A method for positioning and identifying hard pressure plates in intelligent substations

Abstract

The present invention provides a method for positioning and identifying hard pressure plates in intelligent substations. First, the collected pressure plate images are preprocessed for distortion correction; then the images are segmented using HSI color feature extraction technology; then the connected domains are analyzed for morphological features, the boundary information of the connected domains is used to remove the identification plates, and the extreme corrosion algorithm is used to separate the adhesion connected domains; finally, the pressure plate position and number are located using an external rectangle, and the pressure plate state is identified based on the fusion of the boundary feature of the simple external rectangle and the deflection angle feature of the area external rectangle. The present invention is more adaptable in complex lighting environments by separating brightness information from color information, and improves the efficiency of pressure plate identification by reducing the workload of pressure plate identification and simplifying the information in the image using area features. The recognition rate of the pressure plate state of the fuzzy image is improved by using an extreme corrosion algorithm, and the recognition rate is improved by fusing two recognition methods.

Description

Technical Field

[0001] The invention relates to the technical field of hard pressure plates in transformer substations, and in particular to a method for positioning and identifying hard pressure plates in intelligent transformer substations. Background Art

[0002] As the level of intelligence of substations increases, many power companies have put into use or are planning to build centralized remote image monitoring systems, adding remote viewing to the four functions of telemetry, telesignaling, remote regulation, and remote control of power systems, and realizing the visualization of various important monitoring information. Moreover, the development and application of key image recognition technologies provide strong technical support for real-time monitoring of status information of substation-related equipment. In addition, the configuration of operation and inspection personnel is difficult to keep up with the development of the scale of the station, and manual inspection of the status of the pressure plate of the protection cabinet is one of the important tasks of the operation personnel. At present, it is time-consuming and labor-intensive to manually obtain the status of the pressure plate, and it is impossible to monitor it in real time, so intelligent identification of the status of the pressure plate has important research significance.

[0003] Traditional remote identification of substation switches is based on image samples of two different states, open and closed, and feature matching is performed based on images taken on site to identify the switch state. If the highest match is with the image sample of the open state, the switch state is determined to be open; if the highest match is with the image sample of the closed state, the switch state is determined to be closed. However, for some normally closed switches, they are generally closed and only open during maintenance. Therefore, there is a lack of open state samples, and only the closed state can be determined. In the case of insufficient light or rainy weather, the pictures taken do not match the sample images very well, which is prone to misjudgment, resulting in unnecessary investment of manpower and material resources.

[0004] At present, there are two main methods for identifying the status of hard pressure plates in smart substations: one is to transform the pressure plate and add sensors and other facilities to convert the pressure plate opening and closing signals into electrical signals to achieve status identification; the other is to obtain the pressure plate status by image recognition. The method of installing and modifying the pressure plate status identification requires the destruction of the original pressure plate equipment, and the maintenance cost of the equipment is high, which is not suitable for engineering applications; the method of using images to identify the pressure plate status also has the problem of not considering the difference in lighting environment, and due to environmental differences, it is also easy to cause the adhesion problem of the connected domain of the extracted pressure plate; the method of using deep learning to identify the pressure plate status has the problems of high requirements for computer hardware, limited learning samples, and difficulty in optimizing model parameters. Summary of the invention

[0005] The purpose of the present invention is to provide a platen positioning and recognition method based on HSI color space segmentation and image processing technology. The platen image on the protection cabinet is processed by Hough transform straight line detection. The intersection coordinates of the straight lines can be obtained by the gap on the outside of the platen. The inclined image is projected onto a new corrected plane by correction processing. An HSI color model is adopted to avoid the influence of illumination difference on the acquired RGB true color platen image. The color feature of the image is extracted to achieve the segmentation of the platen area. The eight-connected seed filling algorithm is adopted to extract the connected domain of the image to obtain all the connected domains in the image. The Extreme Erosion Algorithm (EEA) is adopted to re-judge the larger connected domain existing in the image after preliminary judgment to determine whether it is an adhesion connected domain. The platen connected domain boundary extracted by the simple minimum circumscribed rectangle method is adopted, which is conducive to expressing the overall morphological characteristics of the connected domain through several specific pixel points. The aspect ratio recognition method and the deflection angle method are respectively adopted for the normal platen and the adhesion platen to improve the recognition accuracy of the adhesion platen. The fusion of the two recognition methods improves the recognition rate.

