High-speed guardrail plate plastic spraying defect detection method and system

Abstract

The invention relates to the technical field of high-speed guardrail plate plastic spraying defect detection, in particular to a high-speed guardrail plate plastic spraying defect detection method and system.The high-speed guardrail plate plastic spraying defect detection method comprises the steps of image preprocessing, data set manufacturing, model training, image detection, image splicing and post-processing; different defects with the same characteristics can be distinguished to obtain a normal image and a defect image; the defect images are further detected, a screening threshold value is set, and the defect images are further distinguished into necessary processing class images and unnecessary processing class images by calculating the number, the area and the distance of the defects; and a plastic spraying defect detection system for the high-speed guardrail plate is built according to the plastic spraying defect detection method for the high-speed guardrail plate. The method has the effects of quantitatively evaluating the severity of the defect and improving the detection efficiency.

Description

Technical Field

[0001] The present application relates to the technical field of plastic spray defect detection for highway guardrails, and in particular to a method and system for detecting plastic spray defect for highway guardrails. Background Art

[0002] Guardrail board spraying is a surface treatment technology, which is mainly used to improve the anti-corrosion performance, aesthetics and service life of guardrail boards. Defects will inevitably occur in the process of guardrail board spraying, and spraying defects may cause the quality, performance and appearance of the guardrail board to not meet the requirements.

[0003] Plastic spray defects usually include: bubbles, particles, sags, pits, abrasions, scratches, damage and deformation, etc. Among these defects, some will affect the anti-corrosion performance of the guardrail, such as bubbles and scratches; while some defects only affect the aesthetics of the guardrail and will not damage the integrity of the plastic spray coating, such as sags; in actual operation, some defects may be accepted or tolerated for various reasons (such as cost, time, process difficulty, etc.). In this case, only the defects that must be handled need to be detected and distinguished, but defects with different degrees of impact may have similar characteristics and need to be distinguished.

[0004] In the existing technology, most methods for detecting defects in spray-painted highway guardrails will detect all types of defects. However, existing image processing methods can only identify the existence of defects but cannot quantitatively assess the severity of the defects. Manual judgment is required as to whether repair is needed, resulting in low detection efficiency. Summary of the invention

[0005] In order to improve detection efficiency, the present application provides a method for detecting plastic spray defects of high-speed guardrail panels.

[0006] In the first aspect, the present application provides a method for detecting defects in spray-coated high-speed guardrails, which adopts the following technical solution: A method for detecting defects in spray-coated high-speed guardrails, comprising the following steps: Image preprocessing: acquiring historical images and performing image preprocessing, wherein the historical images contain different types of defects with the same characteristics; Dataset preparation: annotating different types of defects with the same characteristics in the historical images and preparing training sets and verification sets; Model training: using the training set to train the classification model, and using the validation set to evaluate the classification accuracy of the model to obtain a trained model; Image detection: Obtain the image to be detected, input the image to be detected into the trained model for detection, and obtain the recognition result image with defect recognition marks; Image stitching: stitching multiple recognition result images that are combined into the same guardrail board to synthesize a complete guardrail board image, each guardrail board image contains a guardrail board; according to the recognition result, re-differentiate the guardrail board image into a normal image and a defective image; Post-processing: including quantity counting step, judgment step, repair step and conformity assessment step; Quantity statistics: per unit area The defect images within the range are counted for defect quantity and the unit area is set The threshold of the number of defects allowed within ; Number of defects The calculation model is as follows: = , in, Indicates Unit area The total number of defects in =1, 2, 3…; Indicates unit area The number of defects of type i in ; Judgment: When the number of defects is the number of defects Less than or equal to the set quantity threshold When the number of defects is Greater than the set quantity threshold , dividing the defective image into necessary processing images and performing a repair step; Conformity assessment: the normal images and the non-necessary processing images are qualified images and do not need to be processed; Repair: The images that need to be processed are unqualified images and need to be reworked and repaired.

