Asphalt pavement construction quality detection method and system based on image recognition

Abstract

The invention provides an asphalt pavement construction quality detection method and system based on image recognition, and the method comprises the steps: obtaining a slice image of an asphalt mixture structure generated based on asphalt pavement core drilling sampling, and carrying out the optimization processing of the slice image, and obtaining a target slice image; extracting all characteristic parameters representing the construction quality of the asphalt pavement according to the target slice image, analyzing each characteristic parameter by using a pre-generated construction quality evaluation model, and generating an asphalt pavement construction quality score under each detection index; and when the construction quality scores of the asphalt pavement under each detection index reach the standard, outputting a detection result that the construction quality of the asphalt pavement is qualified. According to the invention, multi-dimensional and quantitative evaluation of the construction quality is realized, and the detection precision and efficiency of the construction quality of the asphalt pavement are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the quality detection technical field of asphalt pavement, and particularly relates to an asphalt pavement construction quality detection method and system based on image recognition. BACKGROUND

[0002] With the rapid development of modern transportation infrastructure construction, the importance of asphalt pavement construction quality detection is increasingly prominent. The pros and cons of construction quality are directly related to the service life of the road, driving safety and maintenance cost.

[0003] Traditional asphalt pavement construction quality detection methods mostly rely on manual field sampling and laboratory analysis, such as Marshall stability test, void ratio determination, etc. after asphalt pavement core sampling. However, this manual detection method is easily affected by subjective factors, and the detection accuracy is unstable, and the analysis of the internal structural characteristics of the asphalt pavement is not deep and comprehensive enough, which is prone to missed detection, misjudgment, etc., and it is difficult to meet the needs of modern asphalt pavement construction for high-quality and high-efficiency detection technology. SUMMARY

[0004] The present application provides an asphalt pavement construction quality detection method and system based on image recognition to improve the detection accuracy and efficiency of asphalt pavement construction quality.

[0005] To solve the above problems, the present application adopts the following technical solutions: On the one hand, the present application provides an asphalt pavement construction quality detection method based on image recognition, comprising: obtaining a slice image of asphalt mixture structure generated based on asphalt pavement core sampling; optimizing the slice image to obtain a target slice image; extracting all feature parameters representing the construction quality of asphalt pavement according to the target slice image, analyzing each feature parameter by using a pre-generated construction quality evaluation model, and generating an asphalt pavement construction quality score under each detection index; When the asphalt pavement construction quality scores under each detection index are up to standard, outputting a detection result that the asphalt pavement construction quality is qualified.

[0006] Preferably, the optimization of the slice image to obtain a target slice image comprises: analyzing the image brightness difference features between the slice image and the preset standard slice image to obtain an image brightness difference value; when it is determined that the image brightness difference value is greater than a preset image brightness difference threshold, the slice image is evenly divided into a plurality of first image blocks, the number of pixels of each brightness value in each first image block is counted, the proportion of the number of pixels of each brightness value in each first image block to the total number of pixels in the corresponding first image block is calculated, and the brightness distribution of each first image block is formed; The difference value between the brightness distribution and the brightness standard deviation of each first image block is calculated respectively, and the reciprocal of the difference value of each first image block is taken as the exponent with a natural constant as the base, so as to obtain the exponential value of each first image block. The image brightness difference value is multiplied by the exponential value of each first image block respectively to obtain the brightness adjustment value of each first image block. According to the brightness adjustment value of each first image block, the brightness value of each first image block is adjusted by using an interpolation algorithm to generate a target slice image.

[0007] Preferably, the slice image is optimized to obtain a target slice image, comprising: The image noise difference between the slice image and a preset standard slice image is analyzed, and the root mean square error between the slice image and the standard slice image is calculated to obtain an image noise difference value; When it is determined that the image noise difference value is greater than a preset image noise difference threshold, each pixel of the slice image is traversed, the variance of a local region centered on each pixel is calculated, the pixel with a variance less than a first threshold is marked as random noise, the pixel with a variance greater than the first threshold and less than a second threshold is marked as texture noise, and the pixel with a variance greater than the second threshold is marked as local artifact; The pixel value corresponding to the target pixel marked as random noise is replaced by the median value in the local region centered on the target pixel, the pixel average value and the pixel standard deviation of the pixel marked as texture noise in the local region are calculated, the pixel value corresponding to the pixel marked as texture noise is adjusted according to the pixel average value and the pixel standard deviation, and the pixel marked as local artifact is processed by using a polynomial fitting method to obtain a target slice image.

[0008] Preferably, all feature parameters representing the construction quality of the asphalt pavement are extracted from the target slice image, comprising: The gray level histogram corresponding to the target slice image is analyzed, the gray level distribution difference between the void region and the mixture region in the target slice image is found, the void region is segmented from the mixture region according to the gray level distribution difference, and a target void region and a target mixture region are formed; calculating a first average gray value of the target void region and a second average gray value of the target mixture region, setting a target gray value between the first average gray value and the second average gray value, determining pixels lower than the target gray value in the target void region as voids, and determining pixels higher than the target gray value in the target mixture region as mixtures; counting the total number of void pixels in the target slice image, and calculating the air voids of the asphalt mixture structure corresponding to the target slice image according to the actual area of the target slice image and the air voids formula.

[0009] Preferably, the extraction of all feature parameters representing the construction quality of the asphalt pavement from the target slice image includes: The morphological opening operation is used to remove the burr structure in the target slice image, and the eight-connected algorithm is used to label the aggregate particles in the target slice image and assign a unique identifier to each aggregate particle. For each labeled aggregate particle, the area equivalent diameter of each aggregate particle is calculated to determine the aggregate particle size, the proportion of the number of aggregates in different aggregate particle size ranges is counted, and an aggregate particle size distribution curve is generated.

[0010] Preferably, the extraction of all feature parameters representing the construction quality of the asphalt pavement from the target slice image includes: The mineral aggregate gap region in the target slice image is constructed as a graph structure; wherein the pixel points in the mineral aggregate gap region are nodes of the graph structure, and the connectivity between adjacent pixel points is the edge of the graph structure. The number of connected components, the size of the largest connected component, and the complexity of the connected path of the graph structure are calculated to form the connectivity feature parameters of the mineral aggregate gap, which are used to evaluate the connectivity of the mineral aggregate gap.

