Pavement crack recognition method and system based on yolov8-seg

Abstract

The application discloses a kind of based on YOLOv8-Seg's road surface crack identification method and system, belong to road engineering pavement maintenance technical field.The method includes: using the trained YOLOv8-Seg segmentation model to the road surface crack slice image is identified, and the mask region of crack is generated;Extract the attribute information of crack, attribute information includes center point coordinate, boundary box, crack area and mask binary graph;The multiple images continuously collected in the same road range are analyzed in space-time continuity, judge whether there is similar shape and similar location crack;If there is similar shape and similar location crack in the image of at least two different points, then it is judged as real crack, and the identification result is output.The method realizes crack pixel level segmentation and attribute extraction, and combines space-time continuity analysis across images, judges crack authenticity, effectively reduces false detection rate, improves identification accuracy and engineering applicability.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of pavement maintenance of road engineering, and particularly relates to a pavement crack identification method and system based on YOLOv8-Seg. BACKGROUND

[0002] Semi-rigid base asphalt pavement structure is widely used in high-grade highways in China. Cracks are easily generated in this structure due to dry-wet cycles, freeze-thaw effects and repeated loads during long-term service. These cracks not only affect driving comfort, but also damage the integrity of the pavement structure. After water intrusion, secondary diseases such as interlayer peeling and structure loosening are easily caused, which significantly shortens the service life of the road. Therefore, it is crucial to detect the cracks of the road base regularly, quickly and accurately.

[0003] At present, the automatic detection method of road diseases based on convolutional neural network (CNN) has become a research hotspot. The YOLO series single-stage target detection algorithm is widely used in crack identification due to its high efficiency and good accuracy. However, this method still has obvious limitations in practical engineering applications: the recognition accuracy is severely dependent on the quantity and quality of the training data set, and the model generalization ability is insufficient, resulting in a high misjudgment rate; only relying on the recognition result of a single image, it is difficult to distinguish between real cracks and transient noise, artifacts and other interference, and the reliability needs to be improved.

[0004] Therefore, there is an urgent need for an intelligent detection method that can effectively improve the accuracy of crack identification, reduce the false detection rate, and be suitable for practical engineering scenarios. SUMMARY

[0005] The purpose of the embodiments of the present application is to provide a pavement crack identification method based on YOLOv8-Seg, which realizes pixel-level segmentation and attribute extraction of cracks through the YOLOv8-Seg model, and combines cross-image spatiotemporal continuity analysis to judge the authenticity of cracks, effectively reduce the false detection rate, and improve the recognition accuracy and engineering applicability, so as to solve at least one technical problem involved in the background art.

[0006] In order to solve the above technical problems, the present application is implemented as follows:

[0007] The embodiments of the present application provide a pavement crack identification method based on YOLOv8-Seg, comprising the following steps:

[0008] Step S1, using a trained YOLOv8-Seg segmentation model to identify the pavement crack slice image, and generating a mask region of the crack;

[0009] Step S2, extracting attribute information of the crack, the attribute information including center point coordinates, a bounding box, a crack area and a mask binary graph;

[0010] Step S3, spatial-temporal continuity analysis is performed on the plurality of images continuously collected in the same road range to determine whether there are cracks similar in shape and close in position;

[0011] Step S4, if there are cracks similar in shape and close in position in the images of at least two different points, it is determined to be a real crack, and the recognition result is output.

[0012] Optionally, in step S2, the attribute information of the crack is extracted by using an image processing function in an OpenCV library, specifically including:

[0013] The crack contour is extracted using the cv2.findContours function;

[0014] The crack center point coordinates are calculated using the cv2.moments function;

[0015] The crack bounding box is obtained using the cv2.boundingRect function;

[0016] The mask is binarized using NumPy to generate a mask binary image.

[0017] Optionally, in step S3, the judgment criteria for cracks similar in shape and close in position include:

[0018] The intersection-over-union IoU between the two crack masks is calculated, and if the IoU is greater than a set threshold, it is considered that the shapes are similar;

[0019] The Euclidean distance between the two crack center points is calculated, and if the Euclidean distance is less than a set threshold, it is considered that the positions are close.

