Road defect detection method and system based on big data

Abstract

The invention relates to the technical field of road crack classification, in particular to a road defect detection method and system based on big data. The method comprises the following steps: firstly, according to a space proportion of a crack region corresponding to each crack category in a historical database, adjusting an initial attribute distinction degree, obtaining a detection distinction degree of a to-be-selected attribute, screening all crack attributes, and obtaining a distinguishing attribute; based on the difference condition of the distinguishing attributes of the to-be-selected crack and the crack region in the crack category, obtaining the category representation degree of the to-be-selected crack in the crack category to which the to-be-selected crack belongs; on the basis of the category representation degree, representative cracks of the crack category are screened out from all the crack areas in the crack category; and classifying the to-be-detected crack based on the difference between the to-be-detected crack and all the distinguishing attributes representing the crack in the crack category. According to the method, the distinguishing effect of the crack attribute on the to-be-detected crack image and the characterization capability of the crack region in the crack type are fully considered, so that the to-be-detected crack is accurately classified.

Description

Technical Field

[0001] The present invention relates to the technical field of road crack classification, and in particular to a road defect detection method and system based on big data. Background Art

[0002] Through road defect detection, problems with the road can be discovered in time, so that corresponding repair measures can be taken to ensure road safety and increase road service life. Road cracks are the initial manifestation of road damage, indicating that the road structure has been damaged to a certain extent. If the cracks are not repaired in time, it will affect the service life of the road and even affect road safety. Since different types of cracks may be caused by different reasons, different repair measures need to be taken. Identifying the type of road cracks is crucial for road crack detection.

[0003] When identifying the crack type in road cracks, the prior art compares the crack to be identified with all crack attributes of cracks of the crack type in the database in sequence to determine the crack type of the crack to be identified. However, when identifying the crack type, the prior art does not fully consider the effect of the crack attributes on the crack image to be identified and the crack characterization capability in the crack type, which makes it difficult to accurately identify the crack type of the crack to be detected. Summary of the invention

[0004] In order to solve the technical problem that it is difficult to accurately classify cracks to be detected in the prior art, the purpose of the present invention is to provide a road defect detection method and system based on big data. The technical solution adopted is as follows:

[0005] A road defect detection method based on big data, the method comprising:

[0006] Obtain the crack areas contained in all historical crack images in the historical database as well as their crack attributes and crack categories; obtain the crack areas contained in the crack images to be detected as well as their crack attributes;

[0007] Taking any crack attribute as a candidate attribute, in the historical database, according to the difference between the candidate attributes in the crack area and the crack category in the crack area, obtaining the initial attribute discrimination of the candidate attribute for each crack category; adjusting the initial attribute discrimination according to the spatial proportion of the crack area corresponding to each crack category in the historical database, obtaining the detection discrimination of the candidate attribute; using the detection discrimination to screen all crack attributes, obtaining the discrimination attribute;

[0008] Taking any crack area in the historical database as a candidate crack, based on the difference between the distinguishing attributes of the candidate crack and the crack area in the crack category to which it belongs, and the detection discrimination of the distinguishing attributes, the category representativeness of the candidate crack in the crack category to which it belongs is obtained; based on the category representativeness, the representative crack of each crack category is screened out from all crack areas of each crack category;

[0009] Each crack region in the crack image to be detected is taken as a crack to be detected, and the crack to be detected is classified based on the difference between the crack to be detected and all distinguishing attributes representing the cracks in the crack category.

[0010] Furthermore, the method for obtaining the initial attribute discrimination degree includes:

[0011] Clustering all crack areas in the historical database based on the attributes to be selected, and obtaining clustering results of the attributes to be selected;

[0012] For any crack category, the initial attribute discrimination of the candidate attribute to the crack category is obtained according to the distribution concentration of the crack category corresponding to the crack area in the clustering result of the candidate attribute.

[0013] Furthermore, the method of obtaining the initial attribute discrimination degree of the candidate attribute to the crack category according to the distribution concentration of the crack category corresponding to the crack area in the clustering result of the candidate attribute includes:

[0014] The crack area corresponding to any crack category is taken as the target crack; the negative correlation mapping result of the total number of all clusters with the target cracks is taken as the first set parameter;

[0015] Count the total number of all target cracks in each cluster, and use the largest total number as the parameter in the second set;

[0016] The product of the first set of parameters and the second set of parameters is calculated to obtain the initial attribute discrimination of the candidate attribute for each crack category.