[0011] Further, before the construction quality evaluation model is used to analyze each feature parameter and generate the asphalt pavement construction quality score under each detection index, the method further includes: obtaining a plurality of training samples, wherein different training samples include different feature parameters representing the construction quality of the asphalt pavement and the asphalt pavement construction quality score under the detection index corresponding to each feature parameter; training the pre-constructed neural network model using the plurality of training samples; wherein the number of input layer neurons of the neural network model is determined according to the number of feature parameters, the middle hidden layer adopts a multi-layer structure, the number of output layer neurons is determined according to the number of divisions of the asphalt pavement construction quality grades, and the number of neurons in each layer is optimized and adjusted through trial and error method and cross-validation method. When the neural network model converges, the trained neural network model is used as a construction quality assessment model.

[0012] Furthermore, after the pre-built neural network model is trained using the multiple training samples, the method further includes: When the neural network model does not converge, optimizing the connection weights and biases of the neural network model based on a genetic algorithm, and searching for a parameter combination that optimizes the performance of the neural network model in a search space; The optimized neural network model is continuously trained using the multiple training samples until the neural network model converges.

[0013] Furthermore, after analyzing each characteristic parameter using the pre-generated construction quality assessment model to generate the asphalt pavement construction quality score under each detection indicator, the method further includes: When the asphalt pavement construction quality score corresponding to the detection index representing the void ratio does not meet the standard, a multivariate linear regression model is established, with the paving temperature, rolling passes, and mixture ratio of the asphalt mixture structure as independent variables and the void ratio as the dependent variable, and the regression coefficient is determined using the least squares method to generate a linear relationship equation between multiple construction parameters and the void ratio; At least one of the construction parameters is continuously adjusted using the linear relationship equation until the asphalt pavement construction quality score corresponding to the detection index characterizing the void ratio meets the standard, and a final adjusted construction parameter combination is output.

[0014] On the other hand, the present invention also provides an asphalt pavement construction quality detection system based on image recognition, comprising: An acquisition module, for acquiring a slice image of an asphalt mixture structure generated based on asphalt pavement core sampling; An optimization processing module, configured to optimize the slice image to obtain a target slice image; An analysis module is used to extract all characteristic parameters representing the construction quality of the asphalt pavement based on the target slice image, analyze each characteristic parameter using a pre-generated construction quality assessment model, and generate an asphalt pavement construction quality score under various detection indicators; The output module is used to output the test result that the asphalt pavement construction quality is qualified when the asphalt pavement construction quality scores under various test indicators meet the standards.

[0015] Compared with the prior art, the technical solution of the present invention has at least the following advantages: The present invention provides an image recognition-based asphalt pavement construction quality inspection method and system. This method optimizes slice images to generate target slice images, improving their quality and ensuring the clarity and accuracy of the image's feature information. Next, all characteristic parameters characterizing the asphalt pavement's construction quality are extracted from the target slice images. Each characteristic parameter is analyzed using a pre-generated construction quality assessment model to generate an asphalt pavement construction quality score for each inspection indicator. This enables more detailed and accurate identification of various characteristic parameters within the pavement mixture, enabling a multi-dimensional, quantitative assessment of construction quality. Furthermore, the automated construction quality assessment model significantly improves inspection efficiency, enabling rapid acquisition of inspection results and timely guidance for construction. Furthermore, a qualified asphalt pavement construction quality inspection result is output only when the scores for each inspection indicator meet the standards, thus ensuring the rigor and reliability of the inspection. This multi-indicator comprehensive assessment system overcomes the incomplete analysis limitations of traditional methods and provides strong technical support for the precise control of asphalt pavement construction quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a flowchart of an embodiment of the asphalt pavement construction quality detection method based on image recognition of the present invention; Figure 2 A slice image of an asphalt mixture structure in one embodiment of the present invention; Figure 3 The slice image after optimization processing in one embodiment of the present invention; Figure 4 This is a structural block diagram of an embodiment of the asphalt pavement construction quality detection system based on image recognition of the present invention. DETAILED DESCRIPTION

[0017] To help those skilled in the art better understand the present invention, the following will provide a clear and complete description of the technical solutions in the embodiments of the present invention, in conjunction with the accompanying drawings. Obviously, the described embodiments are only a portion of the embodiments of the present invention, not all of them. All other embodiments derived by those skilled in the art based on the embodiments of the present invention without creative effort are also within the scope of protection of the present invention.

[0018] Please refer to Figure 1 As shown, the present invention provides an asphalt pavement construction quality detection method based on image recognition, comprising the following steps: S11, obtaining a slice image of the asphalt mixture structure generated based on asphalt pavement core sampling; S12, optimizing the slice image to obtain a target slice image; S13, all feature parameters representing the construction quality of the asphalt pavement are extracted from the target slice image, and a pre-generated construction quality evaluation model is used to analyze each feature parameter to generate a construction quality score of the asphalt pavement under each detection index; S14, when the construction quality score of the asphalt pavement under each detection index meets the standard, a detection result that the construction quality of the asphalt pavement is qualified is output.

[0019] After the construction of the asphalt pavement is completed, a representative section is selected for core sampling. The core sampling equipment needs to ensure that the sampling process minimizes the damage to the pavement and ensures the integrity of the core sample.

[0020] Then, the drilled core sample is processed into a thin slice with a thickness of generally 1-2 millimeters and a smooth surface for subsequent imaging analysis. The slicing process should minimize damage to the internal structure of the mixture. A high-resolution microscope imaging device or a professional industrial CT scanner is used to take a slice image of the slice, and the imaging process needs to ensure that the image is clear and has a high enough resolution to distinguish the aggregates, voids and other details in the mixture.

[0021] For example, the asphalt mixture sample can be processed in a laboratory by slicing technology to cut it into a thin slice, and then a microscope or other imaging device is used to obtain a slice image of the thin slice. Referring to Figure 2 The slice image can be a grayscale image or a color image.

[0022] When the slice image is optimized to obtain a target slice image, the collected color image can be converted into a grayscale image to reduce the data volume and highlight the brightness information. At the same time, median filtering or wavelet transform can be used to remove noise in the image to improve the image quality. The median filtering is suitable for removing salt and pepper noise, and the wavelet transform is suitable for removing Gaussian noise. In addition, histogram equalization or adaptive contrast enhancement technology can be used to improve the contrast of the image, making the details of different gray levels more obvious. Or use an edge detection algorithm to enhance the edge information in the image to highlight the boundaries of aggregates and voids, which is convenient for subsequent feature parameter extraction. For example, referring to Figure 3 The target slice image is a binary processed slice image.