[0020] Optionally, the threshold of the intersection-over-union IoU is set to 0.3 to 0.5, and the threshold of the Euclidean distance is set to 15 to 25 pixels.

[0021] Optionally, the threshold of the intersection-over-union IoU is 0.4, and the threshold of the Euclidean distance is 20 pixels.

[0022] Optionally, in step S1, the pavement crack slice image is obtained by ground penetrating radar and is labeled for training of the YOLOv8-Seg segmentation model.

[0023] Optionally, in step S4, when the recognition result is output, the image with the largest crack area is selected as the representative result, and the crack annotation box and the mask area are drawn in the image.

[0024] The application also provides a pavement crack recognition system for implementing the method, which comprises:

[0025] An image acquisition module for acquiring pavement crack slice images;

[0026] a processing module configured to execute the pavement crack identification method;

[0027] an output module configured to output the crack identification result.

[0028] Compared with the prior art, the present application has the following advantages:

[0029] 1. High identification accuracy and low false detection rate: By introducing the YOLOv8-Seg segmentation model, pixel-level accurate identification of crack morphology is realized; combined with subsequent spatio-temporal continuity analysis, cross-validation is performed using multiple image information, effectively distinguishing real cracks from noise, artifacts and other interference in a single image, significantly reducing the probability of false positives and false negatives.

[0030] 2. Strong engineering applicability and stability: The proposed spatio-temporal continuity determination rules (such as mask IoU and center point distance) can tolerate natural changes in crack morphology in different depth or angle images, are not sensitive to target size and position changes, making the method more robust and reliable in actual complex engineering environments.

[0031] 3. High degree of automation and efficiency improvement: The method realizes full-process automation from image recognition, feature extraction to authenticity judgment, overcoming the shortcomings of traditional manual detection such as strong subjectivity and low efficiency, and providing an effective technical means for rapid and regular automated inspection of large-scale highway networks.

[0032] 4. Providing accurate decision support for preventive maintenance: By accurately identifying early-stage base cracks and tracing their development at different depths, the method provides accurate positioning and morphology information for subsequent grouting repair and other maintenance operations, helping to implement precise preventive maintenance, extend road service life, and save maintenance costs. BRIEF DESCRIPTION OF DRAWINGS

[0033] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor based on these drawings:

[0034] Figure 1 A flowchart of the pavement crack identification method based on YOLOv8-Seg provided by the present application;

[0035] Figure 2 A mask area graph of the crack provided by the present application;

[0036] Figure 3A crack profile map provided by the present application;

[0037] Figure 4 A crack center point map provided by the present application;

[0038] Figure 5 A crack boundary box map provided by the present application;

[0039] Figure 6 A crack mask binary map provided by the present application;

[0040] Figure 7 A crack area map provided by the present application;

[0041] Figure 8 One of the intersection over union maps provided by the present application;

[0042] Figure 9 The second intersection over union map provided by the present application;

[0043] Figure 10 A crack recognition result map provided by the present application. DETAILED DESCRIPTION

[0044] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0045] The terms "first", "second", and the like in the specification and claims of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described herein, and the objects distinguished by "first", "second", etc. are generally of a kind and do not limit the number of objects, for example, the first object can be one or more. In addition, "and / or" in the specification and claims indicates at least one of the connected objects, and the character " / ", generally indicates that the objects before and after are in an "or" relationship.

[0046] Please refer to Figure 1 The embodiments of the present application provide a road surface crack recognition method based on YOLOv8-Seg, which includes the following steps:

[0047] Step S1, using a trained YOLOv8-Seg segmentation model to recognize the road surface crack slice image, and generating a crack mask area;

[0048] Step S2, attribute information of the crack is extracted, and the attribute information includes a center point coordinate, a bounding box, a crack area, and a mask binary graph;

[0049] Step S3, a plurality of images continuously collected in the same road range are subjected to spatio-temporal continuity analysis, and it is judged whether there are cracks similar in shape and close in position;

[0050] Step S4, if there are cracks similar in shape and close in position in images of at least two different points, it is determined to be a real crack, and the recognition result is output.

