A method for clustering convex hull disease information

Abstract

The present invention discloses a method for clustering convex hull disease information, including: mapping the longitude and latitude of the contour point set of any disease on the pavement to a plane rectangular coordinate system; assigning type numbers and access statuses according to the types of the diseases; obtaining the distance between any two diseases, and presetting the minimum radius threshold that the diseases can reach; for diseases of the same type, querying the reachable diseases according to the distance between the diseases and the minimum radius threshold to form a reachable disease set; clustering the diseases in the disease set to obtain a disease set of the same category; using the Graham Scan algorithm to calculate the convex hull of the diseases in the disease set of the same category to obtain the convex hull after clustering. Through the above scheme, the present invention has the advantages of simple logic, accurate and reliable clustering, etc., and has high practical value and promotion value in the field of pavement disease clustering technology.

Description

Technical Field

[0001] The invention relates to the technical field of pavement disease clustering, and in particular to a method for clustering convex hull disease information. Background Art

[0002] The pavement diseases described in this article mainly include cracks, potholes, peeling, bulges, etc. At present, there are many disease recognition algorithms in the prior art, for example: the Chinese invention patent with patent application number "202110228020.4" and name "Airport pavement disease and foreign body detection method based on multiple convolutional neural network fusion", which collects airport pavement disease and foreign body images; constructs an airport pavement disease and foreign body database for training neural networks; builds target detection algorithms YOLOv3 and Mask R-CNN convolutional neural networks; adjusts the convolutional neural network hyperparameters until convergence and the error loss value meets the requirements, saves the network weight parameters at this time, and completes the training of YOLOv3 and Mask R-CNN convolutional neural networks; fuses the trained YOLOv3 and Mask R-CNN convolutional neural networks to construct an intelligent segmentation model for airport pavement diseases and foreign bodies at the pixel level; inputs the test image into the saved model, and outputs the segmentation results of airport pavement diseases and foreign bodies; counts the pixels of the image corresponding to the mask of the segmentation result, and outputs the semantic information of airport pavement diseases and foreign bodies.

[0003] Another example is the Chinese invention patent with patent application number "202110032020.7" and name "A network model and detection method for road surface block disease detection", which includes the first branch to the sixth branch arranged from top to bottom; the first branch is the first convolution layer with a convolution kernel of 1×1; the second branch includes the second convolution layer with a convolution kernel of 1×1 and the third convolution layer with a convolution kernel of 3×3, which are connected and packaged from front to back; the third branch includes the The fourth convolution layer has a convolution kernel of 1×1, the fifth convolution layer has a convolution kernel of 3×3, and the sixth convolution layer has a convolution kernel of 3×3; the fourth branch includes a first average pooling layer of 9×9 and a seventh convolution layer with a convolution kernel of 1×1, which are connected and packaged from front to back; the fifth branch includes a second average pooling layer of 17×17 and an eighth convolution layer with a convolution kernel of 1×1, which are connected and packaged from front to back; the sixth branch includes a third average pooling layer and a ninth convolution layer, which are connected and packaged from front to back.

[0004] Another example is the Chinese invention patent with patent application number "201810674803.3" and name "Image-based Adaptive Constant False Alarm Detection Method for Pavement Crack Defects". It extracts the pixels of the image one by one by column or row to form a one-dimensional signal, and regards the extracted data as a radar pulse echo signal. It uses constant false alarm detection technology to determine whether there are crack features in the data, and retains the signal waveform data with crack information. After all the row or column data of the image are processed, an image containing only cracks is obtained, and finally a classification method is used to distinguish different cracks.

[0005] The above methods can reliably detect and identify diseases and obtain a contour point data set of the disease; however, in the prior art, the diseases remain at the identification process and are not clustered later, making it impossible to perform key area identification processing.

