A multi-information cascade clustering method for power transmission line detection

Abstract

The present invention discloses a multi-information cascade clustering transmission line detection method, comprising the following steps: (1) acquiring a transmission line image and performing grayscale transformation on the image; (2) performing image preprocessing on the grayscale transformed image using an edge enhancement algorithm; (3) performing cumulative probability Hough transform on the enhanced image to extract candidate straight lines in the image; (4) performing cascade clustering analysis on the straight line information according to the length, angle, and position information of the straight line; (5) calculating the distance between all straight lines according to the straight line distance formula, and removing duplicate straight lines; (6) predicting the direction of the transmission line based on the angle position information obtained by calculating the final straight line, and marking the transmission line with a rectangular frame. The present invention can effectively delete interference lines that do not match the transmission line features caused by complex backgrounds, and finally obtain all straight lines corresponding to the edges of the transmission lines, with high accuracy and strong robustness, and good performance under complex backgrounds.

Description

Technical Field

[0001] The invention belongs to the field of digital image processing, and in particular relates to a multi-information cascade clustering transmission line detection method. Background Art

[0002] Transmission lines are usually assumed to be in the wild with complex and changeable environments. Due to their long-term exposure to the outdoor environment, they are subject to severe weather such as high temperatures and heavy snow, which leads to a decrease in transmission efficiency. In order to ensure the stable transmission of electricity, regular inspections of transmission lines are particularly necessary. Traditional manual inspection methods are inefficient, costly, and dangerous; in comparison, drone inspections have the advantages of strong mobility, no terrain restrictions, and easy operation. Using drones to inspect transmission lines can improve inspection efficiency, reduce costs, and avoid risks.

[0003] In power inspections, drones are often equipped with lightweight cameras that can obtain rich environmental information. The cameras are used to obtain aerial images and perform transmission line inspections to prepare for further analysis of transmission line faults. Therefore, it is particularly important to accurately detect the relative position of the transmission line in the aerial images.

[0004] Application number CN201710532067.3, "A method for detecting transmission lines based on aerial images", mainly uses the Hessian matrix to enhance aerial images, then divides the image into blocks according to the distribution characteristics of the transmission lines in the aerial images, applies the Hough line detection algorithm to the transmission line area, and finally screens out the wrong transmission lines based on the line spacing and the slope of the transmission lines. This method can accurately detect transmission lines with parallel features in the image. However, in actual situations, due to the camera perspective, the transmission lines may not have parallel features in the image, which will affect the robustness of the method.

[0005] The application number CN201910492848.3, "Transmission Line Detection Method Based on CBS Straight Segment Detection", mainly uses a steerable filter to segment drone aerial images, then uses the CBS straight line detection method to detect straight line segments in the segmented images, and finally links the detected straight line segments according to the Gestalt principle. This method does not impose conditional constraints on transmission line aerial images and reduces the complexity of the algorithm, but it cannot detect transmission lines in complex backgrounds.

[0006] Application number CN201910566009.1, "A method for detecting power lines under complex backgrounds based on image processing", mainly uses the Canny operator to detect edges on aerial images, then uses the region growing method to remove small connected regions, and then uses the probabilistic Hough transform to detect straight lines, and finally selects the correct straight line segments. This method uses connected region analysis to reduce the impact of complex backgrounds on the algorithm, but the double edge features of the Canny operator affect the final marking of the power lines and increase the complexity of the algorithm.

[0007] Therefore, although there are many studies in the field of transmission line detection at home and abroad, most methods cannot accurately detect transmission lines while coping with complex background interference. Summary of the invention

[0008] Purpose of the invention: The purpose of the present invention is to provide a multi-information cascade clustering transmission line detection method, which has the advantage that it can effectively delete interference lines caused by complex backgrounds that do not match the characteristics of transmission lines, and finally obtain all straight lines corresponding to the edges of transmission lines. It has high accuracy and strong robustness, and performs better under complex backgrounds.

[0009] Technical solution: The multi-information cascade clustering transmission line detection method of the present invention comprises the following steps:

[0010] (1) Obtain the transmission line image and perform grayscale transformation on the image;

[0011] (2) Use edge enhancement algorithm to preprocess the image after grayscale transformation;

[0012] (3) Perform cumulative probability Hough transform on the enhanced image to extract candidate straight lines in the image;

[0013] (4) Perform cascade cluster analysis on the straight line information based on the length, angle, and position information of the straight line;

[0014] (5) Calculate the distances between all straight lines according to the straight-line distance formula and remove duplicate straight lines;

[0015] (6) The resulting straight line is used to predict the direction of the transmission line based on the calculated angle position information, and the transmission line is marked with a rectangular frame.

