A method for quickly recognizing road classification based on skeleton inscribed circle

Abstract

The application discloses a kind of based on skeleton incircle's road grading quick identification method, including the road prediction image obtained by the neural network of well-trained prediction to the exploration work area remote sensing image;Road prediction image is handled at pixel level, extract the skeleton line of road in road prediction image, obtain road skeleton diagram;Extract the contour line of road, obtain road contour line diagram;The road width corresponding to each point in skeleton line is calculated using incircle method, according to the threshold value, each point in skeleton line is marked and classified;Mark the pixel point in the screened area, extract the graded road information, complete the grading of road.The road grading quick identification method based on skeleton incircle proposed in the application combines the advantages of road skeleton and incircle method, can update and calculate the width of each road at pixel point level, improve the calculation efficiency and accuracy, effectively avoid the fuzzy grading operation caused by uneven road width.

Description

TECHNICAL FIELD

[0001] The present application relates to a kind of skeleton inscribed circle-based road classification quick identification method, belong to seismic exploration technical field. BACKGROUND

[0002] The most commonly used seismic source in seismic exploration is explosive source, but nowadays the rapid development of town leads to the exploration work area is more and more complex, so it needs controllable source combined with explosive source to work together, and the layout of controllable source needs to classify the road network in the exploration area, and according to different road width, design different tonnage controllable source vehicle in different areas in advance, to ensure that the maximum improvement of excitation energy and signal-to-noise ratio. But nowadays the town road is very complex and the road width is unevenly distributed, and the quick identification method of road width is rarely studied.

[0003] At present, the road recognition in seismic exploration uses the method of artificial drawing combined with field reconnaissance to recognize, which is suitable for the area with uniform road width, and cannot intelligently classify and recognize the road width, especially the road with uneven width and large bending angle cannot be accurately classified by artificial drawing. SUMMARY

[0004] In order to overcome the problems in the prior art, the present application provides a kind of skeleton inscribed circle-based road classification quick identification method, which is optimized and improved based on the centerline transformation method, has higher calculation efficiency, is more suitable for processing the road recognition result after image segmentation, and the obtained road skeleton graph is more accurate, so that the subsequent road width calculation is more accurate, thereby improving the accuracy and efficiency of road classification, which can effectively guide the design of controllable source vehicle operation scheme.

[0005] The technical scheme provided by the present application to solve the above technical problems is: a kind of skeleton inscribed circle-based road classification quick identification method, comprising the following steps:

[0006] Step S1, the exploration work area remote sensing image is obtained by the trained neural network prediction road prediction image;

[0007] Step S2, the road prediction image is pixelated, and the skeleton line of the road in the road prediction image is extracted, so as to obtain the road skeleton graph;

[0008] Step S3, the contour line of the road is extracted, so as to obtain the road contour line graph;

[0009] Step S4, the road width corresponding to each point in the skeleton line is calculated by using the inscribed circle method, and each point in the skeleton line is marked and classified according to the set threshold;

[0010] Step S5, mark the pixels in the screened region, extract the graded road information, and complete the grading of the road.

[0011] Further technical solutions are that the specific process in step S2 is:

[0012] Step S21, set the background of the road prediction image to 0 and the road to 1, thereby establishing a 3*3 matrix, the definition of the road boundary point is that it is marked as 1 itself and at least one point in its 8 connected domains is marked as 0, consider the boundary point as the center point q1, its 8 domain points are recorded as q2, q3, q4,..., q9 clockwise from the top left, and extract the boundary point satisfying the following conditions:

[0013] ① 2 <= n(q1) <= 6;

[0014] ② s(q1) = 1;

[0015] ③ q2*q4*q6 = 0;

[0016] ④ q4*q6*q8 = 0;

[0017] In the formula: n(q1) is the number of points marked as 1 in the neighborhood of q1; s(q1) is the number of times of changing from 0 to 1 when traversing the neighborhood points q2, q3,..., q9 of q1;

[0018] Step S22, then set the background of the road prediction image to 0 and the road to 1 again, thereby establishing a 3*3 matrix, the definition of the road boundary point is that it is marked as 1 itself and at least one point in its 8 connected domains is marked as 0, consider the boundary point as the center point q1, its 8 domain points are recorded as q2, q3, q4,..., q9 clockwise from the top left, and extract the boundary point satisfying the following conditions:

[0019] ① 2 <= n(q1) <= 6;

[0020] ② s(q1) = 1;

[0021] ③ q2*q4*q8 = 0;

[0022] ④ q2*q6*q8 = 0;

[0023] Step S23, repeat steps 2 and 3 until no point satisfies the above conditions, the remaining points that are not extracted are the skeleton points, and the skeleton points form a skeleton line.

