CN-121978098-A - Visual detection method and device for grading segregation degree and rigidity distribution of large-particle-size macadam foundation

CN121978098ACN 121978098 ACN121978098 ACN 121978098ACN-121978098-A

Abstract

The invention discloses a visual detection method and device for grading segregation degree and rigidity distribution of a large-particle-size macadam foundation, which are characterized in that a matrix multi-view image acquisition unit is constructed, 3-5 industrial cameras are distributed according to the vertex and the center position of an equilateral triangle, synchronous exposure in 0.01s is realized by a synchronous trigger controller, full coverage shooting is carried out on a large-particle-size macadam foundation detection area, a GPS and IMU positioning module is integrated, accurate space anchoring of detection data can be realized, grading-rigidity result fusion visualization is completed through a red-yellow-green three-color coding system, PDF report generation and 4G/5G cloud uploading functions are combined, the overall process closed loop of on-site detection-data operation-result output-cloud supervision is realized, the construction can acquire the accurate positions of segregation and rigidity abnormal areas in real time, and a supervision department can remotely trace source quality problem causes, so that the intelligent and accurate level of quality control of a road foundation is greatly improved.

Inventors

JIANG LIN
ZHANG WEIJIAN
JIANG WEI
Cui Longhu
CHEN WU
XU AO
ZHANG YU
YE JUNTAO
Kong Shaolin
CHANG QI

Assignees

安徽建工公路桥梁建设集团有限公司

Dates

Publication Date: 20260505
Application Date: 20251216

Claims (10)