[0016] In step (1), the transmission line image is obtained and the image is gray-scaled. Specifically, a camera is installed on the top of the nose of the drone, and the camera is used to capture images during the flight of the drone. The captured images are transmitted to the processor for gray-scale conversion.

[0017] In step (2), the edge enhancement algorithm based on the Hessian matrix is ​​used on the grayscale transformed image to extract the edge features belonging to the strong edges in the image, thereby enhancing the edge of the transmission line.

[0018] The calculation formula of the Hessian matrix is:

[0019]

[0020] Where f is the function of the image with respect to the coordinates (x, y) xx , f xy , f yx , f yy They represent the second-order partial derivatives of function f with respect to x, y and their mixtures respectively; the eigenvalues ​​and eigenvectors of the Hessian matrix reflect the intensity and direction of the change in the grayscale curvature of the image.

[0021] In step (3), the cumulative probability Hough transform is performed on the enhanced image to extract candidate straight lines in the image, which specifically includes the following steps:

[0022] (3.1) Within the specified range, the edge pixel information after image enhancement is mapped to the Hough space;

[0023] (3.2) Vote for each edge pixel mapped to the Hough space;

[0024] (3.3) The set of suspected points in the plane within the accumulation range and the points exceeding the set threshold represent a straight line.

[0025] In step (4), the cascade cluster analysis of the straight line information specifically includes the following steps:

[0026] (4.1) Use clustering algorithm to classify the length, angle and position information of the straight line;

[0027] (4.2) Mark the positive class in the line according to the number of votes in step (3);

[0028] (4.3) Perform hard cascade analysis on the results and perform intersection operation on the obtained positive classes. When all the straight lines are classified as positive classes in the three types of information, the straight lines are marked as transmission lines.

[0029] In step (5), the distances between all straight lines are calculated according to the straight-line distance formula, and duplicate straight lines are removed, which specifically includes the following steps:

[0030] (5.1) Load the transmission line image and line parameters;

[0031] (5.2) Estimate the distance from the first straight line to the other straight lines;

[0032] (5.3) Filter all the lines whose distance is less than the threshold, select the longest line to keep, and delete the other lines;

[0033] (5.4) Estimate the distance between the second straight line and other straight lines;

[0034] (5.5) Repeat steps (5.3) and (5.4) until the last straight line is calculated.

[0035] The distance estimation formula is:

[0036]

[0037] In the formula, A, B, and C are the parameters of the general form of a straight line in a two-dimensional rectangular coordinate system, where (x i ,y i ) is any point on one of the straight lines, n is the number of points, is the estimated distance. By setting the threshold D, when It is believed that there is a problem of duplicate detection.

[0038] Beneficial effects: Compared with the prior art, the present invention has the following beneficial effects: (1) By fusing multiple features of transmission lines in aerial images, the influence of complex background on the detection effect of transmission lines is effectively reduced; (2) The constraints of the algorithm on the features of transmission lines in aerial images are reduced, thereby enhancing the robustness of the algorithm; (3) The adopted algorithm has fewer constraints on aerial images, has stronger robustness and accuracy, is more suitable for actual conditions, and is conducive to engineering implementation. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 is a flow chart of the present invention;

[0040] Figure 2 The original aerial photograph of the power lines in the present invention;

[0041] Figure 3 This is an image enhancement effect diagram of the aerial image based on the Hessian matrix in the present invention;

[0042] Figure 4 This is a cumulative probability Hough transform straight line detection effect diagram in the present invention;

[0043] Figure 5 It is a result diagram of multi-information cascade clustering analysis in the present invention;

[0044] Figure 6 The clustering effect diagram of the straight line length information and the straight line spatial position information in the present invention;

[0045] Figure 7 This is a cascade clustering effect diagram of the straight line length, angle, and position information in the present invention;

[0046] Figure 8 This is a diagram of the transmission line detection results in the present invention;

[0047] Fig. 9 This is a diagram showing the effect of the transmission line detection in other scenarios of the present invention. DETAILED DESCRIPTION

[0048] The present invention is further described in detail below in conjunction with specific implementation methods and the accompanying drawings.