[0007] By adopting the above technical solution, the present application first obtains historical images, obtains training sets and verification sets through image preprocessing and data annotation, then trains the classification model with the training set and uses the verification set to evaluate the classification accuracy of the model to obtain a trained model; inputs the image to be detected into the trained model to obtain a recognition result image, splices the recognition result image, combines multiple recognition result images located on the same guardrail board into one, and divides normal images and defective images to facilitate subsequent processing; finally, post-processing is performed to count the number of defects in the defective image and set a quantity threshold , the defect images are further divided into non-essential processing images and necessary processing images, and qualification assessment and repair are carried out; not only can the defect type be identified, but also the number of defects can be counted to determine whether repair is needed, thereby quantifying the severity of the defects, reducing manual identification, effectively improving detection efficiency, and reducing the uncertainty of manual identification.

[0008] Optionally, a reclassification step is provided between the determination step and the conformity assessment step; Reclassification: including the impact health value calculation step and the re-judgment step; Impact health value calculation: Set different weights for different types of defects according to their impact, and calculate the unit area The impact of the number of defects on health value , and set the health value threshold ; Impact health value The calculation model is as follows: , in, Indicates The impact health value per unit area, , It is The number of class defects, , is the corresponding weight, and the sum of the weights is ; Re-judge: When the impact health value Less than or equal to the health value threshold When the impact health value is Greater than the health threshold When the non-essentially processed image is stored again as an essential processed image, the repair step is performed.

[0009] By adopting the above technical scheme, the impact of defects on the quality of guardrails is quantified according to the defect type, tolerability and impact of defects, and different weights are set for the number of different types of defects. Defects with a large impact are given a high weight, and defects with a small impact are given a low weight. This can affect the setting of the health value, increase the proportion of defects with a large impact in the impact health value, and improve the data value and credibility of the impact health value. By evaluating the severity of defects through the impact health value, the quality of defect assessment can be effectively improved, and the determination of whether the defect image needs to be processed can be made more accurate.

[0010] Optionally, between the re-judgment step and the conformity assessment step, a defect area statistics step and a third judgment step are further provided; Defect area statistics: Defect area of ​​non-essentially processed images Calculate and set the defect area limit threshold ; Defect area The calculation model is as follows: , in, Indicates Unit area The total defect area within ; Indicates unit area The area of ​​the i-th defect, ; The third judgment: when the defect area Less than or equal to the set defect area limit threshold When the defect area is Greater than the set defect area limit threshold , the non-necessary processing images are classified into necessary processing images, and the repair step is performed; Defect area limit threshold The calculation model is as follows: , Where P is a proportionality coefficient, The value range is , Indicates unit area.

[0011] By adopting the above technical solution, the non-essentially processed images that meet the defect quantity requirements are processed into defect area Calculation of the defect area Does not exceed the set area threshold When the defect area is Exceeds the set area threshold When the non-essential processing images are divided into the necessary processing images, the number of defects can be reduced to meet the requirements by introducing the defect area, but the product quality problems caused by the large defect coverage area can be solved by setting the area threshold. , which can convert the unit area The defect coverage area is limited to an acceptable range; a comprehensive assessment of the severity of defects through the number of defects and the defect area can reduce the limitations of a single evaluation standard and improve product inspection quality.

[0012] Optionally, a defect merging determination step is further provided between the third determination step and the conformity assessment step; Defect merging and identification: including defect distance statistical steps, distance judgment steps and defect adjustment steps; Defect statistics: Defect spacing between defects in non-essentially processed images Calculate and set the spacing threshold ; Distance judgment: When the defect spacing Calculate and set the spacing threshold ; Distance judgment: When the defect spacing Less than the set spacing threshold When the defect spacing is Greater than or equal to the set spacing threshold When the defect area is maintained No change, and the conformity assessment steps are performed; the spacing threshold is a constant representing the minimum spacing between defects; Defect adjustment: Defect area Adjust the coefficient and update the defect area The calculation model is as follows: , in, Indicates Unit area The total defect area within ; Indicates unit area The area of ​​the i-th defect, , , n represents the defect spacing Less than the set spacing threshold The number of times.

[0013] By adopting the above technical solution, when the defect spacing Less than the set spacing threshold When the coating adhesion between defects is reduced, it is easy to make the defect area Increased, and the defect spacing Less than the set spacing threshold The more times, the greater the defect area The larger the defect area, the larger the defect area can be adjusted. As the defect spacing Less than the set spacing threshold The defect area increases with the increase of the number of The results are more accurate, providing stronger data support for the further classification of non-essential processing images, effectively improving the inspection quality and efficiency of products.