[0006] The Chinese invention patent with application number "202110024616.2" and name "Automatic Detection Method of Underground Structure Defects of Airport Pavement Based on Deep Learning" includes: using a multi-channel ground penetrating radar to collect B-scan images and C-scans of underground structures of airport pavements; constructing a convolutional neural network consisting of two sub-networks, a three-dimensional convolution and a two-dimensional convolution, and using the three-dimensional convolution and two-dimensional convolution sub-networks to respectively extract the three-dimensional feature map containing the correlation between multiple channels of underground diseases in the above-mentioned C-scan and the two-dimensional feature map of underground diseases in the B-scan image; fusing the three-dimensional feature map and the two-dimensional feature map obtained in step 2 to obtain a fused feature map containing rich information about underground diseases; constructing a convolutional neural network consisting of two sub-networks, a three-dimensional convolution and a two-dimensional convolution sub-network, and extracting the three-dimensional feature map containing the correlation between multiple channels of underground diseases in the C-scan and the two-dimensional feature map of underground diseases in the B-scan image; fusing the three-dimensional feature map and the two-dimensional feature map obtained in step 2 to obtain a fused feature map containing rich information about underground diseases; and constructing a convolutional neural network consisting of two sub-networks, a three-dimensional convolution and a two-dimensional convolution sub-network. A 3D deconvolution subnetwork is constructed to upsample the fusion feature map obtained in step 3 to obtain a segmentation map of the same size as the input voxel, so as to achieve pixel-level segmentation of the target in the acquired radar image; the foreground pixels in the segmentation map obtained in step 4 are clustered into clusters of different sizes, and then a 3D circumscribed rectangular box is obtained from all the generated clusters as the initial prediction box; all the initial prediction boxes generated in step 5 are mapped to the corresponding positions of the 3D feature map obtained in step 2 to obtain a 3D feature map of a fixed size; a correction subnetwork is constructed, and the subnetwork and the 3D feature map of a fixed size generated in step 6 are used to correct and classify the initial prediction box obtained in step 5 to obtain a 3D prediction box with more accurate positioning. Its disadvantages are: it is based on the clustering of pixels, and there is no constraint relationship between pixels; it cannot pay attention to all the internal information of each disease, and the clustering result is inaccurate.

[0007] Therefore, there is an urgent need to propose a method for clustering convex hull disease information that is simple in logic, accurate and reliable. Summary of the invention

[0008] In view of the above problems, the purpose of the present invention is to provide a method for clustering convex hull disease information. The technical solution adopted by the present invention is as follows:

[0009] A method for clustering convex hull disease information, comprising the following steps:

[0010] Map the longitude and latitude of the contour point set of any road surface defect to a plane rectangular coordinate system;

[0011] Assign type number and access status according to the type of disease;

[0012] Find the distance between any two diseases and preset the minimum radius threshold that the disease can reach;

[0013] For diseases of the same type, the reachable diseases are queried based on the distance between the diseases and the minimum radius threshold to form a reachable disease set;

[0014] Cluster the diseases in the disease set to obtain a disease set of the same category;

[0015] The Graham Scan algorithm is used to calculate the convex hull of diseases in the same category of disease sets to obtain the clustered convex hull.

[0016] Furthermore, mapping the longitude and latitude of the contour point set of any road surface defect to a plane rectangular coordinate system includes the following steps:

[0017] The origin of the rectangular coordinate system is O(x, y), x=0, y=0, and its position in the latitude and longitude coordinate system is O′(x′, y′), x′=t0, y′=g0;

[0018] The latitude and longitude position of the current disease P′(t1, g1) is converted into rectangular coordinates P(P x , P y ), whose expression is:

[0019]

[0020] Among them, OneMPerLon means that in the longitude and latitude coordinates, the unit distance of longitude is converted to the length in the rectangular coordinates; OneMPerLat means that in the longitude and latitude coordinates, the unit distance of latitude is converted to the length in the rectangular coordinates.

[0021] Furthermore, obtaining the distance between any two diseases includes the following steps:

[0022] The data point sets of diseases D1 and D2 are: D1 = {P 1，1 , P 1，2 , P 1，3 KP 1，N}, D2={P 2，1 , P 2，2 , P 2，3 KP 2，M}; where N is the number of points in the disease D1, and M is the number of points in the disease D2;

[0023] We obtain D1 = {P 1，1 , P 1，2 , P 1，3 KP 1,N Any point in D2 = {P 2,1 ,P 2,2 ,P 2,3 KP 2,M The distance of any point in} is expressed as:

[0024]

[0025] in, It represents the x value of the ith point in the disease D1 in the rectangular coordinate system; It represents the x value of the jth point in the disease D2 in the rectangular coordinate system; It represents the y value of the i-th point in the disease D1 in the rectangular coordinate system; represents the y value of the jth point in the disease D2 in the rectangular coordinate system; 0<i≤N, 0<j≤M;

[0026] Find the minimum value of disease D1 and disease D2:

[0027]

[0028] Furthermore, it also includes:

[0029] Traverse any disease and obtain the access status of the disease. If it has not been visited, assign its clustering category; otherwise, visit the next disease;

[0030] Get the list of disease sets that are reachable by any disease, and visit the diseases in the disease set. If they are not visited, assign their clustering categories; otherwise, visit the next disease until any disease in the disease set is visited;

[0031] Diseases of the same clustering category are grouped into disease sets of the same category.