[0023] The characteristic parameters include void ratio, aggregate distribution, mineral aggregate gap ratio, and skeleton ratio, etc. The void ratio can be calculated by separating the void area from the mixture area in the target slice image through image segmentation technology. The aggregate distribution can be calculated by calculating the particle size distribution, spacing distribution, and distribution uniformity of the aggregate particles in the target slice image. The mineral aggregate gap ratio can be calculated by calculating the void ratio between the mineral aggregate particles in the target slice image to evaluate the compactness of the mixture. The skeleton ratio can be calculated by analyzing the volume ratio and distribution of coarse aggregate in the target slice image.

[0024] Then, the extracted characteristic parameters are input into a pre-trained construction quality evaluation model. The model can be a regression model or a classification model based on machine learning, which is used to map the characteristic parameters to the construction quality score. The construction quality evaluation model can calculate the construction quality score under each detection index according to the input characteristic parameters, and the score range is usually 0-100 points, or divided into excellent, good, medium, and poor levels.

[0025] The embodiment can set the standard conditions of each detection index according to the engineering requirements. For example, the score under the void ratio corresponding detection index needs to be greater than 80 points or the level is excellent, and the score under the aggregate distribution corresponding detection index needs to be greater than 75 points or the level is good and above.

[0026] Check whether the scores of all detection indexes meet the set standard conditions. If all indexes meet the standard, output the detection result of “the construction quality of asphalt pavement is qualified”; otherwise, output the result of “unqualified” and point out the unqualified index. The detection result is output in the form of a report, which can include slice images, characteristic parameter extraction results, score details, and final detection conclusion.

[0027] The application provides an asphalt pavement construction quality detection method based on image recognition. The method optimizes a slice image to obtain a target slice image, thereby improving the quality of the slice image and ensuring the clarity and accuracy of the feature information in the image. Then, all feature parameters representing the construction quality of the asphalt pavement are extracted from the target slice image, and a pre-generated construction quality evaluation model is used to analyze each feature parameter to generate a construction quality score under each detection index, thereby enabling more detailed and accurate identification of various feature parameters inside the pavement mixture and realizing multi-dimensional and quantitative evaluation of the construction quality. Meanwhile, the use of an automated construction quality evaluation model significantly improves the detection efficiency, enabling rapid acquisition of detection results and timely guidance of the construction. In addition, the detection result of the qualified construction quality of the asphalt pavement is output only when the scores under each detection index meet the standards, thereby ensuring the rigor and reliability of the detection. Meanwhile, the multi-index comprehensive evaluation system compensates for the shortcomings of the traditional method of incomplete analysis and provides strong technical support for the accurate control of the construction quality of the asphalt pavement.

[0028] In one of the embodiments, the optimization processing of the slice image to obtain a target slice image comprises: analyzing the image brightness difference features between the slice image and a preset standard slice image to obtain an image brightness difference value; when it is determined that the image brightness difference value is greater than a preset image brightness difference threshold, the slice image is evenly divided into a plurality of first image blocks, the number of pixels of each brightness value in each first image block is counted, the proportion of the number of pixels of each brightness value in each first image block to the total number of pixels in the corresponding first image block is calculated to form the brightness distribution of each first image block; the difference value between the brightness distribution of each first image block and the brightness standard deviation is calculated respectively, and the reciprocal of the difference value of each first image block is taken as the exponent with a natural constant as the base to calculate the exponential value of each first image block, the image brightness difference value is multiplied by the exponential value of each first image block to obtain the brightness adjustment value of each first image block, and the brightness values of each first image block are adjusted according to the brightness adjustment value of each first image block by using an interpolation algorithm to generate a target slice image.

[0029] The preset standard slice image is an "ideal" or "standard" high-quality asphalt mixture structure image, which can be determined through long-term experimental research and experience accumulation. By comparing the slice image obtained with the standard slice image, the pixel values (grayscale or color), texture structure and other features of the two images are compared point by point or region by region. For example, the difference between the two images can be quantified by calculating the sum of squares of the grayscale value difference between the corresponding pixel points in each pair of images, and the texture analysis algorithm can be used to compare the difference in texture patterns.

[0030] The image brightness difference feature reflects the degree of deviation of the actual slice image from the standard slice image in brightness. The brightness difference of each pixel or image region can be calculated by comparing the brightness values of the two images, and finally an image brightness difference value is obtained, so as to quantify the difference between the slice image and the standard image in brightness, and provide a basis for brightness adjustment.

[0031] The preset image brightness difference threshold is a standard value for judging whether the brightness of the slice image needs to be adjusted. When the image brightness difference value exceeds this threshold, it means that there is a significant difference between the brightness of the slice image and the standard image, and adjustment is needed. At this time, the slice image is divided into multiple first image blocks on average, so as to perform local processing on the image. Because different regions of the image may have different brightness characteristics, local adjustment can better improve the brightness uniformity of the image as a whole. For each first image block, the number of pixels corresponding to each brightness value (usually an integer between 0 and 255) is counted. Then, the proportion of the number of pixels of each brightness value in the total number of pixels of the first image block is calculated, thereby obtaining the brightness distribution of each first image block. The brightness distribution can reflect the concentration degree and variation of the pixel brightness within the first image block, and provide detailed information for brightness adjustment.

[0032] The brightness standard deviation can be regarded as an index for measuring the degree of deviation of the brightness distribution from the standard. The difference between the brightness distribution of each first image block and the brightness standard deviation is calculated, which can quantify the deviation degree of the brightness distribution of the first image block from the standard. The reciprocal of the difference is calculated with the natural constant e as the base number, and the obtained exponential value can be regarded as a weight factor for the degree of brightness adjustment. The size of the exponential value reflects the urgency of the adjustment of the first image block. Then, the overall image brightness difference value is multiplied by the exponential value of each first image block to obtain the brightness adjustment value of each first image block, so as to determine the specific brightness value that each first image block needs to adjust by combining the overall brightness difference and the characteristics of the local first image block, so that the brightness adjustment can be differentiated according to the characteristics of different regions.

[0033] The interpolation algorithm is used to smoothly transition the brightness change between adjacent first image blocks when adjusting the brightness of the first image block, so as to avoid obvious block boundaries. According to the brightness adjustment value of each first image block, the new brightness value of each pixel point in the first image block is calculated by using the interpolation algorithm, so as to improve the brightness quality of the slice image as a whole, and obtain the target slice image.

