A method for identifying mountain fires on overhead lines

Abstract

The present invention belongs to the field of intelligent operation and inspection of overhead lines, and in particular relates to a method for identifying wildfires on overhead lines. The present invention uses the HSV color space to analyze the color components of the image background, and uses a specific color threshold to filter and extract pictures suspected of containing wildfires, effectively overcoming the defect that color information of night pictures is not easy to identify, and improving the recognition rate of wildfires at night; regional segmentation is performed by symmetric cross entropy fusion adaptive threshold segmentation technology, and expansion and corrosion morphological image processing methods are used to remove interference from similar color components; connected domains are selected by calculating the distribution differences of different regions through KL divergence, and then adjacent contours are merged, which can further remove connected domains that do not meet the area threshold requirements, and the contours of the remaining specific regions are circumscribed with a minimum rectangle, and the rectangular coordinates are returned to accurately identify the wildfire hazard area.

Description

Technical Field

[0001] The invention belongs to the field of intelligent operation and inspection of overhead lines, and in particular relates to a method for identifying mountain fires on overhead lines. Background Art

[0002] With the upgrading of transmission line maintenance technology, visual remote inspection of transmission line channels has been widely used. On the transmission lines, monitoring equipment takes static pictures at regular intervals to detect hidden dangers in the transmission channels. However, compared with other hidden dangers, forest fire hazards have always been one of the important factors affecting transmission safety.

[0003] In recent years, with the development of deep learning technology, some methods for identifying overhead line wildfires based on convolutional neural networks have emerged. However, deep learning models require a large number of samples for training. Since night scenes are quite different from daytime scenes, they are easily affected by lighting. In addition, the characteristics of night wildfires are also quite different from other scenes. In the RGB space, the color of the center of the flame is white, and the color of the edge area is between red and orange. In addition, it is difficult to collect scene samples at night. Therefore, the model has poor recognition effect for night scenes and cannot achieve good recognition effect. In the early days, color information in RBG space was directly used for wildfire recognition, but the recognition effect was poor and objects with similar colors to wildfires could not be distinguished.

[0004] In summary, how to provide a fast and reliable method for identifying wildfires on overhead lines and reduce the missed reporting rate of wildfire hazard identification at night is one of the problems that technicians in this field need to solve urgently. Summary of the invention

[0005] In view of the shortcomings of the prior art, the purpose of the present invention is to provide a method for identifying wildfires on overhead lines. The present invention uses the HSV color space to analyze the color components of the image background, and uses a specific color threshold to filter and extract pictures suspected of containing wildfires, which effectively overcomes the defect that the color information of night pictures is not easy to identify, and improves the recognition rate of wildfires at night; regional segmentation is performed by symmetric cross entropy fusion adaptive threshold segmentation technology, and the expansion and corrosion morphological image processing methods are used to remove the interference of similar color components; the KL divergence is used to calculate the distribution differences of different regions to select connected domains, and the adjacent contours are merged, which can further remove the connected domains that do not meet the area threshold requirements, and the contours of the remaining specific areas are circumscribed with the minimum rectangle, and the rectangular coordinates are returned to accurately identify the wildfire hazard area.

[0006] The technical problem to be solved by the present invention is achieved by adopting the following technical solution: A method for identifying mountain fires on overhead lines, comprising the following steps:

[0007] a. Obtain the static image transmitted by the monitoring device and convert the color space to convert the RGB image into the HSV space. HSV is a relatively intuitive color model. It expresses color changes in a form that is easier for humans to perceive based on changes in hue, brightness, and color purity. There is no direct relationship between color brightness and light intensity. Therefore, in night scenes, color features will be particularly prominent. HSV can represent more subtle changes in color areas. By converting the representation space into the HSV color space, filtering the specific color areas of night wildfires can be performed, which can well distinguish the wildfire color areas in night scenes.