[0051] In step S1, in the embodiment, the pavement crack slice image is obtained by ground penetrating radar, specifically, a pulse ground penetrating radar system is used, the center frequency is 1.0 GHz, the sampling point number is 512, and the scanning interval is 0.05 m. The obtained radar image is first subjected to background removal and gain correction, then is divided into a slice image every 10 m in the road longitudinal direction, the image size is uniformly adjusted to 640*640 pixels, and is subjected to gray scale normalization processing.

[0052] All image data is manually labeled, and a road engineering professional uses a LabelMe tool to perform pixel-level labeling on a crack area, generates a JSON format labeling file corresponding to the image, and uses the JSON format labeling file for model training and verification.

[0053] Further, the structure of the YOLOv8-Seg model is as follows:

[0054] A standard Backbone-Neck-Head architecture is used to extract multi-scale features, wherein:

[0055] The Backbone uses a CSPDarknet53 structure to extract multi-scale features;

[0056] The Neck uses a PAN-FPN structure to enhance the feature fusion capability;

[0057] The Head adds a mask branch to the detection head, the branch generates a plurality of prototype masks through a small full convolution subnet, and predicts a set of mask coefficients for each detection instance. In the inference process, the prototype mask and the coefficient are linearly combined, and then subjected to Sigmoid activation and up-sampling to obtain the final mask of each instance.

[0058] The training process of the YOLOv8-Seg model is as follows:

[0059] The YOLOv8-Seg model is trained using a self-built pavement crack dataset, the dataset contains 5000 labeled images, and is divided into a training set, a verification set and a test set in a ratio of 8:1:1. The training parameter settings are as follows:

[0060] Pre-trained weights: COCO dataset pre-trained model

[0061] Number of training rounds: 100 epochs

[0062] Batch size: 16

[0063] Optimizer: AdamW with initial learning rate of 0.001

[0064] Loss function: Weighted combination of cross-entropy loss and Dice loss

[0065] Data augmentation: Including random rotation, brightness adjustment, Gaussian noise addition, etc.

[0066] After training, the average precision (mAP@0.5) of the model on the test set reached 89.2%, and the segmentation accuracy (mIoU) was 82.7%.

[0067] As shown in Figure 2 , the trained YOLOv8–Seg model can perform pixel-level segmentation on the input slice image and generate a mask area of the crack. The mask is represented in the form of a binary image, where the white area (pixel value of 1) represents the crack, and the black area (pixel value of 0) represents the background.

[0068] In step S2, the attribute information of the crack is extracted through image processing functions in the OpenCV library, including:

[0069] 1. Contour extraction: Use the cv2.findContours function to extract the crack contour, as shown in Figure 3 ;

[0070] 2. Center point calculation: Use the cv2.moments function to calculate the moment of the contour, and then get the center point coordinates, as shown in Figure 4 ;

[0071] 3. Boundary box acquisition: Use the cv2.boundingRect function to obtain the minimum bounding rectangle of the crack, as shown in Figure 5 ;

[0072] 4. Mask binarization: Use NumPy to perform binarization processing on the mask to generate a mask binary image, as shown in Figure 6 ;

[0073] 5. Crack area calculation: Use the cv2.contourArea function to calculate the contour area, as shown in Figure 7 , the area value is equal to the number of pixels occupied by the crack area, which is used to evaluate the prominence of the crack.

[0074] In step S3,Figure 8 and Figure 9 As shown in the drawings, the criteria for judging the similar morphology and the similar position include:

[0075] Calculate the intersection over union IoU between the two crack masks, and if the IoU is greater than a set threshold, consider the morphology similar;

[0076] Calculate the Euclidean distance between the two crack center points, and if the Euclidean distance is less than a set threshold, consider the position similar.

[0077] The threshold of the intersection over union IoU is set to 0.3-0.5, and the threshold of the Euclidean distance is set to 15-25 pixels.

[0078] The threshold of the intersection over union IoU is 0.4, and the threshold of the Euclidean distance is 20 pixels.

[0079] In step S4, the condition for determining the real crack is that the crack appears in different images in the same road range no less than 2 times.

[0080] When outputting the result, the system automatically selects the image with the largest crack area as the representative result, and draws the crack bounding box and mask area in the image, as shown in Figure 10 .

