A monitoring method for determining whether a target contacts a transmission line

Abstract

The present invention relates to the technical field of image recognition, and discloses a monitoring method for determining whether a target contacts a transmission line, including the following steps: Step 1, collect real-time images of the area where the transmission line is located in real time as background pictures, update the real-time images into background pictures by using a background update strategy, and separate the foreground and background of the background pictures by using an improved Grabcut algorithm; Step 2, determine whether there is a target intrusion in the real-time images; Step 3, predict the operation trend of the intruding target and the detection area demarcated on the transmission line, and feedback and transmit the prediction result to an information feedback system. If the prediction result exceeds a given threshold, it is an abnormal situation, and the information feedback system issues an alarm reminder. By using the improved Grabcut algorithm to separate the image background and foreground, the present invention obtains an image containing only the target for recognition, inputs it into an image recognition system for calculation, reduces the calculation amount, and improves the picture quality.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition, and particularly to a monitoring method for determining whether a target contacts a transmission line. Background Art

[0002] Transmission lines play a core role in the power grid. The safe and reliable operation of transmission lines is related to the operation safety of the entire power system itself. For the situation of external force damage to transmission lines, manual inspections and traditional monitoring devices cannot give early warnings and alarms in a timely and effective manner. Transmission lines are in the wild, with frequent accident hazards and a wide coverage area. In high-altitude and high-plateau areas, the climate, geographical environment are harsh and changeable, and the oxygen is thin, making it more difficult to carry out manual inspections.

[0003] At present, the combination of the development of ultra-high-definition video technology and artificial intelligence is more and more widely used, and it is gradually applied to the foreign object recognition technology of transmission lines, making the development of intelligent inspection technology more perfect. However, for existing images, it is difficult to extract target features, the image samples are insufficient, the environment in high-altitude and high-plateau areas is complex, and the movement of clouds in the sky and the change of light intensity will bring computational complexity and uncertainty to the image recognition operation. Due to the unclear background, inaccurate recognition of the target, and unclear segmentation between the target and the background, misjudgment is likely to occur when the background moves. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a monitoring method for determining whether a target contacts a transmission line for the problem of image recognition in a complex environment. The background segmentation method is adopted. When there is no target in the image, the background is updated, reducing the number of times of using the image recognition algorithm and avoiding power consumption. When a target contour is detected in the image, combined with the image recognition technology, the target is accurately recognized, the abnormal state of the target is checked, and an alarm notification is sent.

[0005] The present invention is realized through the following technical solutions:

[0006] A monitoring method for determining whether a target contacts a transmission line includes the following steps:

[0007] Step 1: Real-time collect the real-time image of the area where the transmission line is located as the background picture, update the real-time image to the background picture by using the background update strategy, and separate the foreground and background of the background picture by using an improved Grabcut algorithm;

[0008] Step 2: Determine whether there is a target intrusion in the real-time image. If there is a target intrusion in the real-time image, use an edge detection algorithm to extract the contour of the target, and use an image recognition algorithm to perform recognition processing on the target to determine the target type;

[0009] Step 3: Predict the operation trend of the intruding target and the detection area defined on the transmission line, determine whether the target is approaching or staying on the transmission line, and feedback and transmit the prediction result to the information feedback system. If the prediction result exceeds the given threshold, it is an abnormal situation, and the information feedback system issues an alarm reminder.

[0010] As an optimization, the improved Grabcut algorithm uses a probabilistic neural network model PNN to replace the Gaussian mixture model, and then uses morphological opening operation and morphological dilation algorithm to reduce the holes in the picture, making the internal and edge information of the image clearer.