[0034] The embodiment can calculate the difference between the brightness distribution and the brightness standard deviation of each first image block, take a natural constant as the base, and take the inverse of the difference value of each first image block as the index, to obtain the index value of each first image block, and multiply the image brightness difference value by the index value of each first image block to obtain the brightness adjustment value of each first image block, so as to realize accurate brightness adjustment and effectively improve the brightness uniformity and visibility of the slice image, so that it is closer to the quality of the standard slice image. At the same time, the brightness adjustment of different regions makes the dark and bright parts in the image better presented. For example, in the asphalt mixture image, after brightness adjustment, the texture of aggregate, the boundary of void and other details can be more clearly displayed, which helps to more accurately analyze the structure of the mixture. In addition, the method of dividing the image into multiple first image blocks for local brightness adjustment can adapt to the brightness characteristics of different regions in the image, and compared with the overall brightness adjustment, this method can better handle the local brightness difference in the image and improve the overall quality of the image.

[0035] In one of the embodiments, the optimization processing of the slice image to obtain a target slice image comprises: analyzing the image noise difference characteristics between the slice image and the preset standard slice image, and calculating the root mean square error between the slice image and the standard slice image to obtain an image noise difference value; when it is determined that the image noise difference value is greater than a preset image noise difference threshold, traversing each pixel of the slice image, calculating the variance of the local region centered on each pixel, marking the pixels with variance less than a first threshold as random noise, marking the pixels with variance greater than the first threshold and less than a second threshold as texture noise, and marking the pixels with variance greater than the second threshold as local artifacts; replacing the pixel value corresponding to the target pixel marked as random noise with the median value in the local region centered on the target pixel, calculating the pixel average value and the pixel standard deviation of the pixels marked as texture noise in the local region, adjusting the pixel value corresponding to the pixels marked as texture noise according to the pixel average value and the pixel standard deviation, and processing the pixels marked as local artifacts by using a polynomial fitting method to obtain a target slice image.

[0036] The embodiment can compare the slice image with the preset standard slice image, analyze the difference between the two images pixel by pixel, and identify the image noise difference characteristics. The noise may be caused by the instability of the imaging device, environmental interference or the unevenness of the sample itself.

[0037] The root mean square error between the slice image and the standard slice image is calculated, and the root mean square error is used to quantify the difference between the two images, that is, the image noise difference value.

[0038] The calculated image noise difference value is compared with a preset image noise difference threshold value. If the image noise difference value is greater than the threshold value, it indicates that the noise level of the slice image is high and needs to be optimized; otherwise, it is considered that the image quality is good and does not need further processing. When optimizing, each pixel of the slice image is traversed, and the variance of the local region centered on each pixel is calculated. The size of the local region can be selected according to actual needs. The variance of the local region reflects the degree of brightness change around the pixel and can be used as a basis for distinguishing different types of noise.

[0039] Specifically, the pixels with a variance less than a first threshold value are marked as random noise. Random noise is usually caused by electronic noise of the imaging device or slight fluctuations of the sample itself, and is characterized by small local brightness changes. The pixels with a variance greater than the first threshold value and less than a second threshold value are marked as texture noise. Texture noise may be caused by the microstructure inside the sample, and is characterized by moderate local brightness changes. The pixels with a variance greater than the second threshold value are marked as local artifacts. Local artifacts are usually caused by uneven sample cutting or abnormalities during the imaging process, and are characterized by large local brightness changes. The first threshold value is less than the second threshold value.

[0040] For the target pixel marked as random noise, the pixel value corresponding to the pixel is replaced with the median value in the local region centered on the pixel. The median value has strong robustness to outliers and can effectively smooth random noise while avoiding excessive blurring of image details. The pixel average value and pixel standard deviation of the pixels marked as texture noise in the local region are calculated. According to the pixel average value and the pixel standard deviation, the pixel value corresponding to the pixel marked as texture noise is adjusted. The specific method can be to adjust the pixel value to the pixel average value plus a small random disturbance (the amplitude of the disturbance depends on the pixel standard deviation) to preserve some texture details. Finally, a polynomial fitting method can be used to process the pixels marked as local artifacts. The polynomial fitting method can construct a smooth polynomial surface according to the pixel distribution of the local region, and use the value of the surface to replace the value of the artifact pixel, thereby smoothing the local artifact, and finally obtaining the target slice image.

[0041] The embodiment can design corresponding optimization processing methods for three main noise types (random noise, texture noise, and local artifacts) in the asphalt mixture slice image, effectively process different types of noise, significantly reduce the noise level of the slice image, and improve the image quality and clarity. When processing texture noise, adjusting by considering the pixel mean value and pixel standard deviation can preserve some texture details while reducing noise, avoiding detail loss caused by excessive smoothing. In addition, the embodiment can also be applied to slice images with different noise levels. By setting different thresholds, the noise classification and processing strategy can be flexibly adjusted to adapt to various actual situations.

[0042] In one embodiment, the feature parameters representing the construction quality of the asphalt pavement are extracted from the target slice image, including: Analyze the gray histogram corresponding to the target slice image to find the gray distribution difference between the void area and the mixture area in the target slice image, and separate the void area from the mixture area according to the gray distribution difference to form a target void area and a target mixture area; Calculate the first average gray value of the target void area and the second average gray value of the target mixture area, set a target gray value between the first average gray value and the second average gray value, judge the pixels below the target gray value in the target void area as voids, and judge the pixels above the target gray value in the target mixture area as mixture; Statistically analyze the total number of void pixels in the target slice image, and calculate the void ratio of the asphalt mixture structure corresponding to the target slice image according to the actual area of the target slice image and the void ratio formula.

[0043] The gray histogram shows the frequency of each gray value in the target slice image. For the slice image of the asphalt mixture structure, the gray histogram usually presents two obvious peak regions: one peak corresponds to the mixture area (usually with a higher gray value because the mixture reflects light more strongly), and the other peak corresponds to the void area (with a lower gray value because the void reflects light more weakly). By observing the gray histogram, the gray distribution difference between the void area and the mixture area is found. The gray value of the void area is usually concentrated in a lower range (such as [0, 100]), while the gray value of the mixture area is concentrated in a higher range (such as [150, 255]). According to the gray distribution difference, the void area is separated from the mixture area to form a target void area and a target mixture area.

[0044] The average grayscale values ​​of the target gap area and the target mixture area are calculated respectively, and a target grayscale value between the first average grayscale value and the second average grayscale value is set. The pixels in the target gap area below the target grayscale value are judged as gaps, and the pixels in the target mixture area above the target grayscale value are judged as mixtures. The division of the gap area and the mixture area is readjusted according to the target grayscale value to ensure more accurate segmentation results.