1. A visual detection method for grading segregation degree and rigidity distribution of a large-particle-size macadam foundation is characterized by synchronously acquiring grading and rigidity parameters based on a non-contact visual detection principle, and specifically comprises the following steps: S1, constructing a matrix multi-view image acquisition unit, distributing 3-5 industrial cameras according to the vertexes and the central positions of an equilateral triangle, realizing the synchronous exposure of each camera in0.01 s through a synchronous trigger controller, and carrying out full coverage shooting on a large-particle-size broken stone base layer detection area; S2, carrying out standardized preprocessing on the multi-source image data, wherein the processing flow sequentially comprises highlight elimination, noise filtration and image registration, wherein the highlight elimination adopts a global threshold screening-local gradient correction compound algorithm, a global highlight threshold is automatically determined through an Otsu algorithm, a highlight region is screened out, a local gradient of the highlight region is calculated through a Sobel operator, and the highlight region texture recovery is realized through gradient reverse correction; S3, carrying out accurate segmentation on the broken stone particles on a standardized image set based on an improved U-Net semantic segmentation model, introducing a double-attention module integrating channel attention and space attention to a coding end by the improved U-Net model, optimizing edge feature extraction by adopting a deformable convolution module (the number of sampling points is set to be 9) at a decoding end, stabilizing the F1 fraction to be more than 97.5% after training a 1200-group marked broken stone image by the model, extracting the closed contour of each broken stone particle by a contour tracing algorithm after segmentation, calculating the equivalent particle diameter (adopting an area equivalent method, wherein S is the contour area of the particles), the particle distribution density (the number of the particles in the unit area) and the slenderness ratio (the aspect ratio of the particles circumscribed rectangle), and establishing a grading feature database containing single particle features and regional statistical features; S4, constructing a grading segregation degree evaluation model by combining highway engineering standards, firstly determining standard grading screen residual ranges corresponding to each screen hole (2.36 mm, 4.75mm, 9.5mm, 19mm, 31.5mm and 63 mm) according to the technical Specification for highway asphalt pavement construction (JTGF 40-2004), then converting the particle occupation ratio counted in a grading characteristic database into equivalent screen residual values, and calculating single screen mesh screen residual deviation rate and accumulated deviation rate, wherein the single screen mesh screen residual deviation rate= (actual measured equivalent screen residual value-standard screen residual median)/standard screen residual median multiplied by 100%, and the accumulated deviation rate is a weighted sum (weight is sequentially 0.3, 0.25, 0.2, 0.15, 0.07 and 0.03 according to the screen mesh particle size, and finally judging the grading segregation grade by combining the single screen mesh deviation and accumulated deviation results; S5, establishing a grading-stiffness association model, wherein model training samples are obtained in a double mode of 'indoor test and on-site actual measurement', 30 groups of test pieces are manufactured according to different grading proportions indoors, a drop hammer type deflection meter (FWD) is adopted to test dynamic rebound modulus, 50 typical road sections are selected on site, image grading data and FWD actual stiffness values are synchronously collected to finally form 1200 groups of effective training samples, a random forest algorithm is adopted to construct a mapping relation, the number of decision trees in the algorithm is set to 100, the maximum depth is 15 layers, the minimum sample division number is 8, model prediction errors are controlled to be within 7.2% after 5-fold cross validation, grading characteristic parameters obtained in the step S3 are input into the model, GPS coordinate information attached to images is combined, and stiffness distribution thermodynamic diagram with grid precision of 1m x 1m is generated through an interpolation algorithm; S6, fusing grading segregation degree evaluation results with a rigidity distribution thermodynamic diagram, adopting a red (severe segregation/rigidity shortage), yellow (moderate segregation/rigidity standard reaching) and green (slight segregation/rigidity excellent) three-color coding system for visual marking, wherein the comprehensive report also needs to contain detection time, road section pile number, environmental parameters (temperature 5-35 ℃ and humidity 30-80%) and model confidence level, and supporting data to be stored locally in a PDF format or uploaded to a highway quality supervision platform through a 4G/5G module.
2. The method according to claim 1, wherein the noise filtering in step S2 is performed by a cascade of gaussian filtering and median filtering, the standard deviation of the gaussian filtering is set to 1.2-1.8, and the window size of the median filtering is set to 5 x 5 pixels.
3. The method according to claim 1, wherein the improved U-Net semantic segmentation model in step S3 optimizes feature extraction by means of a deformable convolution module that introduces fused attention, the F1 score of the model being no less than 97.5%.
4. The method according to claim 1, wherein the grading segregation grade in step S4 is classified into slight segregation, medium segregation and heavy segregation, and the grading screen deviation rates correspond to ±5%, ±5% -10% and ±10% or more, respectively.
5. The method according to claim 1, wherein the training samples of the grading-stiffness correlation model in step S5 comprise more than 1000 sets of different grading parameters and corresponding measured dynamic rebound modulus values, and the model prediction error is less than 8%.
6. A visual inspection device for a large particle size macadam base, for implementing the method of any one of claims 1 to 5, comprising: The image acquisition module consists of 3-5 high-definition industrial cameras and matched LED light supplementing units, the resolution of the cameras is not lower than 500 ten thousand pixels, the frame rate is not lower than 25fps, and the light supplementing units support stepless brightness adjustment; the positioning module integrates a GPS and IMU inertial measurement unit to realize real-time positioning of the detection position, and the positioning accuracy error is smaller than 0.1m; The data processing module adopts an embedded processor, and is internally provided with the image preprocessing algorithm, the segmentation model and the association model according to claim 1, so as to support the real-time operation of data; The output module comprises a touch display screen and a wireless communication unit, and can display the detection result on site and upload the data to the cloud platform.
7. The device of claim 6, wherein the camera lens of the image acquisition module adopts a large depth of field design, and the focal length range is 8-16mm, so that clear imaging of the crushed stone particles within a shooting distance of 2-5m is ensured.
8. The device of claim 6, wherein the LED light filling units are arranged in a circular array, the color temperature adjustment range is 4500K-6500K, and the light filling modes can be automatically switched according to the ambient light intensity.
9. The apparatus of claim 6, wherein the data processing module is further configured with a storage unit having a storage capacity of not less than 128GB, and capable of buffering at least 1000 sets of the detected data and the original image.
10. The device of claim 6, further comprising a portable support, wherein the support is made of carbon fiber, the height adjustment range is 1.2-2.5m, and the lockable universal wheels are arranged at the bottom.