[0011] As an optimization, in Step 1, the specific steps for the improved Grabcut algorithm to separate the foreground and background of the background picture are as follows:

[0012] Step 1.1: Perform preprocessing operations on the background image. Take the internal pixels of the detection area defined on the transmission line as the foreground, and the external pixels of the detection area defined on the transmission line as the background. Establish the gray histograms of the foreground and background, select the pixels with a higher proportion of pixel values as the input samples for training, and input them into the PNN model;

[0013] Step 1.2: Use the input layer in the PNN model to receive the input samples;

[0014] Step 1.3: Use the hidden layer to calculate the distance between the input sample and the center. The input sample x is input into the hidden layer, and the j-th center x of the i-th class in the hidden layer ij The determined input-output relationship is:

[0015]

[0016] where σ is the smoothing factor, d is the dimension of the sample space data, the network is trained with samples, the input samples are divided into i classes, when inputting data, the PNN network identifies the image as background or foreground, i = 0 represents background, and 1 represents foreground;

[0017] Step 1.4: Weight and average the output values of the foreground or background belonging to the same class in the hidden layer and then output. S i represents the relationship between the test data and the foreground or background, and L i represents the number of foreground or background hidden layer centers;

[0018]

[0019] Step 1.5: The output layer normalizes Step 1.4, and takes the largest output value as the output foreground or background probability value argmax(S i );

[0020] Step 1.6: Update the energy function E(·) to solve the weights of t-links using the PNN model, perform image output, and improve the algorithm efficiency;

[0021] E(α, σ, z) = S(α, σ, z) + V(α, z);

[0022] Among them, S(·) is the energy value for cutting t-links;

[0023]

[0024]

[0025] Among them, L n is the number of centers of the nth hidden layer;

[0026] Step 1.7: Perform morphological operations on the output image. After using the opening operation, then use the dilation algorithm to avoid holes in the output image and make the edges of the output image clear.

[0027] As an optimization, the background update strategy is specifically as follows:

[0028] Step 1.8: If the background picture is the same as the real-time picture, update the real-time picture into a new background picture. At the same time, store the old background picture as a historical picture in the memory;

[0029] Step 1.9: If the background picture is different from the real-time picture, then compare the real-time picture with the historical picture. If the real-time picture is similar to the historical picture, update the real-time picture into a new background picture. At the same time, store the old background picture as a historical picture in the memory;

[0030] Step 1.10: If the real-time picture is different from both the background picture and the historical picture, maintain the original background picture.

[0031] As an optimization, in Step 2, the region of interest where the target intrudes is delimited by the ROI algorithm.

[0032] As an optimization, determining whether there is a target intrusion in the real-time image is specifically as follows:

[0033] If there is a target intersecting with the region of interest, it is determined that there is a target intrusion in the real-time image. Use the edge detection algorithm to identify the real-time contour of the target, and frame the contour of the target. When the contour line of the target overlaps with the region of interest, perform image recognition algorithm recognition on the target to determine the target type.

[0034] As an optimization, the Canny edge detection algorithm is used as the edge detection algorithm to obtain the outer contour line of the target.

[0035] As an optimization, the Canny edge detection algorithm determines the optimal threshold T by using the Otsu method, and then calculates the high threshold and low threshold of the Canny edge detection algorithm. Among them, the high threshold is T, and the low threshold is T / 2.

[0036] As an optimization, the detection area on the transmission line is delimited by using the Hough line transform algorithm. Specifically: first, the straight lines on the transmission line are detected, and the length and width of the detected straight lines are expanded outward by a certain distance to form a detection rectangle frame, and when the target invades the rectangle frame, an image recognition operation is performed on the target.

[0037] As an optimization, the pre-judgment result is: the minimum straight-line distance between the contour line of the target and the straight line perpendicular to the rectangle frame. If the minimum straight-line distance is less than the threshold, it is determined that the target is approaching the transmission line; if the minimum straight-line distance is 0, it is determined that the target stays on the transmission line.

[0038] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0039] The present invention separates the image background and foreground by using an improved Grabcut algorithm to obtain an image containing only the target for recognition, and inputs it into the image recognition system for calculation, reducing the calculation amount and improving the picture quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the drawings required for the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention, and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings. In the drawings:

[0041] Figure 1 is a flowchart of a monitoring method for determining whether a target touches a transmission line according to the present invention;

[0042] Figure 2 is a specific flowchart of a monitoring method for determining whether a target touches a transmission line according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0043] To make the purpose, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and the drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0044] Embodiment

[0045] As shown Figure 1-2 in the figure, a monitoring method for determining whether an object contacts a power transmission line includes the following steps:

[0046] Step 1. Collect real-time images of the area where the power transmission line is located as background images in real time. The host computer collects real-time images every T moments, updates the real-time images into background images by using a background update strategy, and separates the foreground and background of the background images by using an improved GrabCut algorithm.