[0014] Optionally, set different quantity thresholds for the outer and inner surfaces of the guardrail , Health value threshold , Defect area limit threshold and spacing threshold ; wherein the outer surface is the main surface that can be directly seen during normal use, the inner surface is the surface that cannot be observed during normal use, and the tolerance of the inner surface to defects is higher than the tolerance of the outer surface to defects.

[0015] By adopting the above technical scheme, different defect detection standards can be set according to the tolerance of different surfaces of the guardrail plate for defects; the outer surface is the main surface, which is easily in direct contact with vehicles, flying sand and stones, wind and rain, etc., and the highest detection standard can be set. The inner surface is the surface that cannot be contacted during normal use. Compared with the outer surface, the detection standard can be relatively lowered, which is conducive to reducing the rework rate of the guardrail plate, saving costs and improving work efficiency.

[0016] Optionally, an edge determination step and a fourth determination step are further provided between the distance determination step and the conformity assessment step; Edge discrimination step: performing edge discrimination on defects on the non-essentially processed images to determine whether the defects are located at the edge of the guardrail plate; The fourth judgment step: when the defect is not at the edge of the guardrail board, the non-essential processing image continues to be stored as a non-essential processing image, and the qualification assessment step is performed; when the defect is located at the edge of the guardrail board, the non-essential processing image is directly classified as a necessary processing image, and the repair step is performed.

[0017] By adopting the above technical solution, guardrail panels with defects at the edge can be directly classified as images that need to be processed and reworked and repaired; because defects at the edge are prone to friction and damage, causing them to fall off, affecting the quality of the guardrail panels.

[0018] In the second aspect, the present application provides a high-speed guardrail spray defect detection system, which adopts the following technical solutions: A high-speed guardrail spray defect detection system, comprising: Image preprocessing module: used to obtain historical images from the database and perform image preprocessing; the image preprocessing includes graying, edge detection and cropping of historical images; Data set making module: the input end is electrically connected to the output end of the image preprocessing module, and is used to annotate data of different types of defects with the same characteristics in historical images and make training sets and verification sets; Model training module: the input end is electrically connected to the output end of the data set making module, and is used to train the classification model using the training set and evaluate the classification accuracy of the model using the validation set to obtain a trained model; Image detection module: including a trained model; used to obtain the image to be detected from the file system, input the image to be detected into the trained model for detection, and obtain the recognition result image with defect recognition marks; Image stitching module: the input end is electrically connected to the output end of the image detection module, and is used to stitch multiple recognition result images of the same guardrail board to synthesize a complete guardrail board image, each guardrail board image contains a guardrail board; according to the recognition result, the guardrail board image is re-classified into a normal image and a defective image; Post-processing module: The input end is electrically connected to the output end of the image stitching module, and is used to perform quantitative analysis on the defects in the defect image. By counting the number of defects per unit area and setting a threshold for the number of defects, it is determined whether the number of defects per unit area meets the requirements, and the defect image is further divided into non-essential processing images and necessary processing images.

[0019] In summary, the present application includes at least one of the following beneficial technical effects: 1. This application can not only identify the defect type, but also determine whether repair is needed by counting the number of defects, thereby quantifying the severity of the defects, reducing manual identification, effectively improving detection efficiency, and reducing the uncertainty of manual identification.

[0020] 2. This application quantifies the impact of defects on the quality of guardrails according to the defect type, tolerability, and impact of the defects, sets different weights for the number of different types of defects, assigns high weights to defects with a large impact, and assigns low weights to defects with a small impact. This can affect the setting of the health value, increase the proportion of defects with a large impact in the impact health value, and improve the data value and credibility of the impact health value. By evaluating the severity of the defects through the impact health value, the quality of defect assessment can be effectively improved, and the determination of whether the defect image needs to be processed can be made more accurate.

[0021] 3. This application calculates the defect area of ​​non-essential processing images that meet the defect quantity requirements. When the defect area does not exceed the set area threshold, the non-essential processing image will continue to be stored as a non-essential processing image; when the defect area exceeds the set area threshold, the non-essential processing image will be classified as a necessary processing image; by introducing the defect area, the product quality problems caused by the defect quantity meeting the requirements but the defect coverage area being too large can be reduced. By setting the area threshold, the defect coverage area per unit area can be limited to an acceptable range; by comprehensively evaluating the severity of the defects by the defect quantity and defect area, the limitations of a single evaluation standard can be reduced and the product inspection quality can be improved.