Description

Visual detection method and device for grading segregation degree and rigidity distribution of large-particle-size macadam foundation Technical Field The invention relates to the technical field of visual detection, in particular to a visual detection method and device for grading segregation degree and rigidity distribution of a large-particle-size macadam foundation. Background The large-grain-size broken stone base layer has the advantages of high strength, good water permeability, strong bearing capacity and the like, is widely applied to pavement base structures of high-grade roads and heavy-load traffic roads, and the grading uniformity and the rigidity distribution of the broken stone base layer directly determine the overall service life and the driving safety of the road pavement, so that the broken stone base layer is a core index for managing and controlling road construction media. The current detection means for large-particle-size broken stone substrates in the industry have a plurality of limitations: 1. In the aspect of grading detection, the traditional method mainly comprises manual on-site sampling and indoor screening, has long detection period (2-3 hours are needed for single sampling and screening), low efficiency, randomness in sampling, and difficulty in realizing full coverage of a detection area, and partial portable screening equipment can be operated on site, still is in contact type detection, is easy to cause secondary disturbance on a basic structure, cannot synchronously acquire particle space distribution characteristics, and is difficult to accurately judge the space range of grading segregation. In addition, the existing visual detection scheme mostly adopts single-view imaging, particle identification omission is easily caused by broken stone particle shielding and uneven illumination, the extraction precision of the segmentation model on the edge characteristics of large-particle-size broken stone is insufficient, and the calculation error of grading parameters is generally more than 10%. 2. In the aspect of rigidity detection, the main flow method is Falling Weight Deflectometer (FWD) single-point detection, the method can only acquire rigidity data of discrete points and cannot form a continuous rigidity distribution map, the rigidity detection and the grading detection are mutually independent, a correlation model of the rigidity detection and the grading detection is lacking, and the cause of abnormal rigidity is difficult to explain from the grading angle, so that the quality problem is low in tracing efficiency. 3. In the aspect of data management, most of traditional detection data are paper records or scattered electronic files, a unified visual presentation and cloud supervision channel is lacking, hysteresis exists between a constructor and data interaction of supervision departments, and real-time early warning and correction of quality problems cannot be realized. In summary, the existing detection technology is difficult to meet the quality detection requirements of high efficiency, accuracy, universe and linkage in the construction of large-particle-size broken stone base layers, and an integrated detection technology capable of synchronously acquiring grading and rigidity parameters is needed. Disclosure of Invention In order to overcome the defects of the prior art, one of the purposes of the invention is to provide a visual detection method and device for grading segregation degree and rigidity distribution of a large-particle-size macadam foundation. One of the purposes of the invention is realized by adopting the following technical scheme: A visual detection method for grading segregation degree and rigidity distribution of a large-particle-size macadam foundation, which is based on a non-contact visual detection principle, realizes synchronous acquisition of grading and rigidity parameters and specifically comprises the following steps: S1, constructing a matrix multi-view image acquisition unit, distributing 3-5 industrial cameras according to the vertexes and the central positions of an equilateral triangle, realizing the synchronous exposure of each camera in0.01 s through a synchronous trigger controller, and carrying out full coverage shooting on a large-particle-size broken stone base layer detection area; S2, carrying out standardized preprocessing on the multi-source image data, wherein the processing flow sequentially comprises highlight elimination, noise filtration and image registration, wherein the highlight elimination adopts a global threshold screening-local gradient correction compound algorithm, a global highlight threshold is automatically determined through an Otsu algorithm, a highlight region is screened out, a local gradient of the highlight region is calculated through a Sobel operator, and the highlight region texture recovery is realized through gradient reverse correction; S3, carrying out accurate segmentation on the broke