[0045] The total number of pixels in the target void area after segmentation is counted, and the void ratio of the asphalt mixture structure corresponding to the target slice image is calculated based on the actual area of ​​the target slice image and the void ratio formula, which specifically includes the following formula: ; Among them, the is the void ratio of the asphalt mixture structure corresponding to the target slice image, is the total number of gap pixels in the target slice image, is the actual area of ​​the target slice image, is the total number of all pixels in the target slice image.

[0046] This embodiment can accurately separate the void area and the mixture area through grayscale histogram analysis and threshold selection, thereby improving the accuracy of segmentation, and can accurately reflect the void situation of the asphalt mixture structure by counting the total number of void pixels and calculating the void ratio.

[0047] In one embodiment, all characteristic parameters characterizing the asphalt pavement construction quality are extracted based on the target slice image, including: Using a morphological opening operation to remove burr structures in the target slice image, marking connected domains of aggregate particles in the target slice image according to an eight-connected algorithm, and assigning a unique identifier to each aggregate particle; For each marked aggregate particle, the area equivalent diameter of each aggregate particle is calculated to determine the aggregate particle size, and the proportion of aggregates in different aggregate particle size ranges is counted to generate the aggregate particle size distribution curve.

[0048] Among them, the morphological opening operation is used to remove small objects and burr structures in the target slice image. The opening operation consists of an erosion operation and a dilation operation. Specifically, a structuring element (usually a small two-dimensional array) is used to scan the target slice image. The erosion operation replaces the pixel values ​​in the area covered by the structuring element with the minimum value in that area, thereby reducing the bright areas (high grayscale value areas) in the target slice image. The eroded target slice image is then dilated. The dilation operation replaces the pixel values ​​in the area covered by the structuring element with the maximum value in that area, thereby expanding the bright areas in the target slice image. The erosion operation first removes small burr structures, and the dilation operation then restores the main target to near its original size, but the burr structures are completely removed.

[0049] The eight-connectedness algorithm is a region labeling method based on pixel connectivity. In the eight-connectedness algorithm, a pixel's neighbors include the pixels above, below, left, right, and diagonally (a total of eight neighbors). By traversing each pixel in the target slice image, connected pixel regions with the same grayscale value are labeled as the same connected domain. Specifically, a labeling matrix is ​​created with the same size as the target slice image, with an initial value of -1 (representing the unlabeled background area). A pixel-by-pixel scan is performed starting from the top left corner of the image. For each unlabeled pixel (labeling matrix value of -1), its eight surrounding neighbors are checked to find a labeled neighbor pixel and assign its label value to the current pixel. If no labeled neighbor is found, a new unique identifier is assigned to the current pixel and it is marked as a new connected domain. During the labeling process, pixels in the same connected domain may be assigned different label values. These conflicts need to be resolved through methods such as union-find to ensure that all pixels in the same connected domain have the same label value.

[0050] The area equivalent diameter is the diameter of a circle with the same area as the aggregate particle. For a marked aggregate particle, its area can be obtained by counting the number of pixels in the connected domain.

[0051] Based on project requirements, aggregate particle size is divided into several intervals, such as [0,2], [2,4], and [4,6]. The area-equivalent diameter of each aggregate particle is assigned to the corresponding particle size interval. The proportion of aggregate particles in each particle size interval to the total aggregate particle size is calculated. An aggregate size distribution curve is then plotted, with the particle size interval as the horizontal axis and the corresponding proportion as the vertical axis. This curve provides a visual representation of the aggregate particle size distribution and can be used to assist in evaluating characteristics such as gradation and uniformity of asphalt mixtures.

[0052] The embodiment can accurately identify and separate each aggregate particle through morphological opening operation denoising and connected domain labeling, ensure the accuracy of aggregate particle size measurement, and generate a particle size distribution curve that can intuitively reflect the distribution of aggregates by statistically analyzing the proportion of aggregates in different particle size ranges. This provides strong data support for evaluating whether the gradation of aggregates meets the design requirements. In addition, through the distribution of aggregate particle size, problems such as uneven distribution of aggregates or unreasonable gradation can be found in a timely manner, providing a scientific basis for adjusting and optimizing the construction process and improving the construction quality of asphalt pavement.

[0053] In one embodiment, the extraction of all feature parameters representing the construction quality of asphalt pavement from the target slice image includes: The mineral aggregate gap region in the target slice image is constructed as a graph structure; wherein the pixel points in the mineral aggregate gap region are nodes of the graph structure, and the connectivity between adjacent pixel points is the edge of the graph structure; The number of connected components, the size of the largest connected component, and the complexity of the connected path of the graph structure are calculated to form the connectivity feature parameters of the mineral aggregate gap, which are used to evaluate the connectivity of the mineral aggregate gap.

[0054] First, the mineral aggregate gap region needs to be identified from the target slice image. Image segmentation techniques, combined with gray threshold or edge detection algorithms, can be used to distinguish the mineral aggregate gap from other parts of the asphalt mixture, such as aggregate and asphalt matrix. Each pixel point in the mineral aggregate gap region is considered as a node in the graph structure. If two pixel points are adjacent in space, it is considered that there is an edge between the two nodes, representing their connectivity.

[0055] In graph theory, a connected component is the largest connected subgraph in a graph. For the graph structure of the mineral aggregate gap region, a connected component means that from any pixel point in the component, all other pixel points in the component can be reached through a continuous path.

[0056] Then, use depth-first search or breadth-first search algorithm to traverse the graph structure. From any unvisited node, visit all nodes connected to it and mark them as the same connected component. Repeat this process until all nodes are visited. The number of connected components obtained finally is the number of connected components of the mineral aggregate gap region.

[0057] The largest connected component is the largest connected component in the mineral aggregate gap region, and its size is usually measured by the number of pixel points (i.e. the number of nodes) in the component. In the process of calculating the number of connected components described above, the size of each connected component is recorded at the same time. Finally, find the largest one among all connected components and record its size, which is the size of the largest connected component.

[0058] The complexity of the connected path can be used to measure the tortuosity of the path in the aggregate gap region. It is usually measured by calculating the ratio of the path length (measured in the number of pixels or edges) to the straight-line distance. Specifically, a number of pairs of nodes in the graph structure can be selected, and the shortest path length between them is calculated. At the same time, the Euclidean distance between the two points is calculated. Then, the ratio of the two is calculated as the path complexity. The path complexities of multiple pairs of nodes are averaged to obtain the overall path complexity.