[0024] Further technical solutions are that the specific process in step S3 is:

[0025] Step S31, perform a denoising convolution operation on the road prediction image;

[0026] Step S32, calculate the gradient amplitude and gradient direction of each pixel point in the road prediction image;

[0027] Step S33, perform non-maximum suppression operation on the edge of the road prediction image according to the gradient amplitude and gradient direction;

[0028] Step S34, set a strong threshold and a weak threshold, mark the pixel point with a gradient amplitude less than the weak threshold as 255; mark the pixel point with a gradient amplitude greater than the strong threshold as 0; for the pixel point with a gradient amplitude less than the strong threshold and greater than the weak threshold, judge whether there is a point greater than the strong threshold in the 8 neighborhood points of the point, if there is, mark the point as 0, if not, mark it as 255;

[0029] Step S35, extract the contour line composed of the marked pixel points, and obtain the road contour line image.

[0030] Further technical solutions are that in the step S31, the Gaussian blur method is used to assign weights to points at different positions in the convolution kernel, and then the Gaussian kernel of the points at different positions is calculated by assigning weights to the points at different positions, and finally the convolution operation is performed on the road prediction image by the Gaussian kernel, so as to achieve the effect of denoising.

[0031] Further technical solutions are that the calculation formula of assigning weights to points at different positions is:

[0032]

[0033] In the formula, G(x, y) is the weight of the point (x, y); and σ is the average of the pixel points in the convolution window.

[0034] Further technical solutions are that the calculation formula in the step S32 is:

[0035]

[0036]

[0037] In the formula, G xy is the gradient amplitude; and θ is the gradient direction.

[0038] Further technical solutions are that in the step S33, it is determined whether the gradient amplitude of the center point is the maximum in its 8 neighborhood, if the center point is not a local maximum, the point is not an edge;

[0039]

[0040] In the formula, G is the gradient amplitude of the center point, and the rest represents the gradient amplitude of the 8 neighborhood points, 1 represents an edge point, and 0 represents a non-edge point.

[0041] Further technical solutions are that the specific process of the step S4 is:

[0042] The step S41, a pixel point coordinate system is established, a minimum rectangle containing a single road is constructed, and the coordinates of the upper left, upper right, lower left and lower right of the rectangle are respectively (x l ,y u ), (x r ,y u ), (x l ,y d ) and (x r ,y d ); for each pixel point coordinate of the road skeleton line, there are a large radius R and a small radius r, so that the constructed circles are partially outside the road contour and fully inside the road contour, and then half_r, which is half of the sum of the large radius R and the small radius r, is taken to construct a circle with half_r as the radius;

[0043] The step S42, whether the edge point of the circle constructed with half_r as the radius exists outside the road contour is judged ; if the point exists outside the road contour, the large radius R takes half_r, and the small radius r remains unchanged; if not, the small radius r takes half_r, and the large radius R remains unchanged;

[0044] The step S43, the above operation is repeated until the difference between the large radius R and the small radius r is within the set accuracy range, and the road width at the skeleton point is obtained as 2r;

[0045] The step S44, a road width threshold range is set, and the skeleton points meeting the requirements are marked according to the threshold range.

[0046] Further technical solutions are that the accuracy calculation formula in the step S43 is:

[0047]

[0048] In the formula, P is the accuracy.

[0049] Further technical solutions are that the specific process of the step S5 is that an external rectangle of the largest inscribed circle with the marked point as the center is constructed, all the pixel points in the rectangle that belong to the road contour are marked as another color, and the grading of the road is completed.

[0050] The present application has the following beneficial effects: the road grading rapid identification method based on the skeleton inscribed circle proposed in the present application combines the advantages of the road skeleton and the inscribed circle method, can perform pixel point level update calculation on the width of each road segment, improves the calculation efficiency and accuracy, and effectively avoids the fuzzy grading operation caused by uneven road width. BRIEF DESCRIPTION OF DRAWINGS

[0051] Figure 1 for a specific embodiment of the present application in the remote sensing image map;

[0052] Figure 2 for a specific embodiment of the present application in the road prediction map;

[0053] Figure 3 for a specific embodiment of the present application in the road skeleton map;

[0054] Figure 4 for a specific embodiment of the present application in the Gaussian kernel calculation flowchart;