[0008] b. Statistically calculate the HSV average component of the standard night picture, find the optimal parameter corresponding ratio, select the night wildfire specific color component to perform color filtering on the static image to extract the picture suspected of containing wildfire; the optimal parameter ratio is the color interval of the wildfire in the HSV space, and the specific selection method is calculated through experimental values. Specifically, take the average from multiple pictures to find the color interval under the current power transmission night scene, in order to find the color interval corresponding to the wildfire, and then segment the wildfire area according to this interval;

[0009] c. Improve the OTSU threshold segmentation method through symmetric cross entropy to eliminate interference areas in images suspected of containing wildfires: perform threshold segmentation to separate specific areas suspected of containing wildfires and obtain binary images; use morphological processing methods to process binary images; since OTSU calculates the threshold based on the inter-class variance method, it will only have a good effect on images with a single peak in inter-class variance. However, when the size ratio of the target to the background is very different, such as excessive background noise, variable edge shape of the flame, etc., it will present multiple peaks, and symmetric cross entropy can better measure the degree of difference in pixel probability distribution. Therefore, using symmetric cross entropy combined with the OTSU method to perform threshold segmentation can make the edge of the segmented image clearer;

[0010] d. Draw the outline of the binary image obtained in step c, select the connected domain by KL divergence, and extract the outline of the specific area. Calculate the area of ​​the approximate outline through the outline points to obtain the polygonal outline area S. Use the area threshold to remove the specific area whose polygonal outline area S does not meet the requirements. After cyclically calculating all the outlines, circumscribe the outline of the remaining specific area with the minimum rectangle, return the rectangular coordinates, and obtain the wildfire hazard area; Use the threshold segmentation method to extract the outline of the specific filter area. Due to the changeable shape of wildfires, calculate the pixel distribution of different neighborhoods by KL divergence, select the connected domain and calculate the outline area, find the outline above the area threshold, delete the interfering outline, and return the coordinate value by circumscribing the minimum rectangle, so as to accurately find the location where the wildfire hazard target occurs and quickly identify it;

[0011] Specifically, after filtering the binary image, the outer contour is detected, and the coordinate points of all approximate contours are retained;

[0012] The eight-neighborhood connectivity method is used to screen the extracted contour points, and the KL divergence of different areas is calculated in blocks to select specific connected domains; KL divergence can measure the distribution differences in different areas. The specific area separated in step b will have many segmented areas, but adjacent areas may be the same target area. The KL divergence is used to select the connected domain, that is, if the distribution difference of the connected areas is small, then the adjacent areas can be connected as a whole area, thereby further improving the recognition accuracy of the flame area. Here, the eight-neighborhood connectivity method is used to judge the difference value from 8 directions, which will have higher accuracy.

[0013] The technical solution of the present invention also includes: in the step b, in the HSV characterization space, the night wildfire specific color component range P is [(5, 12), (43, 255), (46, 255)], and color filtering is performed according to the night wildfire specific color component range P to obtain a filtered image that only exists within the component range, and the night wildfire specific color component range is obtained by statistically calculating the HSV components of the wildfire area in multiple night power transmission scene images.

[0014] The technical solution of the present invention also includes: in step c, the threshold segmentation formula after the symmetric cross entropy improves the OTSU threshold segmentation method is as follows:

[0015] ζ(t)=P a (t)μ a (t)log(μ a (t))+P b (t)μ b (t)log(μ b (t))(I)

[0016] In formula (I), ζ(t) represents the difference between the target image and the background image, t represents the threshold used for segmenting the image, and P a (t) The probability of the corresponding target area image, P b (t) corresponds to the probability of the background area image, μ a (t) and μ b (t) are the grayscale means corresponding to the target area image and the background area image respectively;

[0017] Threshold segmentation through distance measurement can be more robust to targets with variable shapes such as wildfires, separate specific areas, and eliminate noise interference; the OTSU threshold segmentation method is improved by symmetric cross entropy to exclude interference areas and obtain the threshold T. Specifically, the number of each pixel in the grayscale in the entire image is counted, and the probability distribution P of different areas in the entire image can be calculated. The grayscale is traversed and searched, and the probability between the foreground and background classes under the current grayscale value is calculated. Finally, the threshold T corresponding to the intra-class and inter-class variance is calculated, so that threshold segmentation can be performed according to the threshold T to exclude interference areas; the symmetric cross entropy is used to calculate the difference between the foreground and background in the image, and then the threshold T that can distinguish the foreground and background is found. The segmentation effect will be more refined, some interference areas will be removed, and the edge areas will be better segmented. Compared with the OTSU threshold segmentation method, it has stronger noise resistance, and the OTSU threshold segmentation method improved by symmetric cross entropy can reduce the computational complexity and can segment the image faster.