[0017] Furthermore, the method for obtaining the detection discrimination includes:

[0018] According to the spatial proportion of the crack area corresponding to the crack category in the historical database, the detection attention weight of the crack category is obtained;

[0019] The detection attention weight of each crack category is used to weight the initial attribute discrimination degree corresponding to each crack category to obtain the detection discrimination degree of the attribute to be selected.

[0020] Furthermore, the method for acquiring the detection attention weight includes:

[0021] In the historical database, the total area of ​​all crack regions corresponding to the crack category and the total number of all crack regions corresponding to the crack category are forward fused to obtain the detection attention weight of the crack category.

[0022] Furthermore, the method of weighting the initial attribute discrimination of each crack category by using the detection attention weight of each crack category to obtain the detection discrimination of the to-be-selected attribute includes:

[0023] The detection attention weight of each crack category is used to perform weighted summation and normalization on the initial attribute discrimination corresponding to each crack category to obtain the detection discrimination of the selected attribute.

[0024] Furthermore, the method for obtaining the category representativeness includes:

[0025] The crack category to which the selected crack belongs is taken as the target category, and all crack categories except the target category are taken as reference categories;

[0026] The category representativeness is obtained according to a category representativeness formula, wherein the category representativeness formula includes:

[0027] R is the representative degree of the candidate crack in the crack category to which it belongs; D a is the detection discrimination of the ath distinguishing attribute; d e,a is the absolute value of the difference between the e-th crack region and the a-th distinguishing attribute corresponding to the crack to be selected in the target category; d n,a is the absolute value of the numerical difference between the nth crack area and the ath distinguishing attribute corresponding to the selected crack in all the reference categories; E is the total number of all crack areas in the target category; N is the total number of all crack areas in all the reference categories; A is the total number of all the distinguishing attributes; ε is the preset denominator adjustment value.

[0028] Furthermore, the method for classifying cracks to be detected includes:

[0029] Based on the differences between all distinguishing attributes of the crack to be detected and all representative cracks in the crack category and the category representativeness of the representative cracks, a difference index between the crack to be detected and the crack category is obtained;

[0030] The crack category corresponding to the smallest difference index is used as the crack category of the crack to be detected.

[0031] Furthermore, the formula for obtaining the difference index includes:

[0032] P i is the difference index between the crack to be detected and the crack category i; Δrb,q is the absolute value of the difference between the crack to be detected and the bth representative crack in the crack category i corresponding to the qth distinguishing attribute; R b is the category representativeness of the bth representative crack in crack category i; Q is the total number of distinguishing attributes; B is the total number of representative cracks.

[0033] The present invention proposes a road defect detection system based on big data, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the road defect detection method based on big data are implemented.

[0034] The present invention has the following beneficial effects:

[0035] First, considering that the attributes to be selected have different importance in distinguishing different types of crack areas, the initial attribute discrimination is first used to preliminarily reflect the distinguishing effect of the attributes to be selected on the crack categories. In order to accurately and efficiently classify the cracks in the crack image to be detected in the future, it is necessary to screen out the crack attributes with good distinguishing effect for the crack image to be detected. According to the spatial proportion of the crack area corresponding to each crack category in the historical database, the importance of each crack category to be detected in the crack image to be detected is indirectly analyzed, so as to adjust the initial attribute discrimination and obtain the detection discrimination of the attributes to be selected. The detection discrimination reflects the distinguishing effect of the attributes to be selected on the crack image to be detected. All crack attributes are screened by using the detection discrimination to obtain the distinguishing attributes. The distinguishing attributes reflect the crack attributes with good distinguishing effect for the crack image to be detected. In order to accurately and efficiently classify the cracks in the crack image to be detected in the future, it is necessary to screen out the cracks with good representativeness in the crack category. The category representativeness is used to reflect the representative ability of the crack to be selected for the crack category to which it belongs. The larger the value of the category representativeness, the stronger the representative ability of the attribute to be selected for the crack category to which it belongs. Representative cracks of the crack category are screened out from all crack areas in the crack category. The representative cracks reflect the crack regions with better representative ability in the crack category. Based on the differences in all distinguishing attributes between the cracks to be detected and the representative cracks in the crack category, the cracks to be detected are classified to achieve accurate classification of the cracks to be detected. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0037] Figure 1 A flowchart of a road defect detection method based on big data provided by an embodiment of the present invention;

[0038] Figure 2 A flow chart of a method for obtaining initial attribute discrimination provided by an embodiment of the present invention;

[0039] Figure 3 A flow chart of a method for obtaining detection discrimination provided by an embodiment of the present invention;

[0040] Figure 4 A flow chart of a method for classifying cracks to be detected provided by an embodiment of the present invention;

[0041] Figure 5 A structural diagram of a road defect detection system based on big data provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0042] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the road defect detection method and system based on big data proposed by the present invention, its specific implementation, structure, features and effects, in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0043] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0044] The following is a detailed description of a road defect detection method and system based on big data provided by the present invention in conjunction with the accompanying drawings.