[0022] 4. Through defect adjustment, the present application can increase the defect area as the number of times the defect spacing is less than the set spacing threshold increases, making the obtained defect area result more accurate, providing more powerful data support for the further division of non-essential processing images, and effectively improving the product inspection quality and efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 is a flow chart of Example 1 of the present application; Figure 2 is a flow chart of Example 2 of the present application; Figure 3 is a flow chart of Example 3 of the present application; Figure 4 This is a flow chart of Example 4 of the present application. DETAILED DESCRIPTION

[0024] The following combination Figures 1 to 4 This application is described in further detail.

[0025] This embodiment discloses a method for detecting defects in plastic spraying of high-speed guardrails.

[0026] Example 1: Reference Figure 1 , a method for detecting defects of spray-coated high-speed guardrails, comprising the following steps: Image preprocessing: Acquire historical images and perform image preprocessing. The historical images contain different types of defects with the same characteristics. For obvious defects such as missed spraying and scratches, it can be directly determined whether they need to be processed. However, bulging defects are not easy to distinguish with the naked eye due to their similar appearance characteristics, and there are many types of bulging defects with different degrees of influence. Therefore, in this embodiment, bulging defects will be distinguished. Bulging defects usually include: bubbles, particles and sagging. Among them, bubbles are circular or elliptical cavities of varying sizes; particles are granular substances embedded in the layer, mostly in irregular dot shape; sagging is the downward flow of the coating due to gravity during the drying process, forming curtain-like or tear-like marks of varying widths and lengths; according to the characteristics of the defects, bubbles and particles can easily cause peeling and cracking of the coating, seriously affecting the appearance quality and performance of the coating, and sagging can affect the uniformity and aesthetics of the coating.

[0027] Specifically, image preprocessing includes graying, edge detection and cropping of historical images; graying is the process of converting a color image into a grayscale image using a graying algorithm. This embodiment uses a weighted average method to grayscale the image. The grayscale image can highlight the structure and shape information of the image, simplify the image information, and reduce the complexity of image processing; edge detection is used to extract the outline of the defect, providing a basis for subsequent image segmentation algorithms and edge discrimination steps; cropping is used to remove unnecessary parts of the historical image and crop the historical image to a specific size to meet the input requirements of the classification model and unify the image size in the data set; during the cropping process, an image cropping algorithm is used for cropping and the cropped image is numbered to facilitate identification of the initial position of the cropped image, correspondence of the defect position, and subsequent stitching and restoration of the image using an image stitching algorithm.

[0028] Dataset creation: Label different types of defects with the same characteristics in historical images and create training sets and validation sets. Data labeling can be done manually or with semi-automatic labeling tools. Use algorithms to perform preliminary labeling on historical images, and then make manual corrections. During the data labeling process, different attribute values ​​need to be assigned to different types of defects.

[0029] Compared with defect-free images, the number of defective images is very small. To address this problem, upsampling and image enhancement techniques are used to expand the dataset of defective images, increase the diversity and quantity of data, and improve the training effect.

[0030] Model training: Use the training set to train the classification model and use the validation set to evaluate the classification accuracy of the model to obtain a trained model; based on the data characteristics of multi-classification small samples in this embodiment, the classification model uses the ResNet model or the U-Net model, and uses the data set to train and verify the classification model.

[0031] Specifically, the classification accuracy of the evaluation model is evaluated using basic evaluation indicators, including precision, recall, and F1 score; among them, precision can measure the proportion of instances predicted by the model as positive samples that are truly positive samples, recall can measure the model's ability to recognize positive samples, and F1 score is the harmonic mean of precision and recall, which is used to comprehensively measure the performance of the model.

[0032] Image detection: The image to be detected is acquired through a high-resolution camera, and the image to be detected is preprocessed to form a detection data set. The preprocessed image to be detected is input into the trained model for detection to obtain a recognition result image with defect recognition marks; among them, the preprocessing process of the image to be detected is consistent with the preprocessing process of the historical image.