[0059] Finally, the number of connected components, the size of the largest connected component, and the complexity of the connected path of the graph structure are used as the connectivity characteristic parameters of the aggregate gap to evaluate the connectivity of the aggregate gap.

[0060] The embodiment can comprehensively evaluate the connectivity of the aggregate gap by constructing the graph structure and calculating the number of connected components, the size of the largest connected component, and the path complexity, which can reflect the internal structure characteristics of the mixture and provide important basic data for the evaluation of quality indicators such as void ratio and permeability. In addition, the connectivity characteristic parameters provide quantitative indicators for the evaluation of the void structure. For example, more connected components and larger path complexity may indicate more aggregate gaps and tortuous paths, which may affect the strength and durability of the mixture. Through the connectivity characteristic parameters, construction quality problems can be found in time, construction processes can be optimized, and construction quality can be improved.

[0061] In one embodiment, before the pre-generated construction quality evaluation model is used to analyze each characteristic parameter to generate the asphalt pavement construction quality score under each detection indicator, it further includes: Obtain a plurality of training samples, wherein different training samples include different characteristic parameters representing the construction quality of the asphalt pavement and the construction quality score of the asphalt pavement under the detection indicator corresponding to each characteristic parameter; Train the pre-constructed neural network model using the plurality of training samples; wherein the number of input layer neurons of the neural network model is determined according to the number of characteristic parameters, the middle hidden layer adopts a multi-layer structure, the number of output layer neurons is determined according to the number of divisions of the construction quality of the asphalt pavement, and the number of neurons in each layer is optimized and adjusted by trial and error and cross-validation methods; When the neural network model converges, the trained neural network model is used as the construction quality evaluation model.

[0062] The embodiment can collect a large amount of construction quality detection data of the asphalt pavement, which should cover different construction conditions, material characteristics, and construction processes. Each sample includes the following contents: Characteristic parameters: various characteristic parameters extracted from the slice image, such as void ratio, aggregate particle size distribution, aggregate gap connectivity, etc. Construction quality score: The construction quality score corresponding to each characteristic parameter under the detection index can be given by experts according to the actual detection results, or obtained through other existing detection methods (such as physical and mechanical property test).

[0063] Among them, the number of neurons in the input layer can be determined according to the number of characteristic parameters. For example, if there are 10 characteristic parameters, the input layer has 10 neurons. The hidden layer adopts a multi-layer structure, usually including 2-3 layers of hidden layers. The number of neurons in each layer needs to be optimized and adjusted by trial and error method and cross-validation method. The number of neurons in the output layer is determined according to the number of construction quality grades. For example, if the construction quality is divided into 4 grades (excellent, good, medium, and poor), the output layer has 4 neurons.

[0064] The training samples are normalized to scale the values of the characteristic parameters to the range of [0, 1] or [-1, 1] to improve the efficiency and stability of model training. Select a suitable loss function, such as mean square error (MSE) or cross-entropy loss function, to measure the difference between the predicted value and the actual value of the model. Select a suitable optimization algorithm, such as stochastic gradient descent, to adjust the weights and biases of the neural network to minimize the loss function.

[0065] The training samples are input into the neural network model, the predicted value is calculated by forward propagation, and then the weights and biases are updated by back propagation. Repeat this process until the model converges, and use the trained neural network model as the construction quality evaluation model. Use cross-validation method to evaluate the performance of the model to avoid overfitting. For example, observe the change of the loss function to judge whether the model has converged. When the loss function tends to be stable after multiple iterations, and the performance on the validation set no longer improves significantly, it is considered that the model has converged. Save the trained neural network model as the construction quality evaluation model. Finally, evaluate the model to ensure that its performance on the test set meets the expectations and can accurately map the characteristic parameters to the construction quality score.

[0066] The embodiment can automatically map the characteristic parameters to the construction quality score through the neural network model, reducing the subjectivity and complexity of manual evaluation. The neural network model can learn complex nonlinear relationships and provide high-precision evaluation results. At the same time, the training and prediction process of the model can be automated, improving the detection efficiency. In addition, through cross-validation and optimization adjustment, the model can adapt to different construction conditions and quality grades, and has strong generalization ability.

[0067] In one embodiment, after training the pre-constructed neural network model using the plurality of training samples, the method further comprises: When the neural network model does not converge, the connection weights and biases of the neural network model are optimized based on a genetic algorithm to search for a parameter combination that optimizes the performance of the neural network model in a search space. The optimized neural network model is further trained using the plurality of training samples until the neural network model converges.

[0068] During the training process, whether the neural network model converges is determined by observing the change in the loss function value. If the loss function value continues to decrease after multiple iterations, or the performance indicator (such as accuracy) on the validation set does not significantly improve, it indicates that the model has not converged.

[0069] When it is determined that the model does not converge, the connection weights and biases of the neural network model are optimized based on a genetic algorithm. The genetic algorithm is a search algorithm based on the principles of natural selection and genetics, which simulates the selection, crossover and mutation operations in the biological evolution process to find the optimal solution in the search space.

[0070] Specifically, first, a set of random weight and bias parameter combinations are generated to form an initial population. Each parameter combination is called an "individual", and the population size can be selected according to the computing resources and problem complexity. Each individual is evaluated using the training samples to calculate its fitness. Fitness can be measured by the loss function value or the performance indicator (such as accuracy) on the validation set. The higher the fitness, the closer the parameter combination is to the optimal solution. According to the fitness, individuals with higher fitness are selected into the next generation. Two parent individuals are selected, and new offspring individuals are generated by exchanging part of the genetic information. For example, a random selection of two parent individuals can be made to exchange part of the weight and bias parameters. Random mutations are performed on the newly generated offspring individuals to change the values of some genes with a certain probability. Mutation operations can increase the diversity of the population and avoid the algorithm falling into local optimal solutions. Repeat the selection, crossover and mutation operations to iteratively update the population until the termination condition (such as reaching the maximum number of iterations or the fitness reaching a certain threshold) is met, and obtain the optimized connection weights and bias parameter combination.

[0071] The weight and bias parameter combination optimized by the genetic algorithm is used to adjust the parameters of the neural network model, and the optimized neural network model is further trained using the training samples. The change in the loss function value is observed until the model converges. If the optimized model still cannot converge, the above genetic algorithm optimization steps can be repeated, or the structure of the neural network can be further adjusted (such as increasing the number of hidden layer neurons).