[0045] The present invention provides a road defect detection method and system based on big data. Figure 1 , which shows a flow chart of a road defect detection method based on big data provided by an embodiment of the present invention, the method comprising the following steps:

[0046] Step S1: obtaining the crack areas contained in all historical crack images in the historical database and their crack attributes and crack categories; obtaining the crack areas contained in the crack image to be detected and their crack attributes.

[0047] The purpose of the present invention is to classify cracks to be detected, which first requires data acquisition.

[0048] All historical crack images are obtained from the historical database and the crack images to be detected are obtained from the road defect detection system. The specific acquisition steps include using a high-definition camera to select a suitable acquisition position and select a suitable angle for image acquisition to obtain the original image. In order to ensure the image quality of the subsequent image processing process, each original image needs to be subjected to image preprocessing operations to obtain each crack image. The crack images stored in the historical database are used as historical crack images, and the crack images that need to be detected are used as crack images to be detected.

[0049] In order to simplify the subsequent identification of the categories corresponding to the cracks in the crack image to be detected, it is necessary to identify the crack areas in the historical crack images and the crack images to be detected and their crack properties. Specifically, the CNN neural network is used to obtain the crack areas contained in all crack images and the crack properties of the crack areas, that is, to obtain the crack areas in the historical crack images and the crack images to be detected and their crack properties. Using the CNN neural network, the crack image can be input into the trained CNN neural network model, and the crack areas in the crack image and the crack properties of the crack areas are output. The specific CNN neural network model training process is a technical means well known to those skilled in the art, and will not be elaborated here. It should be noted that the CNN neural network is a prior art well known to those skilled in the art, and is only briefly described here. Since the CNN neural network is a deep learning model, it is particularly suitable for image recognition tasks.

[0050] Among them, the crack attributes include crack length, crack width, crack grayscale, crack shape, crack continuity and crack direction. Each crack attribute is segmented using the JieBa word segmentation tool to improve the accuracy of the information. Since the CNN neural network records the information of crack attributes through text, in order to facilitate computer processing, the text information of the crack attributes is first converted into digital information and uploaded to the database. For example, taking crack continuity as an example, the corresponding code for crack continuity is 100, and the crack continuity has four conditions: obviously continuous, relatively continuous, relatively discontinuous, and discontinuous, and the corresponding codes are 1000, 0100, 0010, and 0001 respectively. For example, the crack continuity of a certain crack is relatively continuous, and the corresponding code is 1000100, and its corresponding decimal number is 68. Subsequent values ​​are all expressed in decimal.

[0051] The main purpose of the present invention is to identify the crack area contained in the crack image to be detected based on the crack category of the crack area in the historical database. First, the crack category of the crack area in the historical database needs to be determined. Specifically, all crack areas of all historical crack images are manually checked and the crack categories of the crack areas are marked, such as slip cracks, arch cracks, fatigue cracks, and transmission rod failure cracks. It should be noted that the historical crack images in the historical database and the crack images to be detected in the present invention are for the same specific road.

[0052] In the embodiment of the present invention, the image preprocessing operation includes graying, noise reduction, and contrast enhancement. The embodiment of the present invention uses the weighted average method for graying, uses Gaussian filtering for noise reduction, and uses histogram equalization for contrast enhancement, which can be set by the implementer according to the actual situation. It should be noted that, in order to facilitate the calculation, all the crack attributes involved in the calculation in the embodiment of the present invention are subjected to data preprocessing to eliminate the dimension effect. The specific means of removing the dimension effect are technical means well known to those skilled in the art and are not limited here.

[0053] Step S2: Take any crack attribute as a candidate attribute, and in the historical database, obtain the initial attribute discrimination of the candidate attribute for each crack category based on the differences between the candidate attributes in the crack area and the crack category in the crack area; adjust the initial attribute discrimination based on the spatial proportion of the crack area corresponding to each crack category in the historical database to obtain the detection discrimination of the candidate attribute; use the detection discrimination to screen all crack attributes to obtain the discrimination attribute.