[0047] In this embodiment, the improved GrabCut algorithm uses a probabilistic neural network model PNN to replace the Gaussian mixture model, and then uses morphological opening operation and morphological dilation algorithm to reduce the holes in the image, making the internal and edge information of the image clearer.

[0048] Meanwhile, the specific steps for the improved GrabCut algorithm to separate the foreground and background of the background image are as follows:

[0049] Step 1.1. Perform preprocessing operations on the background image, take the internal pixels of the detection area delimited on the power transmission line as the foreground, take the external pixels of the detection area delimited on the power transmission line as the background, establish the gray histograms of the foreground and background, select the pixels with a relatively high proportion of pixel values as the input samples for training, and input them into the PNN model;

[0050] Step 1.2. Use the input layer in the PNN model to receive the input samples;

[0051] Step 1.3. Use the hidden layer to calculate the distance between the input samples and the centers. The input sample x is input into the hidden layer, and the j-th center x of the i-th class in the hidden layer ij The determined input-output relationship is:

[0052]

[0053] where σ is the smoothing factor, d is the dimension of the sample space data, the network is trained with samples, the input samples are divided into i classes, when inputting data, the PNN network identifies whether the image is background or foreground, i = 0 represents background, and 1 represents foreground;

[0054] Step 1.4. Output the weighted average of the output values of the foreground or background belonging to the same class in the hidden layer. S i represents the relationship between the test data and the foreground or background, and L i represents the number of foreground or background hidden layer centers;

[0055]

[0056] Step 1.5: The output layer normalizes Step 1.4 and takes the largest output value as the foreground or background probability value of the output, argmax(S i );

[0057] Step 1.6: Update the energy function E(·), so as to solve the weights of t-links using the PNN model, perform image output, and improve the algorithm efficiency;

[0058] E(α, σ, z) = S(α, σ, z) + V(α, z);

[0059] where S(·) is the energy value for cutting t-links;

[0060]

[0061]

[0062] where L n is the number of centers of the nth hidden layer;

[0063] Step 1.7: Perform morphological operations on the output image. After using the opening operation, then use the dilation algorithm to avoid holes in the output image and make the edges of the output image clear.

[0064] In this embodiment, the background update strategy is specifically as follows:

[0065] Step 1.8: If the background picture is the same as the real-time picture, update the real-time picture into a new background picture. At the same time, store the old background picture as a historical picture in the memory;

[0066] Step 1.9: If the background picture is different from the real-time picture, then compare the real-time picture with the historical picture. If the real-time picture is similar to the historical picture, update the real-time picture into a new background picture. At the same time, store the old background picture as a historical picture in the memory;

[0067] Step 1.10: If the real-time picture is different from both the background picture and the historical picture, maintain the original background picture.

[0068] Step 2: Determine whether there is a target intrusion in the real-time image. If there is a target intrusion in the real-time image, use an edge detection algorithm to extract the contour line of the target, and use an image recognition algorithm to identify and process the target to determine the target type;

[0069] Define the range of target intrusion in the real-time image through the ROI algorithm.

[0070] In this embodiment, the range of the target intrusion can be delimited by the ROI algorithm. The range of the target intrusion is the region of interest. When the region of interest contacts the contour of the target, that is, when there is an intersection between the target and the region of interest, it is determined that there is a target intruding into the region of interest in the real-time image. The edge detection module is used to quickly identify the real-time contour of the target, and the contour of the target is framed. Then, the target is identified by an image recognition algorithm to determine the target type. The image is introduced into a deep learning algorithm for image recognition operations; the image recognition algorithm returns the image category and coordinates.

[0071] In this embodiment, the edge detection algorithm uses the Canny edge detection algorithm to obtain the outer contour line of the target.