[0072] The genetic algorithm can find the optimal solution in a complex search space by simulating the biological evolution process, effectively optimizing the connection weights and bias parameters of the neural network, and improving the performance and convergence speed of the model. The crossover and mutation operations of the genetic algorithm can increase the diversity of the population, avoid the algorithm falling into a local optimal solution, and improve the global search ability of the model. In addition, by continuing to train the optimized parameter combination, the finally converged model has better generalization ability and can more accurately evaluate the construction quality of asphalt pavement under different construction conditions.

[0073] In one of the embodiments, after the pre-generated construction quality evaluation model is used to analyze each feature parameter and generate the asphalt pavement construction quality score under each detection index, it further includes: When the asphalt pavement construction quality score corresponding to the detection index representing the air void rate is not up to standard, a multiple linear regression model is established, taking the paving temperature, rolling times, and mixture ratio of the asphalt mixture structure as independent variables, and the air void rate as the dependent variable, and using the least squares method to determine the regression coefficients to generate a linear relationship equation between the multiple construction parameters and the air void rate. The linear relationship equation is used to continuously adjust at least one of the construction parameters until the asphalt pavement construction quality score corresponding to the detection index representing the air void rate is up to standard, and the final adjusted construction parameter combination is output.

[0074] In this embodiment, if the score corresponding to the detection index representing the air void rate is lower than the set standard threshold (for example, the air void rate score is lower than 80 points), it is considered that the index is not up to standard, triggering the subsequent optimization process.

[0075] The main construction parameters related to the air void rate are selected, including the paving temperature, rolling times, and mixture ratio, which are taken as independent variables, and the air void rate is selected as the dependent variable. Assuming that there is a linear relationship between the air void rate and the construction parameters, the regression coefficients are determined by minimizing the sum of squares of errors. Specifically, historical construction data including paving temperature, rolling times, mixture ratio, and corresponding air void rate measurement values are collected, and the independent variable data and dependent variable data are constructed as design matrix and vector respectively, and the regression coefficients are calculated using the least squares method formula to generate a linear relationship equation between multiple construction parameters and the air void rate. The fitting degree and prediction ability of the linear relationship equation are verified by cross-validation method to ensure the reliability and generalization ability of the linear relationship equation.

[0076] The at least one construction parameter is continuously adjusted by using the linear relationship equation until the detection index representing the air voids reaches the standard score of the asphalt pavement construction quality, and the final adjusted construction parameter combination is output. Specifically, the initial value of the current construction parameter is selected, for example, the current paving temperature is 150°C, the rolling number is 4 times, and the mixture ratio is a certain specific ratio. The predicted air voids is calculated according to the current construction parameter by using the linear relationship equation, and the predicted air voids is scored by using the construction quality evaluation model. If the score is not up to standard, the corresponding construction parameter is adjusted according to the sign and size of the regression coefficient. For example, if the regression coefficient is negative, it means that the higher the paving temperature, the lower the air voids, and therefore the paving temperature can be appropriately increased. The above process is repeated to gradually adjust the construction parameters until the predicted air voids score reaches the standard. When the predicted air voids score reaches the standard, the final adjusted construction parameter combination, including the paving temperature, the rolling number and the mixture ratio, is recorded and output to the construction management system or provided to the construction personnel in the form of a report for use in subsequent construction.

[0077] The embodiment can automatically quantify the relationship between the construction parameters and the air voids by establishing a multiple linear regression model and determining the regression coefficient by using the least squares method, provide a scientific basis for the optimization of the construction parameters, and reduce the blindness of manual adjustment. When the air voids is not up to standard, the construction quality can be quickly optimized by adjusting the construction parameters, the construction efficiency is improved, and the rework and delay caused by quality problems are avoided.

[0078] Please refer to Figure 4 The embodiment of the present application also provides an asphalt pavement construction quality detection system based on image recognition, which comprises: An acquisition module 11 is configured to acquire a slice image of asphalt mixture structure generated based on asphalt pavement core sampling; An optimization processing module 12 is configured to perform optimization processing on the slice image to obtain a target slice image; An analysis module 13 is configured to extract all feature parameters representing the asphalt pavement construction quality according to the target slice image, analyze each feature parameter by using a pre-generated construction quality evaluation model, and generate an asphalt pavement construction quality score under each detection index; An output module 14 is configured to output a detection result that the asphalt pavement construction quality is qualified when the asphalt pavement construction quality scores under each detection index all reach the standard.

[0079] As to the system in the above embodiment, the specific manner in which each module performs the operation has been described in detail in the embodiment related to the method, and will not be described in detail here.

[0080] In one of the embodiments, the present application further provides a storage medium storing computer readable instructions, which are executed by one or more processors to enable the one or more processors to perform the image recognition based asphalt pavement construction quality detection method. The storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.

[0081] Any combination of the technical features in the above described embodiments can be made, and for the sake of brevity, not all possible combinations of the technical features in the above described embodiments are described, however, as long as the combination of the technical features does not exist in contradiction, it shall be considered as within the scope of the present application.

[0082] The above described embodiments only express several implementation manners of the present application, and the description is relatively specific and detailed, however, it shall not be understood as a limitation on the patent scope of the present application. It shall be noted that, for the ordinary skilled in the art, several modifications and improvements can be made without departing from the concept of the present application, and these shall be within the protection scope of the present application. Therefore, the patent protection scope of the present application shall be subject to the appended claims.

Claims

1. A method for detecting asphalt pavement construction quality based on image recognition, characterized in that: include: Obtain slice images of asphalt mixture structure generated based on asphalt pavement core sampling; performing optimization processing on the slice image to obtain a target slice image; Extracting all characteristic parameters representing the construction quality of the asphalt pavement based on the target slice image, analyzing each characteristic parameter using a pre-generated construction quality assessment model, and generating an asphalt pavement construction quality score under various detection indicators; When the asphalt pavement construction quality scores under various test indicators meet the standards, the test results indicating that the asphalt pavement construction quality is qualified are output.