[0054] Take any crack attribute as a candidate attribute. First, considering that the candidate attribute has different importance in distinguishing different types of crack areas, the initial attribute discrimination is first used to preliminarily reflect the distinguishing effect of the candidate attribute on the crack category. In order to accurately and efficiently classify the cracks in the crack image to be detected in the future, it is necessary to screen out the crack attributes with good distinguishing effect on the crack image to be detected. According to the spatial proportion of the crack area corresponding to each crack category in the historical database, the importance of each crack category in the crack image to be detected is indirectly analyzed, so as to adjust the initial attribute discrimination and obtain the detection discrimination of the candidate attribute. The detection discrimination reflects the distinguishing effect of the candidate attribute on the crack image to be detected. All crack attributes are screened using the detection discrimination to obtain the distinguishing attribute. The distinguishing attribute reflects the crack attribute with good distinguishing effect on the crack image to be detected.

[0055] Considering that the candidate attribute may show high discrimination in some crack categories, but weak discrimination effect in other categories, it is necessary to measure the discrimination effect of the candidate attribute on the crack category. Considering that when clustering under the candidate attribute, a certain crack category is divided into multiple clusters and the proportion of crack areas corresponding to the crack category in the cluster is small, it means that the crack attribute has a weak discrimination effect on the crack category. The initial attribute discrimination is constructed to preliminarily reflect the discrimination effect of the candidate attribute on the crack category. Please refer to Figure 2 , which shows a flow chart of a method for obtaining an initial attribute discrimination degree in one embodiment of the present invention. Preferably, in one embodiment of the present invention, the method for obtaining the initial attribute discrimination degree includes:

[0056] Step S201: clustering all crack regions in the historical database based on the attributes to be selected, and obtaining clustering results of the attributes to be selected.

[0057] By using the candidate attributes to divide the crack areas in the historical database, a clustering result of the candidate attributes is formed, and the clustering result reflects the distinction of the crack areas by the candidate attributes.

[0058] In one embodiment of the present invention, the absolute value of the numerical difference between each two crack regions corresponding to the selected attribute is used as the distance measurement value of each two crack regions, and the k-means clustering algorithm is used to cluster all the crack regions in the historical database to obtain the clustering result of the selected attribute. It should be noted that the k-means clustering algorithm is a technical means well known to those skilled in the art, and will not be described in detail here. Only a brief description of the process of using the k-means clustering algorithm in one embodiment of the present invention to cluster all the crack regions in the historical database according to the distance measurement value of each two crack regions to obtain the clustering result is given:

[0059] Considering that the larger the distance metric value between two crack regions is, the larger the difference in the selected attributes is, the total number of all crack categories is taken as the k value, and the distance metric value between each two crack regions is substituted into the clustering process, and all crack regions in the historical database are clustered to obtain clustering results, that is, to obtain various cluster clusters. The difference in the selected attributes of crack regions in different cluster clusters is made as large as possible, and the difference in the selected attributes of crack regions in the same cluster cluster is made as small as possible. The similarity of the selected attributes of crack regions in each cluster cluster is made as large as possible. In other embodiments of the present invention, the optimal k value of the k-means clustering algorithm can be determined by other methods such as the silhouette coefficient method, which is not limited here.

[0060] Step S202: for any crack category, according to the distribution concentration of the crack category corresponding to the crack area in the clustering result of the candidate attribute, the initial attribute discrimination degree of the candidate attribute to the crack category is obtained.

[0061] The initial attribute discrimination degree is used to preliminarily reflect the discrimination effect of the candidate attribute on the crack category. The larger the value of the initial attribute discrimination degree is, the stronger the discrimination effect of the candidate attribute on the crack category is.

[0062] In one embodiment of the present invention, the fracture area corresponding to any fracture category is taken as the target fracture; the negative correlation mapping result of the total number of all clusters with target fractures is taken as the first centralized parameter; the total number of all target fractures contained in each cluster is counted, and the maximum total number is taken as the second centralized parameter; the product of the first centralized parameter and the second centralized parameter is calculated to obtain the initial attribute discrimination degree of the selected attribute for the fracture category. It should be noted that the negative correlation mapping is a technical means well known to those skilled in the art, and the negative correlation mapping can be in the form of inverse proportion or negative exponential power without limitation.

