan, it is determined whether the extrapolated completion time is less than the preset operation end time. If the result is less, the remaining buffer time between the completion time and the preset operation end time is calculated and recorded as the remaining buffer time. ; Synchronously collect the actual progress of the current maintenance tasks. and the corresponding progress consumption time (Values ​​are uniformly taken in minutes), based on the progress matching and accounting logic: Get the work progress and remaining incremental progress , to the progress of the work and remaining incremental progress The overall progress is accumulated and compared with the preset standard completion progress threshold (the standard completion progress threshold is set to 1). If it is determined that the standard is met, it proves that after the implementation of the diagnostic and treatment plan, the original basic maintenance tasks in the target disease area will not be delayed and will be completed on schedule.

[0026] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement any of the methods described above; A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements any of the methods described above.

[0027] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the protection scope of the technical solutions of the embodiments of the present invention.

Claims

1. An automatic identification method for multiple types of asphalt pavement defects based on artificial intelligence, comprising pre-setting an image acquisition time window and acquiring infrared thermal imaging images of the target detection section, characterized in that, It also includes the following steps: The infrared thermal imaging images are sorted according to the preset labels to form an ordered sequence of infrared thermal imaging images. Then, the suspected disease areas are identified in each frame of the infrared thermal imaging image sequence in turn. The identified suspected disease areas are then extracted in turn to form a set of suspected disease meta-images. Multidimensional feature recognition and analysis were performed on each suspected disease element image in the suspected disease element image set to determine the geometric and temperature feature parameters of the corresponding suspected disease element images. in, Geometric feature parameters include actual physical area, aspect ratio, and circularity; Temperature characteristic parameters include temperature mean, temperature range, and temperature skewness; By combining geometric and temperature feature parameters, the disease status index of each suspected disease element image is determined. Based on the disease status index, the suspected disease element images are comprehensively evaluated, thereby dividing the suspected disease element images into a safe disease element image set and a dangerous disease element image set. Disease type diagnosis is performed on the meta-images in the dangerous disease meta-image set. Based on the diagnosis results, it is determined whether to generate a diagnosis and treatment plan. If the diagnosis and treatment plan is generated, the processing time of the diagnosis and treatment plan is determined.

2. The method for automatic identification of multiple types of asphalt pavement defects based on artificial intelligence according to claim 1, characterized in that, The steps for identifying suspected diseased areas are as follows: For each frame of an infrared thermal imaging image sequence, a uniform grid is divided according to a preset step size. The geometric center of each grid is used as a sampling point, and detection points are set up to construct a set of detection points for each frame of the image. Simultaneously, the temperature values ​​of all detection points in the detection point set are extracted and the average value is calculated to obtain the global reference temperature of each frame of the image; For each frame of image, after obtaining the global reference temperature, the temperature deviation between the temperature value of each detection point and the global reference temperature is calculated. Then, taking each detection point as the center, several neighboring detection points within its neighborhood are selected to form a local detection area. The local standard deviation of the temperature values ​​of all detection points in the local detection area is calculated, and the local standard deviation is used as a dynamic correction factor to construct an adaptive dynamic threshold. If the absolute value of the temperature deviation at a certain detection point is greater than the adaptive dynamic threshold, the temperature change at that detection point is determined to be significantly different from that of the surrounding normal area, and the detection point is marked as an abnormal detection point. The DBSCAN clustering algorithm is used to perform spatial clustering of anomaly detection points, and areas of anomaly detection points that are continuously clustered in space are identified as suspected disease areas.

3. The method for automatic identification of multiple types of asphalt pavement defects based on artificial intelligence according to claim 1, characterized in that, The steps for solving the actual physical area in the geometric characteristic parameters are as follows: The images of each suspected disease element are preprocessed. Then, the total number of pixels in each suspected disease element image is counted. Then, the calibration parameters of the image acquisition system are combined, namely the pixel resolution. The pixel resolution is defined as the actual physical length corresponding to a single pixel. Based on pixel resolution, the total number of pixels in each suspected lesion image is converted into the actual physical area. The conversion logic is as follows: each pixel corresponds to an area of... The actual physical area of ​​each suspected lesion pixel image is equal to the sum of the areas of the tiny physical units corresponding to all pixels.