[0033] Image stitching: Based on the image numbers generated during the image cropping process, an image stitching algorithm is used to stitch together multiple recognition result images of the same guardrail board to synthesize a complete guardrail board image. Each guardrail board image contains a complete guardrail board, which is convenient for counting defects on the same guardrail board and paving the way for subsequent work. Depending on whether there are recognition defects in the stitched image, the guardrail board image is divided into a normal image and a defective image.

[0034] Specifically, stitching is performed according to the shooting position, sequence, and cropping number. After stitching, the location of the defect is determined through visual inspection, and images without defects are classified as normal images, and images with defects are classified as defective images, so as to perform further quantitative analysis on the defective images.

[0035] Post-processing: includes quantity counting step, judgment step, repair step and conformity assessment step.

[0036] Quantity statistics: defect image per unit area The number of defects within the range is counted, and the unit area The selection is performed by setting the displacement and moving step by step from the upper left corner of the defect image. The displacement is determined based on the minimum acceptable defect area and the unit area is set The threshold of the number of defects allowed within ; Number of defects The calculation model is as follows: , in, Indicates Unit area The total number of defects in ; Indicates unit area The number of defects of type i in ; In this embodiment, the first type of defect is set as bubbles, the second type of defect is set as particles, and the third type of defect is set as sagging; according to the detection requirements, the unit area Set to , threshold Set to 3, that is, the first Unit area Total number of defects within , per unit area The maximum number of defects allowed is 3.

[0037] Judgment: When the number of defects Less than or equal to the set quantity threshold When the number of defects is Greater than the set quantity threshold When the defect image is divided into the necessary processing class images, the repair step is performed.

[0038] Qualification assessment: Guardrail images without defects are classified as normal images, and defective images with defect severity within an acceptable range are classified as images that do not require processing. Both normal images and images that do not require processing are qualified images and do not require processing, which can improve the quality and efficiency of defect detection.

[0039] Repair: Defective images whose severity exceeds the acceptable range are classified as necessary processing images. Necessary processing images are unqualified images and need to be reworked and repaired.

[0040] The implementation principle of Example 1 of the present application is: first, historical images are acquired, and a training set and a verification set are obtained through image preprocessing and data annotation. Then, the classification model is trained with the training set and the classification accuracy of the model is evaluated using the verification set to obtain a trained model; the image to be detected is input into the trained model to obtain a recognition result image, the recognition result image is spliced, and multiple recognition result images located on the same guardrail board are combined into one, and normal images and defective images are divided; finally, post-processing is performed to count the number of defects in the defective image and set a quantity threshold , the defect images are further divided into non-necessary processing images and necessary processing images, and qualified assessment and repair are carried out; not only can the defect type be identified, but also the unit area The number of defects within the scope is counted and judged, and the severity of the defects is quantitatively evaluated, which effectively improves the detection efficiency. Example

[0041] Reference Figure 2 The difference between this embodiment and embodiment 1 is that a reclassification step is provided between the judgment step and the conformity assessment step; the reclassification step includes an impact health value calculation step and a re-judgment step.

[0042] Impact health value calculation: Set different weights for different types of defects according to their impact, and calculate the unit area The impact of the number of defects on health value , and set the health value threshold ; Impact health value The calculation model is as follows: , in, Indicates The impact health value per unit area, ; It is The number of class defects, , is the corresponding weight, and the sum of the weights is .

[0043] In this embodiment, according to product characteristics and industry experience, the inside of the bubble is hollow, the coating is thin, and it is easy to be damaged; the inside of the particle contains impurities and is easily detached by external force; and the inside of the sag is a solid structure, which is usually not easy to be damaged. According to the characteristics of the above defects, the impact degree of the defect is set to bubble. Particles Flow, that is ,and , set the health value threshold , and the quantity threshold equal.

[0044] Re-judge: When the impact health value Less than or equal to the health value threshold When the impact health value is Greater than the health threshold When the image is processed in the unnecessary way, it is restored as the necessary way image and the repair step is performed.

[0045] The implementation principle of Example 2 of the present application is: the impact degree of different types of defects is graded, and different weights are set for each type of defect. Defects with a large impact degree are given a high weight, and defects with a small impact degree are given a low weight, so as to increase the impact health value of defects with a large impact degree. The proportion of influence health value data value and credibility.

[0046] Embodiment 3: Reference Figure 3 The difference between this embodiment and embodiment 2 is that a defect area statistics step and a third judgment step are provided between the re-judgment step and the conformity assessment step.