[0055] Figure 5 for a specific embodiment of the present application in the gradient amplitude calculation flowchart;

[0056] Figure 6 for a specific embodiment of the present application in the road profile identification map;

[0057] Figure 7 for a specific embodiment of the present application in the width calculation flowchart;

[0058] Figure 8 for a specific embodiment of the present application in the road classification result map (wide road);

[0059] Figure 9 for a specific embodiment of the present application in the road classification result map (narrow road. DETAILED DESCRIPTION

[0060] The technical solutions of the present application will be described below in conjunction with the drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0061] A road classification rapid identification method based on skeleton inscribed circle of the present application comprises the following steps:

[0062] Step S1, the exploration area remote sensing image (such as shown in FIG. 1) is predicted by a trained neural network to obtain a road prediction image (such as shown in FIG. 2); Figure 1 Figure 2 Step S2, the road prediction image is pixelized and the skeleton line of the road in the road prediction image is extracted, thereby obtaining a road skeleton map (such as shown in FIG. 3);

[0063] Step S21, the background of the road prediction image is set to 0 and the road is set to 1 (such as shown in FIG. 4); Figure 3

[0064] Step S22, the road skeleton image is subjected to a Gaussian kernel calculation (such as shown in FIG. 5) to obtain a Gaussian kernel calculation result (such as shown in FIG. 6);​​Figure 2 The definition of the road boundary point is that the point itself is marked as 1 and at least one of its eight connected domain points is marked as 0. Taking the boundary point as the center point q1, its eight domain points are recorded clockwise as q2, q3, q4,..., q9 from the upper left, and the boundary point satisfying the following conditions is extracted.

[0065] ① 2<=n(q1)<=6; (This condition does not include the case that q1 is a line segment end point and q1 has seven domain points marked as 1)

[0066] ② s(q1)=1; (Avoid breaking the skeleton due to calculation on a single pixel point)

[0067] ③ q2*q4*q6=0; (The case that q1 is a boundary right or lower end point)

[0068] ④ q4*q6*q8=0; (The case that q1 is a boundary left or upper end point)

[0069] In the formula, n(q1) is the number of adjacent points of q1 marked as 1, and s(q1) is the number of times of changing from 0 to 1 when traversing the adjacent points q2, q3,..., q9 of q1.

[0070] Step S22, then the background of the road prediction image is set to 0 and the road is set to 1, thereby establishing a 3x3 matrix. The definition of the road boundary point is that the point itself is marked as 1 and at least one of its eight connected domain points is marked as 0. Taking the boundary point as the center point q1, its eight domain points are recorded clockwise as q2, q3, q4,..., q9 from the upper left, and the boundary point satisfying the following conditions is extracted.

[0071] ① 2<=n(q1)<=6; (This condition does not include the case that q1 is a line segment end point and q1 has seven domain points marked as 1)

[0072] ② s(q1)=1; (Avoid breaking the skeleton due to calculation on a single pixel point)

[0073] ③ q2*q4*q8=0; (The case that q1 is a boundary left or upper end point (i.e., q2=0 and q8=0))

[0074] ④ q2*q6*q8=0; (The case that q1 is a boundary right or lower end point (i.e., q4=0 and q6=0))

[0075] Step S23, repeat steps 2 and 3 until no point satisfies the above conditions, and the remaining points that are not extracted are skeleton points. The skeleton points form a skeleton line.

[0076] Step S3, extracting the contour line of the road, thereby obtaining a road contour line image (as shown in Figure 6

[0077] Step S31, performing a denoising convolution operation on the road prediction image; generally, the value of the center point of the denoising convolution kernel is determined by the average value of the surrounding points, but in this case, if the convolution kernel is too large, the points farthest from the center point and the points closest to the center point have the same weight, which will affect the denoising effect. Here, a Gaussian blur method is used to assign weights to the points in different positions in the convolution kernel. The formula is as follows:

[0078]

[0079] In the formula, G(x, y) is the weight of point (x, y); and σ is the average value of the pixel points in the convolution window.

[0080] The specific calculation process of the Gaussian kernel is shown in Figure 4 The obtained Gaussian kernel is used to perform a convolution operation on the road recognition image, which can achieve the effect of denoising.

[0081] Step S32, calculating the gradient amplitude and gradient direction of each pixel point in the road prediction image.

[0082]

[0083]

[0084] In the formula, G xy is the gradient amplitude; and θ is the gradient direction.