2. The asphalt pavement construction quality detection method based on image recognition according to claim 1 is characterized in that: The step of optimizing the slice image to obtain a target slice image includes: Analyzing image brightness difference characteristics between the slice image and a preset standard slice image to obtain an image brightness difference value; When it is determined that the image brightness difference value is greater than a preset image brightness difference threshold, the slice image is evenly divided into a plurality of first image blocks, the number of pixels of each brightness value in each of the first image blocks is counted, and the ratio of the number of pixels of each brightness value in each first image block to the total number of pixels in the corresponding first image block is calculated to form a brightness distribution of each first image block; The difference between the brightness distribution and the brightness standard deviation of each first image block is calculated respectively, and the exponent value of each first image block is calculated with a natural constant as the base and the inverse of the difference of each first image block as the exponent. The image brightness difference value is multiplied by the exponent value of each first image block to obtain the brightness adjustment value that should be adjusted for each first image block. The brightness value of each first image block is adjusted accordingly according to the brightness adjustment value that should be adjusted for each first image block using an interpolation algorithm to generate a target slice image.

3. The asphalt pavement construction quality detection method based on image recognition according to claim 1 is characterized in that: The step of optimizing the slice image to obtain a target slice image includes: Analyzing image noise difference characteristics between the slice image and a preset standard slice image, and calculating a root mean square error between the slice image and the standard slice image to obtain an image noise difference value; When it is determined that the image noise difference value is greater than a preset image noise difference threshold, traversing each pixel of the slice image, calculating the variance of a local area centered on each pixel, marking pixels with a variance less than a first threshold as random noise, marking pixels with a variance greater than the first threshold and less than a second threshold as texture noise, and marking pixels with a variance greater than the second threshold as local artifacts; The pixel value corresponding to the target pixel marked as random noise is replaced by the median value in the local area centered on the target pixel, the pixel mean and pixel standard deviation of the pixels marked as texture noise in the local area are calculated, the pixel value corresponding to the pixel marked as texture noise is adjusted according to the pixel mean and pixel standard deviation, and the pixels marked as local artifacts are processed using a polynomial fitting method to obtain the target slice image.

4. The asphalt pavement construction quality detection method based on image recognition according to claim 1 is characterized in that: The method of extracting all characteristic parameters characterizing the construction quality of the asphalt pavement according to the target slice image includes: Analyzing a grayscale histogram corresponding to the target slice image to find a grayscale distribution difference between a void region and a mixture region in the target slice image, and segmenting the void region from the mixture region based on the grayscale distribution difference to form a target void region and a target mixture region; Calculating a first average grayscale value of the target gap region and a second average grayscale value of the target mixture region, setting a target grayscale value between the first average grayscale value and the second average grayscale value, determining pixels in the target gap region below the target grayscale value as gaps, and determining pixels in the target mixture region above the target grayscale value as mixtures; The total number of void pixels in the target slice image is counted, and the void ratio of the asphalt mixture structure corresponding to the target slice image is calculated based on the actual area of ​​the target slice image and the void ratio formula.

5. The asphalt pavement construction quality detection method based on image recognition according to claim 1 is characterized in that: The method of extracting all characteristic parameters characterizing the construction quality of the asphalt pavement according to the target slice image includes: Using a morphological opening operation to remove burr structures in the target slice image, marking connected domains of aggregate particles in the target slice image according to an eight-connected algorithm, and assigning a unique identifier to each aggregate particle; For each marked aggregate particle, the area equivalent diameter of each aggregate particle is calculated to determine the aggregate particle size, and the proportion of aggregates in different aggregate particle size ranges is counted to generate the aggregate particle size distribution curve.

6. The asphalt pavement construction quality detection method based on image recognition according to claim 1 is characterized in that: The method of extracting all characteristic parameters characterizing the construction quality of the asphalt pavement according to the target slice image includes: Constructing the mineral gap region in the target slice image into a graph structure; wherein the pixel points in the mineral gap region serve as nodes of the graph structure, and the connectivity between adjacent pixel points serves as edges of the graph structure; The number of connected components, the size of the largest connected component and the complexity of the connected paths of the graph structure are calculated to form connectivity characteristic parameters of the mineral gaps, which are used to evaluate the connectivity of the mineral gaps.

7. The asphalt pavement construction quality detection method based on image recognition according to claim 1 is characterized in that: Before analyzing each characteristic parameter using the pre-generated construction quality assessment model to generate the asphalt pavement construction quality score under each detection indicator, the method further includes: Acquire multiple training samples, where different training samples include different characteristic parameters representing the asphalt pavement construction quality and asphalt pavement construction quality scores under detection indicators corresponding to each characteristic parameter; The pre-constructed neural network model is trained using the multiple training samples; wherein the number of neurons in the input layer of the neural network model is determined according to the number of characteristic parameters, the intermediate hidden layer adopts a multi-layer structure, the number of neurons in the output layer is determined according to the number of asphalt pavement construction quality grades, and the number of neurons in each layer is optimized and adjusted through trial and error and cross-validation methods; When the neural network model converges, the trained neural network model is used as a construction quality assessment model.

8. The asphalt pavement construction quality detection method based on image recognition according to claim 7 is characterized in that: After the pre-built neural network model is trained using the multiple training samples, the method further includes: When the neural network model does not converge, optimizing the connection weights and biases of the neural network model based on a genetic algorithm, and searching for a parameter combination that optimizes the performance of the neural network model in a search space; The optimized neural network model is continuously trained using the multiple training samples until the neural network model converges.

9. The asphalt pavement construction quality detection method based on image recognition according to claim 1, characterized in that: After analyzing each characteristic parameter using the pre-generated construction quality assessment model to generate the asphalt pavement construction quality score under each detection indicator, the method further includes: When the asphalt pavement construction quality score corresponding to the detection index representing the void ratio does not meet the standard, a multivariate linear regression model is established, with the paving temperature, rolling passes, and mixture ratio of the asphalt mixture structure as independent variables and the void ratio as the dependent variable, and the regression coefficient is determined using the least squares method to generate a linear relationship equation between multiple construction parameters and the void ratio; At least one of the construction parameters is continuously adjusted using the linear relationship equation until the asphalt pavement construction quality score corresponding to the detection index characterizing the void ratio meets the standard, and a final adjusted construction parameter combination is output.

10. An asphalt pavement construction quality detection system based on image recognition, characterized in that: include: An acquisition module, for acquiring a slice image of an asphalt mixture structure generated based on asphalt pavement core sampling; An optimization processing module, used for optimizing the slice image to obtain a target slice image; An analysis module is used to extract all characteristic parameters representing the construction quality of the asphalt pavement based on the target slice image, analyze each characteristic parameter using a pre-generated construction quality assessment model, and generate an asphalt pavement construction quality score under various detection indicators; The output module is used to output the test result that the asphalt pavement construction quality is qualified when the asphalt pavement construction quality scores under various test indicators meet the standards.

Images