[0032] Preferably, the Graham Scan algorithm is used to calculate the convex hull of the diseases in the disease set of the same category to obtain the clustered convex hull, which includes the following steps:

[0033] In the rectangular coordinate system, obtain the data point of the disease at the bottom left corner of the disease set of the same category;

[0034] The remaining disease data points are arranged in order of polar angle and distance;

[0035] Take data points a from front to back in the sorting directioni , data point a i+1 and data point a i+2 , find the vector a i a i+1 With vector a i+1 a i+2 Is the cross product of less than 0?

[0036] If it is less than 0, delete the data point a. i+2 , and add data point a i+2 Continue to judge the next data point;

[0037] If it is not less than 0, the three extracted data points are postponed one position in turn, and the judgment continues until all data points are compared to obtain the data contour of the convex hull.

[0038] Furthermore, the location information of the clustered convex hull is converted into longitude and latitude coordinates:

[0039] The rectangular coordinate position P(x1,y1) of the current disease is converted into the longitude and latitude coordinates P'(P t ,P g ), whose expression is:

[0040]

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] (1) The present invention cleverly transforms the linear transformation, maps the longitude and latitude of the disease to a plane rectangular coordinate system, and then converts the rectangular coordinates into longitude and latitude coordinates after the location information of the convex hull is clustered. The advantage is that the difference between the diseases using longitude and latitude coordinates is too small, and the calculation error is large. The present invention facilitates calculation after conversion into a rectangular coordinate system;

[0043] (2) The present invention adds a reachable distance to the disease, which has the advantage that different degrees of clustering results can be obtained according to the size of the reachable distance, and it is helpful to obtain the reachable disease of the disease in advance;

[0044] (3) The present invention searches for reachable diseases based on the distance between diseases and the minimum radius threshold, forms a reachable disease set, and performs clustering access to the diseases in the disease set to obtain a disease set of the same category. The clustering is simple and reliable.

[0045] (4) The present invention uses the Graham Scan algorithm to calculate the convex hull to obtain the convex hull points, which is simple and effective with low time complexity;

[0046] In summary, the present invention has the advantages of simple logic, accurate and reliable clustering, etc., and has high practical value and promotion value in the field of pavement disease clustering technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope of protection. For those skilled in the art, other related drawings can be obtained based on these drawings without creative work.

[0048] Figure 1 It is a schematic diagram of the clustering results of the present invention.

[0049] Figure 2 This is a cluster access flow chart of the present invention. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical scheme and advantages of this application clearer, the present invention is further described below in conjunction with the accompanying drawings and embodiments, and the embodiments of the present invention include but are not limited to the following embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0051] Example

[0052] like Figure 1 As shown, this embodiment provides a method for clustering convex hull disease information, which includes the following steps:

[0053] The first step is to use linear transformation to map the longitude and latitude of the contour point set of any road surface defect to the plane rectangular coordinate system:

[0054] Assume the origin of the rectangular coordinate system is O(x,y), x=0, y=0, and its position in the longitude and latitude coordinate system is O'(x',y'), x'=t0, y'=g0;

[0055] The latitude and longitude position of the current disease P'(t1,g1) is converted into rectangular coordinates P(P x ,P y ), whose expression is:

[0056]

[0057] Among them, OneMPerLon means that in the longitude and latitude coordinates, the unit distance of longitude is converted to the length in the rectangular coordinates, and its value is 0.00001141; OneMPerLat means that in the longitude and latitude coordinates, the unit distance of latitude is converted to the length in the rectangular coordinates, and its value is 0.00000899.

[0058] The second step is to assign type numbers and access statuses according to the types of diseases;

[0059] Visit flag: isVisited, indicating whether the current disease has been visited, true means it has been visited; false means it has not been visited.

[0060] Cluster category: iClusterID. After clustering, all diseases are divided into several categories. Each category contains one or more diseases. iClusterID represents the category number to which a disease belongs. At the beginning of the algorithm, the iClusterID of all diseases is set to -1.