[0006] To achieve the above object, the present invention adopts the following technical solution:

[0007] A method for locating and identifying a hard pressure plate of an intelligent substation comprises the following steps:

[0008] Step 1, using Hough transform line detection and perspective transform to pre-process the collected pressure plate image for distortion correction;

[0009] Step 2: Use HSI color feature extraction technology to extract color features from the image after distortion correction preprocessing to achieve segmentation of the pressure plate area;

[0010] Step 3: Perform morphological feature analysis on the connected domains of the pressure plate area and the signboard area obtained after the pressure plate image is segmented by color features, and use the morphological features of the pressure plate and the signboard to perform binary classification on the ratio of the boundary length to the boundary width of each connected domain to achieve the screening of the connected domains of the pressure plate and the signboard; use the boundary information of the connected domain to remove the signboard, use the extreme erosion algorithm to select the initially determined larger connected domain, and use the circular structure element of the specified radius to continuously erode the connected domain. If the connected domain is cracked during the continuous erosion of the image, the two cracked connected domains are saved, thereby achieving the separation of the sticky connected domains;

[0011] Step 4: Use the circumscribed rectangle method to process the binary grayscale image of the pressure plate obtained in step 3 to locate the position and number of the pressure plate, match the connected domain number with its corresponding pressure plate, and identify the pressure plate state based on the fusion of the simple circumscribed rectangle boundary feature and the deflection angle feature of the area circumscribed rectangle.

[0012] Furthermore, the step 1 specifically includes the following steps:

[0013] Step 1.1: Use Hough transform line detection to obtain the intersection coordinates of the lines

[0014] The horizontal gaps in the detection image are distributed at the top and bottom of all horizontal straight lines. By comparing the sizes of the ordinates of the intersections of all horizontal straight lines and the gap straight lines, the lines with the smallest and largest ordinates are the upper and lower straight lines where the horizontal gaps are located. Finally, the four coordinates P of the intersection of the two horizontal gaps and the two vertical gaps are obtained. 1 (x 1 ,y 1 ), P 2 (x 2 ,y 2 ), P 3 (x 3 ,y 3 ), P 4 (x 4 ,y 4 ), from left to right and from top to bottom;

[0015] Step 1.2: Use image perspective transformation to correct the image and obtain the corrected image

[0016] The basic principle of perspective transformation is to project the tilted image onto a new corrected plane. The projection between planes involves point-to-point mapping. The formula for perspective transformation is as follows:

[0017]

[0018] The calculation formula of the corrected horizontal coordinate is as follows:

[0019]

[0020] The calculation formula of the corrected ordinate is as follows:

[0021]

[0022] Where x, y are the coordinates (x, y) of a pixel in the tilted image, x', y' are the new pixel coordinates (x', y') after the image is corrected; a 11 、a 12 、a 13 、a 21 、a 22 、a 23 、a 31 、a 32There are 8 parameters to be determined; 8 equations need to be constructed to solve the 8 parameters. First, the coordinates of the 4 intersection points can be obtained by line detection. By calculating the geometric distance of these 4 coordinates, the mapping coordinates P after image correction can be approximately estimated. 1 ′、P 2 ′、P 3 ′、P 4 ′, bring the 4 intersection coordinates and the corresponding 4 mapping coordinates into the perspective transformation matrix, get 8 parameter equations, solve the perspective transformation matrix parameters, and finally get the corrected image.

[0023] Furthermore, the step 2 specifically includes:

[0024] Step 2.1: Convert RGB model to HSI model

[0025] In the HSI color model, H represents the hue component, S represents the saturation component, and I represents the brightness component. This color model is used to separate the color information, brightness information, and saturation information of the image. Different light intensities will only affect the size of the I value, but have no effect on the H value and the S value. Therefore, thresholds are set for the H component and the S component in the HSI color model to achieve image color extraction; the mapping relationship from the RGB model to the HSI model is as follows:

[0026]

[0027] Where: R, G, B represent the three components of the RGB color model; H, S, I represent the three components of the HSI;

[0028] Step 2.2: Set the platen color feature threshold to achieve image segmentation

[0029] The protection function pressure plate is marked in yellow, the trip pressure plate is marked in red, and the standby pressure plate is marked in light camel. By extracting the red and yellow components of the image in the HSI color model, a binary image containing only the pressure plate and the identification plate can be obtained. According to the H and S value ranges of red and yellow, the binary classification threshold is obtained. By setting thresholds for the H and S values, the red and yellow pixels in the pressure plate image can be screened out to achieve image segmentation.