[0047] Among them, the defect area statistics are the defect area of ​​non-essential processing images. Calculate and set the defect area limit threshold ; Defect area The calculation model is as follows: , in, Indicates Unit area The total defect area within ; Indicates unit area Neidi The area of ​​the defect class, In this embodiment, Unit area Total defect area within .

[0048] The third judgment: when the defect area Less than or equal to the set defect area limit threshold When the defect area is Greater than the set defect area limit threshold , the non-necessary processing class images are divided into necessary processing class images, and the repair step is performed; Defect area limit threshold The calculation model is as follows: , in, is a proportionality coefficient, The value range is , Indicates unit area; in this embodiment, according to the detection standard of the number of defects allowed per unit area during the spraying detection process, within the range of The ridges of the ridges shall not exceed 3, and the value of P is set to , that is, per unit area Defect area within The proportion does not exceed .

[0049] Furthermore, a defect merging judgment step is provided between the third judgment step and the conformity assessment step; the defect merging judgment step is applicable to defects of the type that are easy to fall off, and the defect merging judgment includes a defect distance statistics step, a distance judgment step and a defect adjustment step.

[0050] Defect statistics: Defect spacing between defects in non-essentially processed images Calculate and set the spacing threshold , which is used to limit the distance between defects.

[0051] Specifically, the minimum distance between two defect edges is set as the defect spacing , and the defect spacing With spacing threshold for comparison.

[0052] Distance judgment: When the defect spacing Less than the set spacing threshold When the defect spacing is Greater than or equal to the set spacing threshold When the defect area is maintained unchanged and perform the conformity assessment steps; due to the acceptable defect spacing is a fixed value and does not change with other variables, so the spacing threshold is a constant representing the minimum acceptable spacing between defects.

[0053] Defect adjustment: Defect area Adjust the coefficient and update the defect area The calculation model is as follows: , in, Indicates Unit area The total defect area within ; Indicates unit area Neidi The area of ​​the defect, , , Indicates the defect spacing Less than the set spacing threshold The number of times.

[0054] In this embodiment, Unit area Total defect area within , that is, as the defect spacing Less than the set spacing threshold The number of times the defect area increases The faster the growth rate.

[0055] The implementation principle of Example 3 of the present application is: the defect area of ​​the defect image that meets the defect quantity requirement in Example 2 is calculated. The defect images are divided into unnecessary processing images and necessary processing images to screen out single defects with larger areas; and according to the defect spacing Adjust the defect for the defect that is easy to fall off and update the defect area The calculation result includes the area between two defects that are close to each other, so that the defect area More precise.

[0056] In other embodiments, different detection standards are set for different surfaces of the guardrail, which generally include an outer surface and an inner surface. The outer surface is the main surface, which is easily in direct contact with vehicles, splashing sand and stones, wind and rain, etc., and has a high detection standard; the inner surface is a surface that is not in contact during normal use, and the detection standard is lower than that of the outer surface; different quantity thresholds can be set according to specific circumstances , Health value threshold , Defect area limit threshold and spacing threshold , reduce the repair rate of guardrail panels, save costs and improve work efficiency.

[0057] In other embodiments, edge discrimination is performed on non-essential processing images, and defects on the edge of the guardrail plate are directly classified as necessary processing images regardless of the degree of impact, and rework and repair are performed. Example

[0058] Reference Figure 4 This embodiment discloses a high-speed guardrail plate spray defect detection system, the detection system comprising: Image preprocessing module: used to obtain historical images from the database and perform image preprocessing. Image preprocessing includes graying, edge detection and cropping of historical images.

[0059] Dataset production module: The input end is connected to the output end of the image preprocessing module through electrical signal connection. It is used to annotate different types of defects with the same characteristics in historical images and produce training sets and verification sets. Different attribute values ​​are assigned to different types of defects. The data set can also be expanded through upsampling and image enhancement technology.

[0060] Model training module: The input end is connected to the output end of the data set production module by electrical signals. It is used to train the classification model using the training set and evaluate the classification accuracy of the model using the validation set to obtain a trained model; the classification accuracy of the model is evaluated using basic evaluation indicators to comprehensively measure the performance of the model.