[0018] The morphological processing method specifically includes corrosion and dilation operations in sequence, first removing small areas of similar color components and then expanding the target area contour; in order to remove the interference of some similar color components such as lights, the interference area is eliminated through image processing methods such as dilation and corrosion. Lights of similar color components are generally small areas as noise points. Corrosion is to use the surrounding pixel areas to remove noise points, the overall contour will be reduced, and then expanded to the original size. The noise points are removed because the connected domain is too small, and they disappear during expansion, so that only the target area is retained; multiple fire points can be distinguished according to the image segmentation threshold as long as the connected domains are not adjacent, and they need to be segmented if they meet the threshold T.

[0019] The technical solution of the present invention also includes: in step d, the area threshold includes the minimum wildfire area threshold P1 and the street lamp area value LP, and the specific area whose polygon contour area S is less than the minimum wildfire area threshold P1 and the specific area equal to the street lamp area value LP is removed.

[0020] Although the interference areas of some lights with similar color components can be removed through threshold segmentation and morphological processing methods, some light color components overlap with wildfire color components and cannot be completely removed; and some wildfire areas are too far away from the monitoring equipment to accurately identify their specific locations, that is, they pose no threat to the current transmission lines. In order to further eliminate these interference factors, the outline areas of wildfires and street lights of similar colors in multiple transmission night scenes are pre-calculated to obtain the minimum wildfire area threshold P1 and the street light area value LP; on the one hand, because the color components and outline areas of all street lights in the transmission scene are basically the same, and the street light position is fixed, the outer outline will not change, and the calculated current outline area is compared with the street light area. The preset street lamp area value LP is compared and judged, and the outline of the light area with the same area value LP as the street lamp, that is, the light area where the color component overlaps with the wildfire color component, is removed; on the other hand, based on the actual scenario, each transmission line has a certain number of monitoring equipment points. Since the monitoring equipment has a certain monitoring distance and accuracy, and the area that is too far away from one monitoring equipment will be close to another monitoring equipment point, the minimum wildfire area threshold P1 is set. When the monitoring area is too far away, that is, the image area is smaller than the set threshold, it will be filtered out, which can more quickly identify the target area and its specific location, that is, determine the nearest monitoring equipment, avoid repeated identification, improve recognition efficiency, and at the same time improve recognition accuracy.

[0021] The present invention also provides an application of an overhead line wildfire identification method for wildfire graded warning. Specifically, by calculating the area of ​​the smallest rectangle circumscribing the wildfire hazard area, the severity level is divided into general and emergency according to the area value, and an alarm message is sent according to the severity level.

[0022] Beneficial effects of the present invention:

[0023] The present invention processes night pictures in the HSV space, and extracts pictures suspected of containing wildfires by filtering the specific color components of wildfires at night, which effectively overcomes the defect that the color information of night pictures is difficult to identify, reduces the number of later picture processing, and improves the efficiency of wildfire identification.

[0024] The present invention performs threshold segmentation on images suspected of containing wildfires by improving the OTSU threshold segmentation method through symmetric cross entropy, which can clearly separate specific areas suspected of containing wildfires, and remove some noise points through morphological processing methods, which can make the edges of specific areas suspected of containing wildfires clearer, effectively overcoming the defects of unclear and changeable edges of wildfire images.

[0025] Finally, the outline of the specific area suspected of containing wildfires is drawn, and the connected domains are selected and merged through KL divergence. By calculating the area of ​​the merged outline, the connected domains that do not meet the area threshold requirements can be further removed, and the outline of the remaining specific area is circumscribed with the minimum rectangle, and the rectangular coordinates are returned to accurately identify the wildfire hazard area, providing technical support for the detection of transmission line hazards.