[0030] Step 2.3: Platen image color extraction

[0031] Based on the thresholds of H and S values, each pixel in the pressure plate image is screened, and the pixels that meet the conditions are retained and their grayscale values ​​are set to 255, and the pixels that do not meet the conditions are set to 0, and finally a binary grayscale image containing the pressure plate and the sign is obtained;

[0032] Step 2.4: Image filtering

[0033] Morphological filtering is performed on the initially segmented binary grayscale image, and the filtering process includes corrosion processing, expansion processing, opening operation and closing operation.

[0034] Furthermore, the specific steps of step 2.3 are as follows:

[0035] (1) Input the RGB color plate image and obtain the first pixel f 1 The values ​​of the R, G, and B components of (x, y);

[0036] (2) Using the color model conversion formula, obtain the pixel point f 1 (x, y) H and S component values ​​in the HSI color model;

[0037] (3) If the pixel point f 1 If the H and S values ​​of (x, y) satisfy the following formula, then the pixel is red, and its gray value is Gray[f i (x, y)] = 255 Output:

[0038]

[0039] In the formula, f r (x, y) represents a red pixel;

[0040] If the H and S values ​​do not satisfy the above formula, but satisfy the following formula, then the pixel is yellow, and its gray value is Gray[f i (x, y)] = 255 Output:

[0041]

[0042] In the formula, f y (x, y) represents a yellow pixel;

[0043] If neither of the above two equations is satisfied, then it is a pixel point in the non-pressing plate area, and the gray value of the pixel point is Gray[f i (x, y)] = 0 output;

[0044] (4) After the first pixel is determined, the pixel points f are obtained from left to right and from top to bottom. i (x, y) order, repeat steps 1, 2, 3 to determine the pixel point f i (x, y), determine f i The gray value of (x, y) is calculated until all the pixels of the image are determined, and a binary image with a black background including the pressure plate and the identification plate is obtained, thereby realizing the initial segmentation of the pressure plate image.

[0045] Furthermore, the step 3 specifically includes the following steps:

[0046] Step 3.1: Extract image connected areas

[0047] After the pressure plate image is segmented by color features, the pressure plate area and the sign area arranged in sequence are obtained. Each area is defined as a connected domain. The eight-connected seed filling algorithm is used to extract the connected domains of the image. Finally, all the connected domains in the image are obtained and digital labels are attached to all the connected domains in sequence.

[0048] Step 3.2: Signboard removal based on boundary information analysis

[0049] By comparing and analyzing the morphological characteristics of the pressure plate and the signboard, it is found that a single pressure plate presents a vertical rectangle, while the signboard presents a horizontal rectangle. The ratio of the boundary length to the boundary width of each connected domain is classified into two categories to realize the screening of the connected domains of the pressure plate and the signboard. The following formula is the formula for screening the pressure plate and the signboard:

[0050]

[0051] In the formula, x i末 ,y i末 represents the horizontal and vertical coordinates of the last pixel in the i-th connected domain; x i首 ,y i首 Represents the horizontal and vertical coordinates of the first pixel in the i-th connected domain;

[0052] Step 3.3: Initial determination of adhesion connected domains

[0053] The area of ​​the adhesion connected domain is generally twice the area of ​​the normal single platen connected domain. The relationship between the area of ​​each connected domain in the image and the average area of ​​all connected domains is analyzed to preliminarily screen out the adhesion connected domains in the image. The screening formula for adhesion connected domains is as follows:

[0054]

[0055] In the formula, S i is the area of ​​the i-th connected domain; k represents the total number of connected domains in the image after removing the interfering light spots; S (re-max3)j is the area of ​​the jth connected domain after excluding the three largest connected domains;