[0049] Taking into account the problem of a large number of pseudo straight lines encountered when detecting transmission lines in complex backgrounds, the present invention combines the length, angle, and position information of the transmission lines in aerial images to perform cascade clustering analysis on the lines, thereby improving the robustness and accuracy of the transmission line detection algorithm.

[0050] like Figure 1 As shown, the multi-information cascade clustering transmission line detection method of the present invention comprises the following steps:

[0051] (1) Obtain the image of the power transmission line and perform grayscale conversion on the image. Specifically, a camera is installed on the top of the UAV’s nose. The camera captures the image during the flight of the UAV. The captured image is transmitted to the processor for grayscale conversion. The original image is as follows: Figure 2 As shown,

[0052] (2) In order to facilitate edge detection of power lines in images, aerial images need to be preprocessed. Since RGB images contain a large number of redundant color features, the input image needs to be grayed to reduce the overall image data volume. The straight lines obtained after the image is grayed are not obvious, so the edges of the power lines in the image need to be enhanced. The straight line information corresponding to the power lines belongs to the strong edges in the image. The image enhancement algorithm based on the Hessian matrix is ​​used to extract the features belonging to the strong edges. The two-dimensional Hessian matrix is ​​expressed as:

[0053]

[0054] Where f is the function of the image with respect to the coordinates (x, y) xx , f xy , f yx , f yy Represent the second-order partial derivatives of function f with respect to x, y and their mixtures respectively; the eigenvalues ​​and eigenvectors of the Hessian matrix reflect the intensity and direction of the change in the grayscale curvature of the image. By calculating the Hessian matrix, the linear structure is enhanced according to the anisotropy of the linear structure in the image, such as Figure 3 shown.

[0055] (3) Performing cumulative probability Hough transform on the enhanced image to extract candidate straight lines in the image, specifically including the following steps:

[0056] (3.1) Within the specified range, the edge pixel information after image enhancement is mapped to the Hough space;

[0057] (3.2) Vote for each edge pixel mapped to the Hough space;

[0058] (3.3) The set of suspected points in the plane within the accumulation range and the points exceeding the set threshold represent a straight line, thereby reducing the amount of calculation and shortening the calculation time.

[0059] (4) Perform cascade cluster analysis on the straight line information based on the length, angle, and position information of the straight line; Figure 4 As shown in the figure, according to the cumulative Hough transform principle, when the peak value of the intersection of the line segments in the Hough space reaches the threshold, the system will automatically identify the edge as a straight line. However, the edges that do not belong to the transmission line will generate false peaks. These false peaks may be generated by dense leaves or the edges of the trunks of houses in the background. Cascade clustering analysis is performed on the straight line information, which specifically includes the following steps:

[0060] (4.1) Use clustering algorithm to classify the length, angle and position information of the straight line;

[0061] (4.2) Mark the positive class in the line according to the number of votes in step (3);

[0062] (4.3) Perform hard cascade analysis on the results and perform intersection operation on the obtained positive classes. When all the straight lines are classified as positive classes in the three types of information, the straight lines are marked as transmission lines.

[0063] Cluster analysis is performed on the length information and spatial position information of the straight lines in the image, such as Figure 6 As shown in Figure 2, none of them can effectively filter out negative classes. Therefore, after cascading cluster analysis of all information, we get Figure 5 The results shown in the figure are as follows. The black dots in the figure are the positive classes determined by the algorithm. The analysis results are displayed on the original image, and we can get Figure 7 It can be seen that all the transmission lines in the figure are detected, and there are no interference lines, but it contains the influence of repeated detection.

[0064] (5) Calculate the distances between all straight lines according to the straight-line distance formula and remove duplicate straight lines; specifically, the following steps are included:

[0065] (5.1) Load the transmission line image and line parameters;

[0066] (5.2) Estimate the distance between the first straight line and other straight lines; the distance estimation formula is:

[0067]

[0068] In the formula, A, B, and C are the parameters of the general form of a straight line in a two-dimensional rectangular coordinate system, where (x i ,y i ) is any point on one of the straight lines, n is the number of points, is the estimated distance. By setting the threshold D, when It is believed that there is a problem of duplicate detection.