[0061] The third step is to find the distance between any two diseases and preset the minimum radius threshold that the disease can reach;

[0062] Assume two diseases are D1 and D2, each disease contains a number of data points that make up the disease: D1 = {P 1,1 ,P 1,2 ,P 1,3 KP 1,N}, D2={P 2,1 ,P 2,2 ,P 2,3 KP 2,M}, N is the number of points in the disease D1, M is the number of points in the disease D2, represents the x value of the mth point of the nth disease in the rectangular coordinate system. Similarly, represents the y value of the mth point of the nth disease in the rectangular coordinate system. 1,i (0<i≤N), P 2,j (0<j≤M), calculate the distance d(i,j) between two points:

[0063]

[0064] in, It represents the x value of the ith point in the disease D1 in the rectangular coordinate system; It represents the x value of the jth point in the disease D2 in the rectangular coordinate system; It represents the y value of the i-th point in the disease D1 in the rectangular coordinate system; represents the y value of the jth point in the disease D2 in the rectangular coordinate system; 0<i≤N, 0<j≤M;

[0065] The distance between two diseases Take the minimum value of d and use the formula:

[0066]

[0067] The rule for determining whether the diseases are reachable is: the distance between two diseases is less than the cluster radius r, that is It is considered that D1 and D2 can reach each other among the two diseases.

[0068] Step 4: For diseases of the same category, according to the distance between diseases and the minimum radius threshold, query reachable diseases to form a set of reachable diseases;

[0069] Step 5: Cluster access to the diseases in the disease set to obtain a disease set of the same category;

[0070] The clustering process can be divided into an inspection stage and an expansion stage:

[0071] Inspection stage:

[0072] Traverse each disease in the list and check the status of the disease. If it has not been visited, set it as visited, assign its clusterID to a new ID (set as i), which is the category to which the disease belongs, and enter the expansion stage. If it has been visited, skip it and check the next disease. Until all diseases in the list have been checked.

[0073] Expansion stage:

[0074] View all ArrivalList list items of this category, check all contents of the list items, traverse the diseases in the list. If the status of the disease is not visited, set it as visited, assign its clusterID to the ID of the initial node, that is, i, which is the category to which the disease belongs, and pass the disease into the expansion stage again. If it has been visited, skip it and check the next disease until all list items have been checked.

[0075] As Figure 2 shown, the clustering process is as follows:

[0076] Suppose the length of the disease is L, that is, there are L diseases to be clustered. First, take i = 0, ID = 0, and the status of each disease is not visited.

[0077] 1) Check whether i < L holds. If it holds, go to 2); if not, go to 7).

[0078] 2) Check whether the i-th disease has been visited. If it has been visited, then i + 1, and go to 1). If it has not been visited, go to 3).

[0079] 3) Update the status of this disease to visited, determine its category as ID, and at the same time obtain the reachable list of this disease. Get the length K of the reachable list, and initialize j = 0.

[0080] 4) Check whether j < K holds. If it holds, take the j-th disease; if not, go to 6).

[0081] 5) Check whether the jth disease has been visited. If it has been visited, then go to j+1 and go to 4). If it has not been visited, go to 3).

[0082] 6) All the indirect and directly accessible diseases of the i-th category have been clustered, update i+1, ID+1. Go to 1).

[0083] 7) All diseases have been clustered and the algorithm ends.

[0084] In the sixth step, the Graham Scan algorithm is used to calculate the convex hull of the diseases in the disease set of the same category to obtain the convex hull after clustering. The Graham Scan algorithm first sets a starting point, usually the leftmost point and the rightmost point, and then checks each point one by one. If the convexity of the enclosing area formed by the newly added point and the previously found point does not change, continue to check the next point, otherwise delete the last point found and compare the convexity again until the convexity does not change. Scan the upper and lower enclosing areas separately and merge them together to form the final convex hull.

[0085] Specifically:

[0086] 1. The points in each overall disease are a point set. Select the point in the lower left corner of the overall disease, that is, the point with the smallest x, followed by the smallest y.

[0087] 2. The remaining points are sorted by polar angle. If the polar angles are equal, the one closest to the pole is given priority.

[0088] 3. Start from the first two points, let the first two points be A and B, insert the third point C in sequence, calculate whether the cross product of AB and BC is less than 0 after adding the third point. If it is less than 0, it means that the concavity has changed in the reverse direction. Delete point C, add the next point, and continue to judge. If it is not less than 0, it means that the concavity has not changed, add point D, and judge the cross product of BC and CD, and so on.