[0056] Step 3.4: Determine the separation adhesion connected domain

[0057] (1) Obtain the initially determined larger connected domain L and continuously erode L using a circular structure element with a radius of 2;

[0058] (2) If a connected domain is split during the continuous erosion of the image, the two connected domains L that are split are saved. 1 , L 2If no connected domain is cracked during the corrosion process and a seed point is eventually formed, the connected domain L is determined to be a connected domain formed by a single pressing plate;

[0059] (3) For two connected domains L that have undergone cleavage 1 , L 2 , continue to perform extreme corrosion on it. During the corrosion process, the connected domain with a smaller area will be corroded into a seed point g first 1 (x 1 ,y 1 ), at this time, the pixel coordinates of the first seed point are saved and no corrosion is performed on this seed point;

[0060] (4) Continue to corrode the remaining connected domain until the second seed point g is formed after corrosion. 2 (x 2 ,y 2 ) and save the pixel coordinates of this seed point;

[0061] (5) Analyze the pixel coordinate position information of the two seed points. If (x 1 -x 2 ) 2 -(y 1 -y 2 ) 2 >0, indicating that the two seed points are distributed left and right, and the adhesion connected domain is determined to be formed by the adhesion of two adjacent pressure plates, so the two connected domains L generated after cleavage are retained. 1 , L 2 ; If (x 1 -x 2 ) 2 -(y 1 -y 2 ) 2 <0, indicating that the two seed points are distributed up and down, and it is judged that the adhesion connected domain is formed by the adhesion of the pressure plate and the signboard below it;

[0062] (6) For the connected domain determined to be the adhesion between the pressure plate and the signboard, if y 1 >y 2 , determine L 1 For the identification plate, L 2 As the pressure plate, let L 1 The gray value of each pixel in the connected domain is 1, and it is set as the background area; if y 1 <y 2 , determine L 1 is the pressure plate, L2 is the identification plate;

[0063] (7) By re-determining the connected domain information obtained, the adhesion connected domains are separated, the signboard connected domain is eliminated, the pressure plate connected domain is left, and the connected domain is renumbered.

[0064] Furthermore, the step 4 specifically includes the following steps:

[0065] Step 4.1: Positioning the bounding rectangle

[0066] The minimum rectangular frame enclosing the connected domain is formed by two boundary lines parallel to the X-axis and two boundary lines parallel to the Y-axis. The minimum circumscribed rectangle refers to the minimum circumscribed rectangular frame enclosing the connected domain, and this circumscribed rectangular frame reflects the deflection angle of the pressing plate.

[0067] Step 4.2: Identification of normal platen status

[0068] The connected domain that has not been processed by the extreme corrosion algorithm is called a normally segmented pressure plate. A simple minimum circumscribed rectangle is used to locate it. The rectangular frame in the "open" and "closed" states has unique boundary characteristics. When the pressure plate is in the "open" state, the ratio of length to width is close to 1; when it is in the "closed" state, the ratio of length to width is generally less than 0.5.

[0069] The state classification of the i-th connected domain can be realized as follows:

[0070]

[0071] Where x j1 ,y j1 Indicates the horizontal and vertical coordinates of the upper left corner of the circumscribed rectangular box; x j4 ,y j4 Indicates the horizontal and vertical coordinates of the lower right corner of the circumscribed rectangular box;

[0072] Step 4.3: Identification of the sticking platen status

[0073] For the adhesion connected domain, the smallest circumscribed rectangle is selected to judge its state. When the pressure plate is in the "open" state, the deflection angle of the smallest circumscribed rectangle is about 45°; when the pressure plate is in the "closed" state, the deflection angle of the smallest circumscribed rectangle is about 0°.

[0074] The following formula can be used to classify the status of the left and right adhesion pressure plates:

[0075]

[0076] In the formula, x m1 ,y m1 Indicates the horizontal and vertical coordinates of the upper left corner of the smallest circumscribed rectangular box; x m3 ,y m3 Represents the horizontal and vertical coordinates of the lower left corner of the minimum circumscribed rectangular box.