[0061] Image detection module: The input end is electrically connected to the output end of the model training module, including a trained model; it is used to obtain the image to be detected from the file system. The image to be detected can be obtained through a high-resolution camera, and the image to be detected is preprocessed to form a detection data set. The preprocessing of the detection data set is consistent with the standard of the training data set. The preprocessed image to be detected is input into the trained model for detection to obtain a recognition result image with a defect recognition mark.

[0062] Image stitching module: The input end is electrically connected to the output end of the image detection module, and is used to stitch multiple recognition result images that are combined into the same guardrail board to synthesize a complete guardrail board image, in which each guardrail board image contains a guardrail board; based on whether there are recognition defects in the stitched image, the guardrail board image is re-differentiated into a normal image and a defective image.

[0063] Post-processing module: The input end is connected to the output end of the image stitching module for electrical signal connection, and is used to perform quantitative analysis on the defects in the defective image. By counting the number of defects per unit area and setting a defect number threshold, it is judged whether the number of defects per unit area meets the requirements. When the number of defects is less than or equal to the set number threshold, the defective image is classified as a non-necessary processing image and is evaluated as a qualified image without processing. When the number of defects is greater than the set number threshold, the defective image is classified as a necessary processing image, and the repair steps are executed for rework and repair.

[0064] Furthermore, by setting the impact health value, different weights are set for different types of defects, and the impact health value of the number of defects per unit area is counted, and a health value threshold is set; when the impact health value is less than or equal to the health value threshold, the non-essential processing class image continues to be stored as a non-essential processing class image, and is evaluated as a qualified image without processing; when the impact health value is greater than the health value threshold, the non-essential processing class image is re-stored as a necessary processing class image, and the repair steps are executed for rework and repair.

[0065] Furthermore, by counting the defect area within a unit area and setting a defect area limit threshold, when the defect area is less than or equal to the set defect area limit threshold, the non-essential processing image will continue to be stored as a non-essential processing image and will be evaluated as a qualified image without processing; when the defect area is greater than the set defect area limit threshold, the non-essential processing image will be classified as a necessary processing image, and the repair steps will be executed for rework and repair.

[0066] Furthermore, for defects that are easy to fall off, the defect spacing between defects in non-essential processing images is calculated and a spacing threshold is set; when the defect spacing is greater than or equal to the set spacing threshold, the defect area is kept unchanged and is evaluated as a qualified image without processing; when the defect spacing is less than the set spacing threshold, the defect adjustment step is executed, and the defect area is judged on the adjusted area.

[0067] The above are all preferred embodiments of the present application, and the protection scope of the present application is not limited thereto. Therefore, any equivalent changes made according to the structure, shape, and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for detecting defects in spray-coated high-speed guardrails, characterized in that: The following steps are involved: Image preprocessing: acquiring historical images and performing image preprocessing, wherein the historical images contain different types of defects with the same characteristics; Dataset preparation: annotating different types of defects with the same characteristics in the historical images and preparing training sets and verification sets; Model training: using the training set to train the classification model, and using the validation set to evaluate the classification accuracy of the model to obtain a trained model; Image detection: Obtain the image to be detected, input the image to be detected into the trained model for detection, and obtain the recognition result image with defect recognition marks; Image stitching: stitching multiple recognition result images that are combined into the same guardrail board to synthesize a complete guardrail board image, each guardrail board image contains a guardrail board; according to the recognition result, re-differentiate the guardrail board image into a normal image and a defective image; Post-processing: including quantity counting step, judgment step, repair step and conformity assessment step; Quantity statistics: per unit area The defect images within the range are counted for defect quantity and the unit area is set The threshold of the number of defects allowed within ; Number of defects The calculation model is as follows: = ， in, Indicates Unit area The total number of defects in ; Indicates unit area The number of defects of type i in ; Judgment: When the number of defects Less than or equal to the set quantity threshold When the number of defects is Greater than the set quantity threshold , dividing the defective image into necessary processing images and performing a repair step; Conformity assessment: the normal images and the non-necessary processing images are qualified images and do not need to be processed; Repair: The images that need to be processed are unqualified images and need to be reworked and repaired.