[0063] For the above steps, the parameters in the first set reflect the distribution and concentration of the target cracks in multiple clusters. The smaller the parameters in the first set, the more clusters the target cracks are scattered to, indicating that the selected attribute has a weaker effect on distinguishing the crack category. The total number of all target cracks in each cluster is counted, and the largest total number reflects the total number of target cracks in the clusters where the target cracks are mainly concentrated. The larger the parameters in the second set, the more the target cracks are mainly concentrated in a certain cluster, indicating that the selected attribute has a stronger distinguishing effect on the crack category. Finally, the product of the parameters in the first set and the parameters in the second set is calculated to obtain the initial attribute discrimination of the crack category by the selected attribute. The larger the value of the initial attribute discrimination, the stronger the distinguishing effect of the selected attribute on the crack category.

[0064] In order to select crack attributes with good distinguishing effect on crack images to be detected, a detection discrimination index is constructed to reflect the distinguishing effect of the selected attributes on crack images to be detected. Figure 3 , which shows a flow chart of a method for obtaining detection discrimination in one embodiment of the present invention. Preferably, in one embodiment of the present invention, the method for obtaining detection discrimination includes:

[0065] Step S211: Obtain the detection attention weight of the crack category according to the spatial proportion of the crack area corresponding to the crack category in the historical database.

[0066] Considering that the historical crack images in the historical database and the crack images to be detected reflect the crack conditions of the same specific road, the more times and the larger the area of ​​the crack area corresponding to a certain crack category in the historical database, the more serious the impact of this crack category on this specific road, and the more attention should be paid to this crack category in the crack images to be detected, and the detection attention weight of the crack category is constructed. The larger the detection attention weight, the more attention should be paid to the crack category.

[0067] In one embodiment of the present invention, a method for acquiring a detection attention weight includes:

[0068] In the historical database, the total area of ​​all crack regions corresponding to the crack category and the total number of all crack regions corresponding to the crack category are forward fused to obtain the detection attention weight of the crack category. Specifically, in the historical database, the product of the total area of ​​all crack regions corresponding to the crack category and the total number of all crack regions corresponding to the crack category is calculated and normalized to obtain the detection attention weight of the crack category. It should be noted that weighted summation and normalization are both technical means well known to those skilled in the art, and normalization can be performed by linear normalization, etc., which is not limited here.

[0069] Step S212: weighting the initial attribute discrimination degree corresponding to each crack category by using the detection attention weight of each crack category to obtain the detection discrimination degree of the attribute to be selected.

[0070] Taking into account the initial attribute discrimination to preliminarily reflect the discrimination effect of the candidate attribute on the crack category, the detection attention weight reflects the need to focus on the crack category in the crack image to be detected. The initial attribute discrimination of each crack category is weighted by the detection attention weight of each crack category to obtain the detection discrimination of the candidate attribute. The detection discrimination reflects the discrimination effect of the candidate attribute for the crack image to be detected. The larger the detection discrimination value, the stronger the discrimination effect of the candidate attribute for the crack image to be detected.

[0071] In one embodiment of the present invention, the detection discrimination method includes: using the detection attention weight of each crack category to perform weighted summation and normalization processing on the initial attribute discrimination corresponding to each crack category, and obtain the detection discrimination of the selected attribute. It should be noted that weighted summation and normalization processing are both technical means well known to those skilled in the art, and the normalization processing can be linear normalization, etc., which is not limited here. Among them, the initial attribute discrimination corresponding to the crack category is the initial attribute discrimination of the selected attribute to the crack category.

[0072] For the above steps, in order to evaluate the distinguishing effect of the selected attributes when detecting crack images, the importance of each crack category in the crack image to be detected is indirectly analyzed according to the spatial proportion of the crack area corresponding to each crack category in the historical database. The crack category that needs more attention is given a greater weight. Therefore, the initial attribute discrimination of each crack category is weightedly summed and normalized using the detection attention weight of each crack category to obtain the detection discrimination of the selected attribute. The larger the detection discrimination value, the better the distinguishing effect of the selected attribute on the crack image to be detected.

[0073] Considering that the larger the detection discrimination value is, the better the discrimination effect of the candidate attribute on the crack image to be detected is, the distinguishing attribute is constructed to reflect the crack attribute with better discrimination effect on the crack image to be detected. Preferably, in one embodiment of the present invention, the method for obtaining the distinguishing attribute includes:

[0074] The crack attribute with a detection discrimination greater than the preset attribute threshold is used as the distinguishing attribute. In one embodiment of the present invention, the preset attribute threshold is 0.51, and the implementer can set it according to the implementation scenario.