[0077] The present invention has the following beneficial effects:

[0078] 1. The present invention adopts a color extraction technology based on HSI color space to separate brightness information from color information. Compared with the traditional RGB color extraction method, it can avoid the influence of illumination factors on image segmentation, making the segmentation method of the present invention more adaptable to complex illumination environments;

[0079] 2. The present invention adopts color feature separation technology to segment the pressing plate, which can directly extract the effective pressing plate and eliminate the spare pressing plate, greatly reducing the workload of pressing plate identification and improving the efficiency of pressing plate identification;

[0080] 3. The present invention utilizes the area characteristics of the pressing plate and the identification plate, removes the connected domain of the identification plate in the image, simplifies the information in the image, and improves the recognition efficiency;

[0081] 4. The present invention adopts an extreme erosion algorithm, and determines whether it is an adhesion connected domain based on the number of seed points obtained by the extreme erosion algorithm, determines the type of the adhesion connected domain based on the pixel coordinates of the seed points, and completes the segmentation of the adhesion connected domain, thereby realizing the segmentation processing of the fuzzy image and improving the recognition rate of the fuzzy image pressure plate state;

[0082] 5. The present invention adopts different recognition methods for the normal pressing plate and the adhesion pressing plate, respectively, thereby improving the recognition accuracy of the adhesion pressing plate, and integrating the two recognition methods to improve the recognition rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] Figure 1 This is a flow chart of a method for locating and identifying a hard pressure plate in an intelligent substation according to an embodiment of the present invention;

[0084] Figure 2 A Hough transform straight line detection diagram used in the present invention;

[0085] Figure 3 The image distortion correction processing result adopted by the present invention;

[0086] Figure 4 is the color threshold in the HSI space adopted by the present invention;

[0087] Figure 5 It is a flow chart of the color extraction of the pressing plate image adopted by the present invention;

[0088] Figure 6 This is a diagram of the initial segmentation result of the pressing plate used in the present invention;

[0089] Figure 7 This is the result diagram after the image filtering process adopted by the present invention;

[0090] Figure 8 A schematic diagram of the connected region extraction method used in the present invention;

[0091] Fig. 9 This is a result diagram of the signboard removal used in the present invention;

[0092] Fig.10 This is a flow chart of the treatment of the adhesion platen used in the present invention;

[0093] Fig.11 This is a diagram showing the adhesion treatment result of the pressing plate identification plate used in the present invention;

[0094] Fig.12 The minimum bounding rectangle diagram used in the present invention;

[0095] Fig.13 It is a simple minimum circumscribed rectangular frame feature map adopted by the present invention;

[0096] Fig.14 It is the normal platen state recognition result adopted by the present invention;

[0097] Fig.15 It is the minimum area circumscribed rectangular frame feature map used in the present invention;

[0098] Fig.16 The result of identifying the state of the adhesion platen used in the present invention;

[0099] Fig.17 The minimum circumscribed rectangular diagram of the bonding platen used in the present invention;

[0100] Fig.18 is the aspect ratio recognition result adopted by the present invention;

[0101] Fig.19 The deflection angle recognition result of No. 5 and No. 6 pressing plates used in the present invention;

[0102] Fig. 20 A statistical table for identifying the pressing plate samples used in the present invention;

[0103] Fig.21 The following are the comparison results of the recognition rate and recognition efficiency of the three methods used in the present invention. DETAILED DESCRIPTION

[0104] The technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the present invention.

[0105] See also Figure 1 The embodiment of the present invention provides a method for locating and identifying a hard pressure plate of an intelligent substation, comprising the following steps:

[0106] Step 1: performing distortion correction preprocessing on the collected pressure plate image; specifically, step 1 includes the following steps:

[0107] Step 1.1: Use Hough transform line detection to obtain the intersection coordinates of the lines

[0108] like Figure 2 is the Hough transform straight line detection map, Figure 2 It can be seen that the horizontal gaps in the detection image are distributed at the top and bottom of all horizontal straight lines. By comparing the sizes of the ordinates of the intersections of all horizontal straight lines and the gap straight lines, the lines with the smallest and largest ordinates are the upper and lower straight lines where the horizontal gaps are located. Finally, the four coordinates P of the intersection of the two horizontal gaps and the two vertical gaps are obtained. 1 (x 1 ,y 1 ), P 2 (x 2 ,y 2 ), P 3 (x 3 ,y 3 ), P 4 (x 4 ,y 4 ), from left to right and from top to bottom.