2. The method for detecting defects of high-speed guardrail panels sprayed with plastic according to claim 1, characterized in that: There is also a reclassification step between the judgment step and the conformity assessment step; Reclassification: including the impact health value calculation step and the re-judgment step; Impact health value calculation: Set different weights for different types of defects according to their impact, and calculate the unit area The impact of the number of defects on the health value W, and set the health value threshold ; Impact health value The calculation model is as follows: ， in, Indicates The impact health value per unit area, ; It is The number of class defects, , is the corresponding weight, and the sum of the weights is ; Re-judge: When the impact health value is Less than or equal to the health value threshold When the impact health value is Greater than the health threshold When the non-essentially processed image is stored again as an essential processed image, the repair step is performed.

3. The method for detecting defects of high-speed guardrail spraying according to claim 2 is characterized in that: Between the re-judgment step and the conformity assessment step, there are also a defect area statistics step and a third judgment step; Defect area statistics: Defect area of ​​non-essentially processed images Calculate and set the defect area limit threshold ; Defect area The calculation model is as follows: ， in, Indicates Unit area The total defect area within ; Indicates unit area The area of ​​the i-th defect, ; The third judgment: when the defect area Less than or equal to the set defect area limit threshold When the defect area is Greater than the set defect area limit threshold , the non-necessary processing images are classified into necessary processing images, and the repair step is performed; Defect area limit threshold The calculation model is as follows: ， in, is a proportionality coefficient, The value range is , Indicates unit area.

4. The method for detecting defects of high-speed guardrail spraying according to claim 3 is characterized in that: Between the third judgment step and the conformity assessment step, there is also a defect merging judgment step; Defect merging and identification: including defect distance statistical steps, distance judgment steps and defect adjustment steps; Defect statistics: Defect spacing between defects in non-essentially processed images Calculate and set the spacing threshold ; Distance judgment: When the defect spacing Less than the set spacing threshold When the defect adjustment step is executed; When the defect spacing Greater than or equal to the set spacing threshold When the defect area is maintained unchanged and carry out conformity assessment steps; Spacing Threshold is a constant representing the minimum spacing between defects; Defect adjustment: Defect area Adjust the coefficient and update the defect area The calculation model is as follows: ， in, Indicates Unit area The total defect area within , Indicates unit area The area of ​​the i-th defect, , Indicates the defect spacing Less than the set spacing threshold The number of times.

5. The method for detecting defects of plastic spraying of high-speed guardrails according to any one of claims 1 to 4, characterized in that: Set different quantity thresholds for the outer and inner surfaces of the guardrail , Health value threshold , Defect area limit threshold and spacing threshold ; wherein the outer surface is the main surface that can be directly seen during normal use, the inner surface is the surface that cannot be observed during normal use, and the tolerance of the inner surface to defects is higher than the tolerance of the outer surface to defects.

6. The method for detecting defects of high-speed guardrail spraying according to claim 4 is characterized in that: Between the distance judgment step and the conformity assessment step, an edge judgment step and a fourth judgment step are also provided; Edge discrimination step: performing edge discrimination on defects on the non-essentially processed images to determine whether the defects are located at the edge of the guardrail plate; The fourth judgment step: when the defect is not at the edge of the guardrail board, the non-essential processing image continues to be stored as a non-essential processing image, and the qualification assessment step is performed; when the defect is located at the edge of the guardrail board, the non-essential processing image is directly classified as a necessary processing image, and the repair step is performed.

7. A high-speed guardrail spray defect detection system, the detection system is applicable to the detection method according to any one of claims 1 to 6, characterized in that: include: Image preprocessing module: used to obtain historical images and perform image preprocessing; Dataset creation module: used to annotate different types of defects with the same characteristics in historical images and create training sets and verification sets; Model training module: used to train the classification model using the training set and evaluate the classification accuracy of the model using the validation set to obtain a trained model; Image detection module: used to obtain the image to be detected, input the image to be detected into the trained model for detection, and obtain the recognition result image; Image stitching module: used to stitch multiple recognition result images of the same guardrail board to synthesize a complete guardrail board image, where each guardrail board image contains one guardrail board; based on the recognition results, the guardrail board image is divided into a normal image and a defective image; Post-processing module: used to perform quantitative analysis on the defects in the defect image, by counting the number of defects per unit area and setting a threshold for the number of defects, to determine whether the number of defects per unit area meets the requirements, and to further divide the defect image into non-essential processing images and necessary processing images.

Images