[0085] The Sobel operator includes two 3x3 matrices, G x and G y . In actual image processing, the convolution operation is performed on the image, that is, the image data is multiplied by the operator value at the corresponding position and added, to obtain G x in the x direction and G y in the y direction. G x and G y calculated above are squared and added at the corresponding positions to obtain G xy . Each pixel point in the image has a gradient amplitude G xy , which is used for threshold screening later. The calculation process is shown in Figure 5 ;

[0086] Step S33, performing a non-maximum suppression operation on the edges of the road prediction image based on the gradient amplitude and the gradient direction.

[0087] That is, the local maximum value of the pixel point is found, and it is determined whether the gradient amplitude of the center point is the maximum in its 8-neighborhood. If the center point is not the local maximum, the point is not an edge (formula 4).​

[0088]

[0089] wherein: is the central point gradient amplitude, the rest represents the 8-neighbor point gradient amplitude, 1 represents an edge point, and 0 represents a non-edge point;

[0090] Step S34, a strong threshold and a weak threshold are set, a pixel point with a gradient amplitude less than the weak threshold is marked as 255; a pixel point with a gradient amplitude greater than the strong threshold is marked as 0; for a pixel point with a gradient amplitude less than the strong threshold and greater than the weak threshold, it is determined whether there is a point greater than the strong threshold in the 8-neighbor points of the point, if there is, the point is marked as 0, if not, it is marked as 255 (formula 5);

[0091]

[0092] wherein: pixel_v represents the central point pixel value, represents the 8-neighbor point gradient amplitude of the central point;

[0093] Step S35, a contour line is extracted from the marked pixel points, and a road contour line map is obtained;

[0094] Step S4, the road width corresponding to each point in the skeleton line is calculated, and each point in the skeleton line is marked and classified according to a set threshold;

[0095] Step S41, a pixel point coordinate system is established, a minimum rectangle containing a single road is constructed, and the coordinates of the upper left, upper right, lower left, and lower right of the rectangle are set as (x l ,y u ), (x r ,y u ), (x l ,y d ), and (x r ,y d ), respectively. For each pixel point coordinate of the road skeleton line, there is a large radius R and a small radius r, so that the constructed circles are partially outside the road contour and fully inside the road contour, respectively. Then, half_r (formula 6) is taken as half of the sum of the large radius R and the small radius r, and a circle with half_r as the radius is constructed;

[0096]

[0097] Step S42, it is determined whether the edge point (formula 7) of the circle constructed with half_r as the radius exists outside the road contour, if there is a point outside the road contour, the large radius R takes half_r, and the small radius r remains unchanged; if not, the small radius r takes half_r, and the large radius R remains unchanged;

[0098]

[0099] Step S43, repeat the above operation until the difference between the large radius R and the small radius r is within the set precision P (formula 8), and obtain the road width 2r at the skeleton point;

[0100]

[0101] Step S44, set a road width threshold range, and mark the skeleton points that meet the requirements according to the threshold range.

[0102] Step S5, construct the circumscribed rectangle of the largest inscribed circle with the marked point as the center, mark all the pixel points within the rectangle that belong to the road contour as another color, extract the graded road information, and complete the grading of the road (as shown in Figure 8 and Figure 9 ).

[0103] The above description is not intended to limit the present application in any form, although the present application has been disclosed by the above examples, however, it is not intended to limit the present application, any skilled person in the art, without departing from the technical solution of the present application, can make some changes or modifications to the equivalent examples of equivalent changes by using the above disclosed technical content, as long as it does not deviate from the technical solution of the present application, any simple modification, equivalent change and modification made to the above examples according to the technical essence of the present application, still belongs to the scope of the technical solution of the present application.