[0072] In this embodiment, the Canny edge detection algorithm determines the optimal threshold T by using the OTSU method, and then calculates the high threshold and low threshold of the Canny edge detection algorithm. Among them, the high threshold is T, and the low threshold is T / 2.

[0073] Step 3: Predict the operation trend of the target and the detection area delimited on the transmission line, determine whether the target is approaching or staying on the transmission line, and feedback the prediction result to the information feedback system. If the prediction result exceeds the threshold, it is an abnormal situation, and the information feedback system issues an alarm reminder.

[0074] In this embodiment, the area to be detected of the transmission line is delimited by using the Hough line transform algorithm.

[0075] In this embodiment, the Hough transform line detection algorithm is specifically as follows: First, the straight lines on the transmission line are detected, and the length and width of the detected straight lines are expanded outward by a certain distance to form a detected rectangular frame. When the target invades the rectangular frame, image recognition operations are performed on the target.

[0076] In this embodiment, specifically how much distance to expand is set according to the actual situation.

[0077] In this embodiment, the prediction result is: the minimum straight-line distance between the contour line of the target and the straight line perpendicular to the rectangular frame. If the minimum straight-line distance is less than the given threshold, it is determined that the target is approaching the transmission line; if the minimum straight-line distance is 0, it is determined that the target is staying on the transmission line. Set the threshold for the target's distance from the transmission line. If it exceeds the threshold, it is judged that the state is abnormal and an alarm notification is triggered; otherwise, it is considered that the state of the detected target is normal.

[0078] If the image coordinates exceed the threshold with respect to the position of the transmission line, an alarm is issued. If it does not exceed the threshold, image recognition operations are continued on the image.

[0079] It should be noted that the ROI algorithm, Canny edge detection algorithm, and Hough transform line detection algorithm are all existing technologies, and the specific detection processes will not be elaborated here.

[0080] Regarding the problem that a large amount of manual inspection is required for traditional video surveillance systems to detect abnormal situations, there are problems such as complex deep learning algorithms, long target recognition time, and delayed alarms in case of abnormal situations. The present invention can more accurately identify abnormal targets and give timely warnings in case of abnormalities.

[0081] For existing video surveillance systems, severe interference from the background increases the computational load for target detection, reduces the accuracy, and results in unsatisfactory recognition effects. The non-update of the background causes the inability to recognize various real-time detection situations, and targets that have invaded the recognition area cannot be recognized. Prolonged use of image recognition algorithms increases the power calculation pressure and algorithm complexity, leading to an increase in the power consumption of the power system and making the power system overwhelmed. The present invention uses an improved Grabcut algorithm for background segmentation. Background segmentation can improve the image quality, quickly lock in suddenly emerging targets, and improve the recognition sensitivity. When detecting an object invasion, the contour of the target is obtained through edge detection. When the target contour appears in the image, image recognition technology is used to recognize the target. The present invention accurately frames the contour of the image and filters out the background. At the same time, it can simplify the image recognition algorithm, reduce power consumption, accurately recognize suddenly emerging targets, simplify the task of image recognition, and improve the detection accuracy.

[0082] The specific embodiments described above further elaborate on the purpose, technical solutions, and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A monitoring method for determining whether a target contacts a transmission line, characterized in that, It includes the following steps: Step 1: Real-time collect the real-time image of the area where the transmission line is located as the background image, update the real-time image as the background image using the background update strategy, and separate the foreground and background of the background image using the improved Grabcut algorithm; In the improved Grabcut algorithm, the Gaussian mixture model is replaced by the probability neural network model PNN, and then the morphological opening operation and morphological dilation algorithm are used to reduce the holes in the image, making the internal and edge information of the image clearer; The specific steps for the improved Grabcut algorithm to separate the foreground and background of the background image are as follows: Step 1.1: Perform preprocessing operations on the background image. Take the internal pixels of the detection area delimited on the transmission line as the foreground, and the external pixels of the detection area delimited on the transmission line as the background. Establish the grayscale histograms of the foreground and background, select the pixels with a higher proportion of pixel values as the input samples for training, and input them into the PNN model; Step 1.2: Use the input layer in the PNN model to receive the input samples; Step 1.