[0069] (5.3) Filter all the lines whose distance is less than the threshold, select the longest line to keep, and delete the other lines;

[0070] (5.4) Estimate the distance between the second straight line and other straight lines;

[0071] (5.5) Repeat steps (5.3) and (5.4) until the last straight line is calculated;

[0072] (6) The resulting straight line is used to predict the direction of the transmission line based on the calculated angle position information, and the transmission line is marked with a rectangular box. The result is as follows: Figure 8 As shown, all transmission lines were successfully marked.

[0073] The present invention enhances the strong edge features belonging to the edge of the transmission line through the Hessian matrix, detects the straight line by the cumulative probability Hough transform, and then enhances the robustness of the algorithm by cascading cluster analysis of the length, angle, and position information of the straight line in the image. Finally, after repeated line segment screening, a result graph with high accuracy is obtained. In order to demonstrate the scope of application of the present invention, the method is verified in other contexts, such as Fig. 9 shown.

Claims

1. A multi-information cascade clustering transmission line detection method, characterized in that: The following steps are involved: (1) Obtain the transmission line image and perform grayscale transformation on the image; (2) Use edge enhancement algorithm to preprocess the image after grayscale transformation; (3) Perform cumulative probability Hough transform on the enhanced image to extract candidate straight lines in the image; (4) Perform cascade cluster analysis on the straight line information based on the length, angle, and position information of the straight line; (5) Calculate the distances between all straight lines according to the straight-line distance formula and remove duplicate straight lines; (6) The resulting straight line is used to predict the direction of the transmission line based on the calculated angle position information, and the transmission line is marked with a rectangular frame; In step (2), the edge enhancement algorithm based on the Hessian matrix is ​​used on the grayscale transformed image to extract edge features belonging to strong edges in the image, thereby enhancing the edge of the transmission line; The calculation formula of the Hessian matrix is: Where f is the function of the image with respect to the coordinates (x, y) xx , f xy , f yx , f yy They represent the second-order partial derivatives of function f with respect to x, y and their mixtures respectively; the eigenvalues ​​and eigenvectors of the Hessian matrix reflect the intensity and direction of the change in the grayscale curvature of the image; In step (4), the cascade cluster analysis of the straight line information specifically includes the following steps: (4.1) Use clustering algorithm to classify the length, angle and position information of the straight line; (4.2) Mark the positive class in the line according to the number of votes in step (3); (4.3) Perform hard cascade analysis on the results and perform intersection operation on the obtained positive classes. When all the straight lines are classified as positive classes in the three types of information, the straight lines are marked as transmission lines.

2. The multi-information cascade clustering transmission line detection method according to claim 1, characterized in that: In step (1), the transmission line image is obtained and the image is gray-scaled. Specifically, a camera is installed on the top of the nose of the drone, and the camera is used to capture images during the flight of the drone. The captured images are transmitted to the processor for gray-scale conversion.

3. The multi-information cascade clustering transmission line detection method according to claim 1, characterized in that: In step (3), the cumulative probability Hough transform is performed on the enhanced image to extract candidate straight lines in the image, which specifically includes the following steps: (3.1) Within the specified range, the edge pixel information after image enhancement is mapped to the Hough space; (3.2) Vote for each edge pixel mapped to the Hough space; (3.3) The set of suspected points in the plane within the accumulation range and the points exceeding the set threshold represent a straight line.

4. The multi-information cascade clustering transmission line detection method according to claim 1, characterized in that: In step (5), the distances between all straight lines are calculated according to the straight-line distance formula, and duplicate straight lines are removed, which specifically includes the following steps: (5.1) Load the transmission line image and line parameters; (5.2) Estimate the distance from the first straight line to the other straight lines; (5.3) Filter all the lines whose distance is less than the threshold, select the longest line to keep, and delete the other lines; (5.4) Estimate the distance between the second straight line and other straight lines; (5.5) Repeat steps (5.3) and (5.4) until the last straight line is calculated.

5. The multi-information cascade clustering transmission line detection method according to claim 4, characterized in that: The distance estimation formula is: In the formula, A, B, and C are the parameters of the general form of a straight line in a two-dimensional rectangular coordinate system, where (x i ,y i ) is any point on one of the straight lines, n is the number of points, is the estimated distance; by setting the threshold D, when It is believed that there is a problem of duplicate detection.

Images