[0026] The present invention also analyzes the severity of wildfires by the size of the area of ​​the smallest rectangle circumscribing the wildfire hazard area, and carries out graded early warning, which can provide more on-site information for subsequent processing and facilitate adjustment of the processing plan. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a schematic diagram of the flow chart of the overhead line mountain fire identification method of the present invention;

[0028] Figure 2 A schematic diagram of the HSV color space according to an embodiment of the present invention;

[0029] Figure 3 is a color filter graph according to an embodiment of the present invention;

[0030] Figure 4 is an interference area processing diagram of an embodiment of the present invention,

[0031] Figure 5 is a contour detection diagram according to an embodiment of the present invention;

[0032] Figure 6 2 is a diagram showing the detection results of an embodiment of the present invention. DETAILED DESCRIPTION

[0033] In order to enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0034] like Figure 1 As shown, a method for identifying a mountain fire on an overhead line comprises the following steps:

[0035] a. Obtain the static image transmitted by the monitoring device and convert the color space to convert the RGB image into the HSV space.

[0036] b. Statistically calculate the HSV average component of the standard night picture, find the optimal parameter corresponding ratio, select the night wildfire specific color component to perform color filtering on the static image to extract pictures suspected of containing wildfires.

[0037] In step b, in the HSV representation space, the range P of the specific color component of the night wildfire is [(5, 12), (43, 255), (46, 255)]. Colors are filtered according to the range P of the specific color component of the night wildfire to obtain a filtered image that only exists within the component range.

[0038] c. Improve the OTSU threshold segmentation method through symmetric cross entropy to eliminate interference areas in pictures suspected of containing wildfires: perform threshold segmentation to separate specific areas suspected of containing wildfires and obtain a binary image.

[0039] The threshold segmentation formula after the symmetric cross entropy improved OTSU threshold segmentation method is as follows:

[0040] ζ(t)=P a (t)μ a (t)log(μ a (t))+P b (t)μ b (t)log(μ b (t))(I)

[0041] In formula (I), ζ(t) represents the difference between the target image and the background image, t represents the threshold used for segmenting the image, and P a (t) The probability of the corresponding target area image, P b (t) corresponds to the probability of the background area image, μ a (t) and μ b (t) are the grayscale mean values ​​of the target area image and the background area image, respectively. According to the image binarization, the OTSU threshold segmentation method is used to improve the defect of insensitivity to the target shape. The symmetric cross entropy is added for distance measurement, that is, the cross entropy is used to measure the difference. The asymmetric cross entropy does not meet the symmetry of the distance measurement. Compared with the asymmetric cross entropy, the symmetric cross entropy has better noise resistance and runs faster. The edge of the wildfire has motion characteristics. Because the cross entropy can describe the degree of difference in probability distribution, the threshold segmentation is performed based on the cross entropy. The flame shape is irregular. OTSU itself only roughly separates the target and the background. Irregular targets cannot measure the edge detail information well. The symmetric cross entropy has good robustness to the change of the shape of the moving area by comparing the images before and after segmentation. It can accurately extract the moving target of a smaller area, that is, the image with irregular shape edges, through the distribution probability of the grayscale value of the pixel point.

[0042] The binary image is processed using morphological processing methods, which specifically include sequential erosion and dilation operations, first removing small areas of similar color components, and then expanding the contour of the target area.

[0043] d. Draw the outline of the binary image obtained in step c, select the connected domain using KL divergence, extract the outline of the specific area, calculate the area of ​​the approximate outline through the outline points, obtain the polygonal outline area S, use the area threshold to remove the specific area whose polygonal outline area S does not meet the requirements, and after looping all the outlines, circumscribe the outline of the remaining specific area with the minimum rectangle, return the rectangular coordinates, and obtain the wildfire hazard area. In step d, the area threshold includes the wildfire minimum area threshold P1 and the street lamp area value LP, and the specific area whose polygonal outline area S is less than the wildfire minimum area threshold P1 and the specific area equal to the street lamp area value LP are removed.