3. Calculate the distance between the input sample and the center using the hidden layer. The input sample x is input into the hidden layer, and the j-th center x of the i-th class in the hidden layer ij The determined input-output relationship is as follows: Among them, σ is the smoothing factor, d is the dimension of the sample space data. Train the network with samples, divide the input samples into i categories. When inputting data, the PNN network identifies whether the image is the background or the foreground. i = 0 represents the background, and 1 represents the foreground; Step 1.4: Output the weighted average of the output values of the foreground or background belonging to the same class in the hidden layer, S i indicating the relationship between the test data and the foreground or background, L i indicating the number of centers of the foreground or background hidden layer; Step 1.

5. The output layer normalizes Step 1.4, and takes the largest output value as the foreground or background probability value of the output, argmax(S i ); Step 1.6: Update the energy function E(·), so as to solve the weights of the t-links using the PNN model, perform image output, and improve the algorithm efficiency; E(α, σ, z) = S(α, σ, z) + V(α, z); Among them, S(·) is the energy value of the cutting t-links; where L n is the number of centers of the n-th hidden layer; Step 1.7: Perform morphological operations on the output image. Use the opening operation and then the dilation algorithm to avoid holes in the output image and make the edges of the output image clear; Step 2: Judge whether there is a target intrusion in the real-time image. If there is a target intrusion in the real-time image, use the edge detection algorithm to extract the contour of the target, and use the image recognition algorithm to identify and process the target to judge the target type; Step 3: Predict the running trend of the intruding target and the detection area delimited on the transmission line, judge whether the target is approaching or staying on the transmission line, and feedback the prediction result to the information feedback system. If the prediction result exceeds the given threshold, it is an abnormal situation, and the information feedback system issues an alarm reminder.

2. The monitoring method for determining whether a target contacts a power transmission line according to claim 1, wherein The background update strategy is specifically as follows: Step 1.8: If the background image is the same as the real-time image, update the real-time image into a new background image. At the same time, store the old background image as a historical image in the memory; Step 1.9: If the background image is different from the real-time image, compare the real-time image with the historical image. If the real-time image is similar to the historical image, update the real-time image into a new background image. At the same time, store the old background image as a historical image in the memory; Step 1.10: If the real-time image is different from both the background image and the historical image, maintain the original background image.

3. The monitoring method for determining whether a target contacts a transmission line according to claim 1, characterized in that In Step 2, the ROI algorithm is used to delimit the region of interest where the target intrudes.

4. The monitoring method for determining whether a target contacts a transmission line according to claim 3, characterized in that, To determine whether there is a target intrusion in the real-time image, specifically: If a target intersects with the region of interest, it is determined that there is a target intrusion in the real-time image. The edge detection algorithm is used to identify the real-time contour of the target, and the contour of the target is framed. When the contour line of the target overlaps with the region of interest, the target is identified by the image recognition algorithm to determine the target type.

5. The monitoring method for judging whether a target contacts a power transmission line according to claim 1, wherein, The edge detection algorithm uses the Canny edge detection algorithm to obtain the contour line of the target.

6. The monitoring method for determining whether an object contacts a power transmission line according to claim 5, characterized in that, The Canny edge detection algorithm determines the optimal threshold T by using the OTSU method, and then calculates the high threshold and low threshold of the Canny edge detection algorithm. Among them, the high threshold is T, and the low threshold is T / 2.

7. The monitoring method for determining whether a target contacts a transmission line according to claim 5, wherein The detection area on the transmission line is delimited by the Hough line transform algorithm. Specifically: First, the straight lines on the transmission line are detected, and the length and width of the detected straight lines are expanded outward by a certain distance to form a detection rectangle frame, and the target is subjected to image recognition operations when the target invades the rectangle frame.

8. The monitoring method for determining whether a target contacts a transmission line according to claim 7, characterized in that, The pre-judgment result is: the minimum straight-line distance between the contour line of the target and the straight line perpendicular to the rectangle frame. If the minimum straight-line distance is less than the threshold, it is determined that the target is approaching the transmission line; if the minimum straight-line distance is 0, it is determined that the target stays on the transmission line.

Images