[0075] Step S3: Take any crack area in the historical database as a candidate crack, and obtain the category representativeness of the candidate crack in the crack category to which it belongs based on the difference in distinguishing attributes between the candidate crack and the crack area in the crack category to which it belongs, as well as the detection discrimination of the distinguishing attributes; based on the category representativeness, select representative cracks of each crack category from all crack areas of each crack category.

[0076] In order to accurately and efficiently classify the cracks in the crack image to be detected, it is necessary to screen out the cracks with better representativeness in the crack category. The category representativeness is used to reflect the representativeness of the candidate crack to the crack category to which it belongs. The larger the category representativeness value, the stronger the representativeness of the candidate attribute to the crack category to which it belongs. The representative cracks of the crack category are screened out from all the crack areas in the crack category. The representative cracks reflect the crack areas with better representativeness in the crack category.

[0077] Considering that the higher the similarity between the candidate fracture and the fracture region of the fracture category to which it belongs and the higher the difference between the candidate fracture and the fracture region of other categories, the stronger the representative ability of the candidate fracture to the fracture category to which it belongs is, the category representativeness is used to reflect the representativeness of the candidate fracture to the fracture category to which it belongs. Preferably, in one embodiment of the present invention, the method for obtaining the category representativeness includes:

[0078] The crack category to which the selected crack belongs is taken as the target category, and all crack categories except the target category are taken as reference categories;

[0079] The category representativeness is obtained according to the category representativeness formula. The category representativeness formula includes:

[0080] R is the representative degree of the candidate crack in the crack category to which it belongs; D a is the detection discrimination of the a-th distinguishing attribute; d e,a is the absolute value of the difference between the e-th crack region and the a-th distinguishing attribute corresponding to the selected crack in the target category; d n,ais the absolute value of the difference between the nth crack region and the ath distinguishing attribute corresponding to the selected crack in all reference categories; E is the total number of all crack regions in the target category; N is the total number of all crack regions in all reference categories; A is the total number of all distinguishing attributes; ε is the preset denominator adjustment value. In one embodiment of the present invention, the preset denominator adjustment value is 0.001 to prevent the denominator from being 0.

[0081] In the formula, It reflects the similarity between the candidate crack and its category in the ath distinguishing attribute. The larger the value, the greater the similarity. It reflects the difference between the candidate crack and other categories in the a-th distinguishing attribute. The larger the value, the greater the difference. a It reflects the distinguishing effect of the a-th distinguishing attribute on the crack image to be detected. The category representativeness is finally obtained by weighting. The larger the category representativeness value, the greater the similarity between the candidate crack and its category and the greater the difference between it and other categories, indicating that the candidate crack has a stronger representative ability for the crack category to which it belongs. Similarly, the category representativeness of each crack area in the historical database relative to the crack category to which it belongs can be obtained. According to the category representativeness, the representative cracks of each crack category can be screened out from all crack areas in the crack category.

[0082] In order to determine a fracture region with better representativeness among fracture categories, preferably, in one embodiment of the present invention, a method for obtaining representative fractures includes:

[0083] In a crack category, all crack regions in the crack category are sorted in descending order of category representativeness, and the first preset number of crack regions are used as representative cracks of the crack category. In one embodiment of the present invention, the preset number of regions is 5, which can be set by the implementer according to the implementation scenario.

[0084] Step S4: taking each crack region in the crack-to-be-detected image as a crack-to-be-detected, and classifying the cracks to be detected based on the differences between the cracks to be detected and all distinguishing attributes representing cracks in the crack category.

[0085] Through the above steps, the distinguishing attributes with good distinguishing effect on the crack image to be detected and the representative cracks with good representative ability in the crack category are classified based on the differences between all distinguishing attributes of the crack to be detected and the representative cracks in the crack category, so as to achieve accurate classification of the cracks to be detected.

[0086] See also Figure 4, which shows a flow chart of a method for classifying cracks to be detected provided by an embodiment of the present invention. Preferably, in an embodiment of the present invention, the method for classifying cracks to be detected includes:

[0087] Step S401: based on the differences between the crack to be detected and all the distinguishing attributes of all the representative cracks in the crack category and the category representativeness of the representative cracks, a difference index between the crack to be detected and the crack category is obtained.

[0088] The difference index reflects the difference between the crack to be detected and the crack category. The larger the value of the difference index is, the greater the difference between the crack to be detected and the crack category is.