[0044] An application of an overhead line wildfire identification method is used for wildfire graded early warning. Specifically, by calculating the area of ​​the smallest rectangle circumscribing the wildfire hazard area, the severity level is divided into general and emergency according to the area value, and an alarm message is sent according to the severity level.

[0045] Example 1

[0046] In a certain province, there is a risk of wildfire at night in the area under the transmission line. A series of abnormal images collected by the front-end device are used to detect the hidden danger.

[0047] The process of testing abnormal images is as follows:

[0048] Image name to be tested: 99000843200301_20210304163003.jpg

[0049] 1) Convert the HSV color space, analyze the color components of a specific target, and process the image as shown below: Figure 2 express.

[0050] 2) Continue to filter the color components of the specific target area and perform OTSU threshold segmentation on the image. The effect of the processed image is shown in the figure below: Figure 3 express.

[0051] 3) Corrosion and expansion are performed on the segmented image to remove interference from irrelevant areas. The effect of the processed image is shown in the figure below: Figure 4 shown.

[0052] 4) Detect the outer contour, retain all contour points, select the connected domain and draw the outer contour, and the effect diagram of the extracted contour is as follows: Figure 5 express.

[0053] Calculate the outer contour area, find the smallest circumscribed rectangle of the specific target area, and identify the wildfire hazard in the original image. The detection results are as follows: Figure 6 express.

[0054] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for identifying mountain fires on overhead lines, characterized in that: The following steps are involved: a. Obtain the static image transmitted by the monitoring device and convert the color space to convert the RGB image into the HSV space; b. Statistically calculate the HSV average component of the standard night picture, select the night wildfire specific color component to perform color filtering on the static image to extract the picture suspected of containing wildfire; c. Improve the OTSU threshold segmentation method through symmetric cross entropy to eliminate interference areas in images suspected of containing wildfires: perform threshold segmentation to separate specific areas suspected of containing wildfires and obtain binary images; use morphological processing methods to process the binary images; d. Draw the contour of the binary image obtained in step c, use KL divergence to select the connected domain, and extract the contour of the specific area. Calculate the area of ​​the approximate contour through the contour points to obtain the polygonal contour area S. Use the area threshold to remove the specific areas whose polygonal contour area S does not meet the requirements. After looping all contours, enclose the contours of the remaining specific areas in the minimum rectangle, return the rectangle coordinates, and obtain the wildfire hazard area.

2. The method for identifying an overhead line fire according to claim 1, characterized in that: In step b, in the HSV representation space, the range P of the specific color component of the night wildfire is [(5, 12), (43, 255), (46, 255)]. Colors are filtered according to the range P of the specific color component of the night wildfire to obtain a filtered image that only exists within the component range.

3. The method for identifying an overhead line fire according to claim 1, characterized in that: In step c, the threshold segmentation formula after the symmetric cross entropy improved OTSU threshold segmentation method is as follows: ζ(t)=P a (t)μ a (t)log(μ a (t))+P b (t)μ b (t)log(μ b (t)) (I) In formula (I), ζ(t) represents the difference between the target image and the background image, t represents the threshold used for segmenting the image, and P a (t) The probability of the corresponding target area image, P b (t) corresponds to the probability of the background area image, μ a (t) and μ b (t) are the grayscale means corresponding to the target area image and the background area image respectively; The morphological processing method specifically includes performing erosion and dilation operations in sequence, first removing small areas of similar color components, and then expanding the outline of the target area.

4. The method for identifying an overhead line fire according to claim 1, characterized in that: In step d, the area thresholds include the minimum wildfire area threshold P1 and the street lamp area value LP, and the specific areas whose polygonal contour areas S are smaller than the minimum wildfire area threshold P1 and equal to the street lamp area value LP are removed.

5. A method for identifying an overhead line wildfire as described in any one of claims 1 to 4, characterized in that: Used for graded wildfire warning. Specifically, by calculating the area of ​​the smallest rectangle circumscribing the wildfire hazard area, the severity level is divided into general and emergency according to the area value, and an alarm message is sent according to the severity level.

Images