[0089] 4. Finally, the convex hull points are obtained.

[0090] The seventh step is to convert the point set of the convex hull from the rectangular coordinate system to the position value of the longitude and latitude coordinates again. The expression is:

[0091]

[0092] The above embodiments are only preferred embodiments of the present invention and are not intended to limit the protection scope of the present invention. Any changes that adopt the design principles of the present invention and are made through non-creative labor on this basis should fall within the protection scope of the present invention.

Claims

1. A method for clustering convex hull disease information, characterized in that: The following steps are involved: Map the longitude and latitude of the contour point set of any road surface defect to a plane rectangular coordinate system; Assign type number and access status according to the type of disease; Find the distance between any two diseases and preset the minimum radius threshold that the disease can reach; For diseases of the same type, according to the distance between diseases and the minimum radius threshold, the reachable diseases are queried to form a reachable disease set; Traverse any disease and obtain the access status of the disease. If it has not been visited, assign its clustering category; otherwise, visit the next disease; Get the list of disease sets that are reachable by any disease, and visit the diseases in the disease set. If they are not visited, assign their clustering categories; otherwise, visit the next disease until any disease in the disease set is visited; Diseases of the same clustering category are grouped into disease sets of the same category; The Graham Scan algorithm is used to calculate the convex hull of diseases in the same category of disease sets to obtain the clustered convex hull.

2. The method for clustering convex hull disease information according to claim 1, characterized in that: Mapping the longitude and latitude of the contour point set of any road surface defect to a plane rectangular coordinate system includes the following steps: The origin of the rectangular coordinate system is O(x,y), x=0, y=0, and its position in the latitude and longitude coordinate system is O'(x',y'), x'=t0, y'=g0; The latitude and longitude position of the current disease P'(t1,g1) is converted into rectangular coordinates P(P x ,P y ), whose expression is: Among them, OneMPerLon means that in the longitude and latitude coordinates, the unit distance of longitude is converted to the length in the rectangular coordinates; OneMPerLat means that in the longitude and latitude coordinates, the unit distance of latitude is converted to the length in the rectangular coordinates.

3. The method for clustering convex hull defects information according to claim 1, characterized in that: Obtaining the distance between any two diseases includes the following steps: The data point sets of diseases D1 and D2 are: D1 = {P 1,1 ,P 1,2 ,P 1,3 KP 1,N }, D2={P 2,1 ,P 2,2 ,P 2,3 KP 2,M }; where N is the number of points in the disease D1, and M is the number of points in the disease D2; We obtain D1 = {P 1,1 ,P 1,2 ,P 1,3 KP 1,N Any point in D2 = {P 2,1 ,P 2,2 ,P 2,3 KP 2,M The distance of any point in} is expressed as: in, It represents the x value of the ith point in the disease D1 in the rectangular coordinate system; It represents the x value of the jth point in the disease D2 in the rectangular coordinate system; It represents the y value of the i-th point in the disease D1 in the rectangular coordinate system; Indicates the y value of the jth point in the disease D2 in the rectangular coordinate system; 0<i≤N, 0<j≤M; Find the minimum value of disease D1 and disease D2:

4. The method for clustering convex hull disease information according to claim 1, characterized in that: The GrahamScan algorithm is used to calculate the convex hull of the diseases in the same category of disease set to obtain the clustered convex hull, including the following steps: In the rectangular coordinate system, obtain the data point of the disease at the bottom left corner of the disease set of the same category; The remaining disease data points are arranged in order of polar angle and distance; Take data points a from front to back in the sorting direction i , data point a i+1 and data point a i+2 , find the vector a i a i+1 With vector a i+1 a i+2 Is the cross product of less than 0? If it is less than 0, delete the data point a. i+2 , and add data point a i+2 Continue to judge the next data point; If it is not less than 0, the three extracted data points are postponed one position in turn, and the judgment continues until all data points are compared to obtain the data contour of the convex hull.

5. The method for clustering convex hull defects information according to claim 1, characterized in that: Also included: Convert the location information of the clustered convex hull into longitude and latitude coordinates: The rectangular coordinate position P(x1,y1) of the current disease is converted into the longitude and latitude coordinates P'(P t ,P g ), whose expression is:

Images