4. The method for automatic identification of multiple types of asphalt pavement defects based on artificial intelligence according to claim 1, characterized in that, The steps for solving the aspect ratio and roundness in the geometric feature parameters are as follows: Extract the boundary contour point set within each suspected lesion element image. Then, calculate the minimum bounding rectangle of the contour based on the equivalent algorithm. Its length corresponds to the maximum physical span of the suspected lesion element image in the principal axis direction, and its width corresponds to the minimum physical span of the suspected lesion element image in the vertical principal axis direction. Based on the length-to-width ratio of the minimum bounding rectangle of each suspected lesion element image, the aspect ratio of each suspected lesion element image is obtained. The circularity is calculated by extracting the perimeter of the boundary contour and the actual physical area of ​​each suspected lesion element image.

5. The method for automatic identification of multiple types of asphalt pavement defects based on artificial intelligence according to claim 1, characterized in that, The steps for solving the temperature characteristic parameters, including the temperature mean, temperature range, and temperature skewness, are as follows: Temperature sample set is formed by extracting the temperature values ​​corresponding to all pixels in each suspected disease element image, and temperature feature parameters are calculated based on the temperature sample set. In this process, all temperature values ​​in the temperature sample set are summed and then divided by the total number of pixels corresponding to the suspected disease element images to obtain the arithmetic mean, which is used as the temperature mean. Traverse the temperature sample set, find the maximum and minimum temperature values ​​respectively, calculate the difference between the two, and obtain the temperature range; Temperature skewness is obtained by the ratio of the third central moment to the cube of the standard deviation.

6. The method for automatic identification of multiple types of asphalt pavement defects based on artificial intelligence according to claim 1, characterized in that, The steps for calculating the disease status index are as follows: The geometric and temperature feature parameters of each suspected disease element image are linearly normalized, and then the dimensionless normalized results are multiplied by their respective weight coefficients and summed to obtain the disease status index of each suspected disease element image.

7. The method for automatic identification of multiple types of asphalt pavement defects based on artificial intelligence according to claim 1, characterized in that, The steps to divide suspected disease pixel images into safe disease pixel image sets and dangerous disease pixel image sets are as follows: The disease status index of each suspected disease element image. Compared with the preset disease risk threshold Compare; When the disease status index ≥Preset disease risk threshold At that time, the disease status index will be... The corresponding suspected disease element images are recorded as dangerous disease element images, and a dangerous disease element image set is constructed from them; When the disease status index <Preset disease risk threshold> At that time, the disease status index will be... The corresponding suspected disease element images are recorded as safe disease element images, and a safe disease element image set is constructed from them.

8. The method for automatic identification of multiple types of asphalt pavement defects based on artificial intelligence according to claim 1, characterized in that, When generating a diagnostic treatment plan, the following steps are performed: The disease-related area corresponding to the current matching diagnosis and treatment plan is recorded as the target disease area. The disease treatment operation progress and the preset overall operation end time corresponding to the target disease area are retrieved. The theoretical processing completion time of the disease repair operation is generated by combining the extracted processing time with the current system benchmark time. Determine whether the processing completion time is less than the job end time. If the result is less than, calculate the time interval between the processing completion time and the job end time and record it as the remaining buffer time. The system collects the actual progress of the current maintenance task and the corresponding time consumed. Based on the progress matching and calculation logic, it obtains the work progress and the remaining incremental progress. The work progress and the remaining incremental progress are accumulated to obtain the overall progress. The overall progress is compared and verified with the preset standard completion progress threshold. If it is determined to meet the standard, it proves that after the implementation of the diagnostic and treatment plan, it will not delay the completion of the original basic maintenance tasks in the target disease area on schedule.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method steps of any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method steps of any one of claims 1 to 8.

Images