[0089] Preferably, in one embodiment of the present invention, the formula for obtaining the difference index includes:

[0090] The difference index is obtained according to the difference index formula, which includes:

[0091] P i is the difference index between the crack to be detected and the crack category i; Δr b,q is the absolute value of the difference between the crack to be detected and the bth representative crack in the crack category i corresponding to the qth distinguishing attribute; R b is the category representativeness of the bth representative crack in crack category i; Q is the total number of distinguishing attributes; B is the total number of representative cracks.

[0092] In the formula, Δr b,q The local part reflects the difference between the crack to be detected and the bth representative crack in the crack category i at the qth distinguishing attribute. b Reflects the representative ability of the bth representative crack, using R b For Δr b,q The difference index value is obtained by weighting, and the difference index value comprehensively reflects the difference between the crack to be detected and the crack category i. The larger the difference index value is, the greater the difference between the crack to be detected and the crack category is.

[0093] Step S402: taking the crack category corresponding to the smallest difference index as the crack category of the crack to be detected.

[0094] Considering that the smaller the difference index is, the higher the similarity between the crack to be detected and the crack category is, the crack category corresponding to the smallest difference index is taken as the crack category of the crack to be detected. The crack category can accurately reflect the crack category to which the crack to be detected belongs.

[0095] The present invention also proposes a road defect detection system based on big data, please refer to Figure 5, which shows a structure diagram of a road defect detection system based on big data provided by an embodiment of the present invention, the system includes: a data acquisition module 101, a distinguishing attribute analysis module 102, a representative crack analysis module 103 and a classification module 104.

[0096] The data acquisition module 101 is used to acquire the crack areas contained in all historical crack images in the historical database and their crack attributes and crack categories; and to acquire the crack areas contained in the crack images to be detected and their crack attributes.

[0097] The distinguishing attribute analysis module 102 is used to take any crack attribute as a candidate attribute, and obtain the initial attribute discrimination of the candidate attribute for each crack category in the historical database based on the differences between the candidate attributes in the crack area and the crack category in the crack area; adjust the initial attribute discrimination based on the spatial proportion of the crack area corresponding to each crack category in the historical database to obtain the detection discrimination of the candidate attribute; use the detection discrimination to screen all crack attributes to obtain the distinguishing attribute.

[0098] The representative crack analysis module 103 is used to take any crack area in the historical database as a candidate crack, and obtain the category representativeness of the candidate crack in the crack category to which it belongs based on the difference in the distinguishing attributes between the candidate crack and the crack area in the crack category to which it belongs, as well as the detection discrimination of the distinguishing attributes; based on the category representativeness, the representative crack of each crack category is screened out from all crack areas of each crack category.

[0099] The classification module 104 is used to take each crack region in the crack image to be detected as a crack to be detected, and classify the crack to be detected based on the difference between the crack to be detected and all distinguishing attributes representing the cracks in the crack category.

[0100] It should be noted that the system provided in the above embodiment is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer device is divided into different functional modules to complete all or part of the functions described above. In addition, the road defect detection system based on big data and the road defect detection method based on big data provided in the above embodiment belong to the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0101] In summary, the embodiment of the present invention provides a road defect detection method and system based on big data. First, the initial attribute discrimination is adjusted according to the spatial proportion of the crack area corresponding to each crack category in the historical database to obtain the detection discrimination of the attribute to be selected; all crack attributes are screened using the detection discrimination to obtain the distinguishing attribute; based on the difference in the distinguishing attribute between the crack to be selected and the crack area in the crack category, and the detection discrimination of the distinguishing attribute, the category representativeness of the crack to be selected in the crack category to which it belongs is obtained; based on the category representativeness, representative cracks of the crack category are screened from all crack areas in the crack category; based on the difference in all distinguishing attributes between the crack to be detected and the representative cracks in the crack category, the cracks to be detected are classified. In the embodiment of the present invention, the effect of crack attributes on the distinction of the crack image to be detected and the characterization ability of the crack area in the crack type are fully considered to achieve accurate classification of the cracks to be detected.

[0102] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0103] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A road defect detection method based on big data, characterized in that: The method comprises: Obtain the crack areas and crack attributes and crack categories contained in all historical crack images in the historical database; obtain the crack areas and crack attributes contained in the crack images to be detected; Taking any crack attribute as a candidate attribute, in the historical database, according to the difference between the candidate attributes in the crack area and the crack category in the crack area, obtaining the initial attribute discrimination of the candidate attribute for each crack category; adjusting the initial attribute discrimination according to the spatial proportion of the crack area corresponding to each crack category in the historical database, obtaining the detection discrimination of the candidate attribute; using the detection discrimination to screen all crack attributes, obtaining the discrimination attribute; Taking any crack area in the historical database as a candidate crack, based on the difference between the distinguishing attributes of the candidate crack and the crack area in the crack category to which it belongs, and the detection discrimination of the distinguishing attributes, the category representativeness of the candidate crack in the crack category to which it belongs is obtained; based on the category representativeness, the representative crack of each crack category is screened out from all crack areas of each crack category; Each crack region in the crack image to be detected is taken as a crack to be detected, and the crack to be detected is classified based on the difference between the crack to be detected and all distinguishing attributes representing the cracks in the crack category.