[0081] The present application also provides a pavement crack identification system for implementing the method, which comprises:

[0082] An image acquisition module for acquiring pavement crack slice images;

[0083] A processing module for executing the pavement crack identification method;

[0084] An output module for outputting the crack identification result.

[0085] The pavement crack identification method based on YOLOv8-Seg provided by the present application is compared with the related "bounding box fusion + appearance frequency threshold" method, and the results are shown in Table 1.

[0086] Table 1 Experimental results

[0087]

[0088] As can be seen from the experimental results, the method provided by the present application has more accurate spatial position, supports shape change, and has stronger robustness, and is less likely to misjudge or miss.

[0089] Experimental effect: on a test road with a length of 2 kilometers, the false detection rate of the method is reduced from 11.2% of the single YOLOv8-Seg model to 2.8%, and the false detection crack number is reduced from 8 to 2, the false detection rate is reduced by 8.4%, which shows significant technical advantages.

[0090] It should be noted that in this document, the terms "comprise", "comprising", or any other variant thereof are intended to cover non-exclusive inclusions, so that processes, methods, articles, or devices that include a series of elements not only include those elements, but also include other elements not explicitly listed, or inherent to such processes, methods, articles, or devices. Without more limitations, the element defined by the statement "comprises a" does not exclude the presence of other identical elements in the process, method, article, or device that includes the element.

[0091] In addition, it should be pointed out that the scope of the method and system in the embodiments of the present application is not limited to performing functions in the order shown or discussed, but can also include performing functions in a substantially simultaneous manner or in the opposite order according to the functions involved, for example, the described method can be performed in an order different from that described, and various steps can also be added, omitted, or combined. In addition, the features described with reference to certain examples can be combined in other examples.

[0092] The embodiments of the present application are described above in combination with the drawings, but the present application is not limited to the above specific embodiments, and the above specific embodiments are only illustrative, not limiting, and those skilled in the art can make many forms under the inspiration of the present application without departing from the purpose of the present application and the scope protected by the claims, which all belong to the protection of the present application.

Claims

1. A method for identifying pavement cracks based on YOLOv8-Seg, characterized in that, Includes the following steps: Step S1: Use the trained YOLOv8-Seg segmentation model to identify road surface crack slice images and generate mask regions for the cracks. Step S2: Extract the attribute information of the crack, including the center point coordinates, bounding box, crack area, and mask binary image; Step S3: For multiple images continuously acquired within the same road area, calculate the intersection-union ratio (IoU) between two crack masks and the Euclidean distance between the center points of the two cracks. If the crossover ratio (IoU) is greater than a set threshold, the crack morphology is considered similar. If the Euclidean distance is less than a set threshold, the crack locations are considered to be close. Step S4: If cracks with similar shapes and close positions exist in images at at least two different locations, they are determined to be real cracks, and the recognition result is output.

2. The method according to claim 1, characterized in that, In step S2, the attribute information of the crack is extracted using image processing functions in the OpenCV library, specifically including: Use the cv2.findContours function to extract the crack contour; Use the cv2.moments function to calculate the coordinates of the crack center point; Use the cv2.boundingRect function to obtain the crack bounding box; Use NumPy to binarize the mask and generate a binary image of the mask.

3. The method according to claim 1, characterized in that, The threshold for the Intersection over Union (IoU) is set to 0.3 to 0.5, and the threshold for the Euclidean distance is set to 15 to 25 pixels.

4. The method according to claim 3, characterized in that, The threshold for the Intersection over Union (IoU) is 0.4, and the threshold for the Euclidean distance is 20 pixels.

5. The method according to claim 1, characterized in that, In step S1, the road surface crack slice image is acquired by ground penetrating radar and labeled for use in training the YOLOv8-Seg segmentation model.

6. The method according to claim 1, characterized in that, In step S4, when outputting the recognition results, the image with the largest crack area is selected as the representative result, and the crack annotation box and mask area are drawn in the image.

7. A pavement crack identification system for implementing the method as described in any one of claims 1 to 6, characterized in that, include: The image acquisition module is used to acquire slice images of road surface cracks; The processing module is used to execute the road surface crack identification method; The output module is used to output the crack identification results.

Images