Claims

1. A method for fast recognition of road classification based on in-skeleton inscribed circle, characterized in that, The method comprises the following steps: Step S1, predicting a road prediction image from a remote sensing image of an exploration work area through a trained neural network; Step S2, performing pixel-level processing on the road prediction image to extract a skeleton line of the road in the road prediction image, thereby obtaining a road skeleton image; Step S3, extracting a contour line of the road, thereby obtaining a road contour line image; Step S4, calculating a road width corresponding to each point in the skeleton line by using an inscribed circle method, and marking and classifying each point in the skeleton line according to a set threshold value; Step S41, establish a pixel point coordinate system, construct a minimum rectangle containing a single road, set the coordinates of the upper left, upper right, lower left and lower right of the rectangle as (x l ,y u ), (x r ,y u ), (x l ,y d ), (x r ,y d ), for each pixel point coordinate of the road skeleton line, there are a large radius R and a small radius r, so that the constructed circles are partially outside the road contour and fully inside the road contour, then take half_r which is half of the sum of the large radius R and the small radius r to construct a circle with half_r as the radius; Step S42, judging the edge point of the circle with half_r as the radius whether there is a point outside the road profile, if there is a point outside the road profile, the large radius R takes half_r, and the small radius r remains unchanged; if not, the small radius r takes half_r, and the large radius R remains unchanged; Step S43, repeating the above operation until a difference between the large radius R and the small radius r is within a set accuracy range, and obtaining a road width of 2r at the skeleton point; Step S44, setting a road width threshold value range, and marking the skeleton points meeting the requirements according to the threshold value range; Step S5, marking the pixel points in the screened area, and extracting graded road information, thereby completing the grading of the road. The specific process of step S5 is as follows: constructing an external rectangle of the largest inscribed circle with the marked point as the center, marking all the pixel points belonging to the road contour in the rectangle as another color, and completing the grading of the road.

2. The method according to claim 1, wherein, The specific process in step S2 is as follows: Step S21, setting the background of the road prediction image as 0 and the road as 1 to establish a 3*3 matrix, the definition of the road boundary point is that it is marked as 1 itself, and at least one point in its eight connected domains is marked as 0, considering the boundary point as a center point q1, its eight domain points are recorded as q2, q3, q4,..., q9 clockwise from the top left, and the boundary point meeting the following conditions is extracted; ① 2<=n(q1)<=6; ② s(q1)=1; ③ q2*q4*q6=0; ④ q4*q6*q8=0; In the formula: n(q1) is the number of adjacent points marked as 1 of q1; s(q1) is the number of times of changing from 0 to 1 when traversing the adjacent points q2, q3,..., q9 of q1; Step S22, then setting the background of the road prediction image as 0 and the road as 1 to establish a 3*3 matrix, the definition of the road boundary point is that it is marked as 1 itself, and at least one point in its eight connected domains is marked as 0, considering the boundary point as a center point q1, its eight domain points are recorded as q2, q3, q4,..., q9 clockwise from the top left, and the boundary point meeting the following conditions is extracted; ① 2<=n(q1)<=6; ② s(q1)=1; ③ q2*q4*q8=0; ④ q2*q6*q8=0; Step S23, repeating steps 2 and 3 until no point meets the above conditions, and the remaining points that are not extracted are skeleton points, and the skeleton points form a skeleton line.

3. The method of claim 1, wherein, The specific process in step S3 is as follows: Step S31, performing a denoising convolution operation on the road prediction image; Step S32, calculating a gradient amplitude and a gradient direction of each pixel point in the road prediction image; Step S33, performing a non-maximum suppression operation on the edge of the road prediction image according to the gradient amplitude and the gradient direction; Step S34, setting a strong threshold and a weak threshold, marking the pixel point with a gradient amplitude less than the weak threshold as 255; marking the pixel point with a gradient amplitude greater than the strong threshold as 0; for the pixel point with a gradient amplitude less than the strong threshold and greater than the weak threshold, judging whether there is a point greater than the strong threshold in the 8 neighborhood points of the point, if there is, marking the point as 0, if not, marking the point as 255; Step S35, extracting the marked pixel points to form a contour line, thereby obtaining a road contour line graph.

4. The method according to claim 3, wherein, In the step S31, a Gaussian blur method is used to assign weights to points at different positions in the convolution kernel, then the Gaussian kernel of the points at different positions is calculated through the weight assignment of the points at different positions, and finally the convolution operation is performed on the road prediction image through the Gaussian kernel, thereby achieving the effect of denoising.

5. The method according to claim 4, wherein, The calculation formula of the weight assignment of the points at different positions is as follows: In the formula, G(x, y) is the weight of the point (x, y); and σ is the mean value of the pixel points in the convolution window.

6. The method of claim 3, wherein the method is characterized by, The calculation formula in the step S32 is as follows: where: G xy is the gradient magnitude; θ is the gradient direction; G x and G y are two 3x3 matrices of Sobel operator, respectively.

7. The method of claim 3, wherein the method is characterized by, In the step S33, it is determined whether the gradient amplitude of the center point is the maximum in its 8 neighborhood points, if the center point is not a local maximum, the point is not an edge; In the formula: is the center point gradient magnitude, the rest represent the 8-neighbor point gradient magnitudes, 1 represents an edge point, and 0 represents a non-edge point.

8. The method of claim 1, wherein, The precision calculation formula in the step S43 is as follows: In the formula, P is the precision.

Images