2. According to the big data-based road defect detection method of claim 1, it is characterized in that: The method for obtaining the initial attribute discrimination degree includes: Clustering all crack areas in the historical database based on the attributes to be selected, and obtaining clustering results of the attributes to be selected; For any crack category, the initial attribute discrimination of the candidate attribute to the crack category is obtained according to the distribution concentration of the crack category corresponding to the crack area in the clustering result of the candidate attribute.

3. According to claim 2, a road defect detection method based on big data is characterized in that: The method for obtaining the initial attribute discrimination degree of the candidate attribute to the crack category according to the distribution concentration of the crack category corresponding to the crack area in the clustering result of the candidate attribute includes: The crack area corresponding to any crack category is taken as the target crack; the negative correlation mapping result of the total number of all clusters with the target cracks is taken as the first set parameter; Count the total number of all target cracks in each cluster, and use the largest total number as the parameter in the second set; The product of the first set of parameters and the second set of parameters is calculated to obtain the initial attribute discrimination of the candidate attribute for each crack category.

4. According to the big data-based road defect detection method of claim 1, it is characterized in that: The method for obtaining the detection discrimination comprises: According to the spatial proportion of the crack area corresponding to the crack category in the historical database, the detection attention weight of the crack category is obtained; The detection attention weight of each crack category is used to weight the initial attribute discrimination degree corresponding to each crack category to obtain the detection discrimination degree of the attribute to be selected.

5. According to the big data-based road defect detection method of claim 4, it is characterized in that: The method for acquiring the detection attention weight includes: In the historical database, the total area of ​​all crack regions corresponding to the crack category and the total number of all crack regions corresponding to the crack category are forward fused to obtain the detection attention weight of the crack category.

6. The road defect detection method based on big data according to claim 4, characterized in that: The method of weighting the initial attribute discrimination of each crack category by using the detection attention weight of each crack category to obtain the detection discrimination of the to-be-selected attribute includes: The detection attention weight of each crack category is used to perform weighted summation and normalization on the initial attribute discrimination corresponding to each crack category to obtain the detection discrimination of the selected attribute.

7. The road defect detection method based on big data according to claim 1, characterized in that: The method for obtaining the category representativeness includes: The crack category to which the selected crack belongs is taken as the target category, and all crack categories except the target category are taken as reference categories; The category representativeness is obtained according to a category representativeness formula, wherein the category representativeness formula includes: R is the representative degree of the candidate crack in the crack category to which it belongs; D a is the detection discrimination of the ath distinguishing attribute; d e,a is the absolute value of the difference between the e-th crack region and the a-th distinguishing attribute corresponding to the crack to be selected in the target category; d n,a is the absolute value of the numerical difference between the nth crack area and the ath distinguishing attribute corresponding to the selected crack in all the reference categories; E is the total number of all crack areas in the target category; N is the total number of all crack areas in all the reference categories; A is the total number of all the distinguishing attributes; ε is the preset denominator adjustment value.

8. The road defect detection method based on big data according to claim 1, characterized in that: The method for classifying cracks to be detected comprises: Based on the differences between all distinguishing attributes of the crack to be detected and all representative cracks in the crack category and the category representativeness of the representative cracks, a difference index between the crack to be detected and the crack category is obtained; The crack category corresponding to the smallest difference index is used as the crack category of the crack to be detected.

9. The road defect detection method based on big data according to claim 8, characterized in that: The formula for obtaining the difference index includes: P i is the difference index between the crack to be detected and the crack category i; Δr b,q is the absolute value of the difference between the crack to be detected and the bth representative crack in the crack category i corresponding to the qth distinguishing attribute; R b is the category representativeness of the bth representative crack in crack category i; Q is the total number of distinguishing attributes; B is the total number of representative cracks.

10. A road defect detection system based on big data, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of a road defect detection method based on big data as described in any one of claims 1 to 9 are implemented.

Images