[0109] Step 1.2: Use image perspective transformation to correct the image and obtain the corrected image

[0110] The basic principle of perspective transformation is to project the tilted image onto a new corrected plane. The projection between planes involves point-to-point mapping. The formula for perspective transformation is as follows:

[0111]

[0112] The calculation formula of the corrected horizontal coordinate is as follows:

[0113]

[0114] The calculation formula of the corrected ordinate is as follows:

[0115]

[0116] In the formula, x, y are the coordinates (x, y) of a pixel point in the tilted image, and x', y' are the new pixel coordinates (x', y') after the image is corrected. 11 、a 12 、a 13 、a 21 、a 22 、a 23 、a 31 、a 32 There are 8 parameters to be requested.

[0117] Eight parameters need to be solved by constructing eight equations. First, the coordinates of four intersection points can be obtained by line detection. By calculating the geometric distances of these four coordinates, the mapping coordinates P after image correction can be approximately estimated. 1 ′、P 2 ′、P 3 ′、P 4 ′. Substitute the 4 intersection coordinates and the corresponding 4 mapping coordinates into the perspective transformation matrix, and get 8 parametric equations. The perspective transformation matrix parameters can be solved, and finally the corrected image can be obtained. Figure 3 This is the image after correction and cutting out the information outside the straight line.

[0118] Step 2: Use HSI color feature extraction technology to segment the image to achieve segmentation of the pressure plate area; specifically, step 2 includes the following steps:

[0119] Step 2.1: Convert RGB model to HSI model

[0120] Since the three component values ​​of the true color plate image R (red component), G (green component), and B (blue component) are not only affected by color, but also by brightness and saturation, the values ​​of the three components of the same color under different lighting conditions are different. Therefore, in the RGB color model, illumination has always been an important factor affecting the quality of image segmentation. However, in the HSI color model, H represents the hue component, S represents the saturation component, and I represents the brightness component. This color model can separate the color information, brightness information, and saturation information of the image. Different illumination intensities will only affect the size of the I value, but have no effect on the H and S values. Therefore, in the HSI color model, thresholds are set for the H and S components to achieve image color extraction. The mapping relationship from the RGB model to the HSI model is as follows:

[0121]

[0122] Where: R, G, B represent the three components of the RGB color model; H, S, I represent the three components of the HSI;

[0123] Step 2.2: Set the platen color feature threshold to achieve image segmentation

[0124] Color is the most prominent feature of the pressure plate. The State Grid Corporation of China stipulates that the protection function pressure plate should be marked in yellow, the trip pressure plate should be marked in red, and the standby pressure plate should be marked in light camel. Therefore, by extracting the red and yellow components of the image in the HSI color model, a binary image containing only the pressure plate and the sign can be obtained. According to the H value and S value range of red and yellow, the binary classification threshold is obtained as follows Figure 4 By setting thresholds for the H value and the S value, the red pixels and yellow pixels in the platen image can be screened out to achieve image segmentation.

[0125] Step 2.3: Platen image color extraction

[0126] Based on the thresholds of H and S values, each pixel in the pressure plate image is screened. The pixels that meet the conditions are retained and their grayscale values ​​are set to 255, and the pixels that do not meet the conditions are set to 0. Finally, a binary grayscale image containing the pressure plate and the sign is obtained. The process of color extraction of the pressure plate image is as follows: Figure 5 , the specific steps are as follows:

[0127] (1) Input the RGB color plate image and obtain the first pixel f 1 The values ​​of the R, G, and B components of (x, y).

[0128] (2) Using the color model conversion formula, obtain the pixel point f 1 (x, y) are the H and S component values ​​in the HSI color model.

[0129] (3) If the pixel point f 1 If the H and S values ​​of (x, y) satisfy the following formula, then the pixel is red, and its gray value is Gray[f i (x, y)] = 255 Output:

[0130]

[0131] In the formula, f r (x, y) represents a red pixel;

[0132] If the H and S values ​​do not satisfy the above formula, but satisfy the following formula, then the pixel is yellow, and its gray value is Gray[f i (x, y)] = 255 Output:

[0133]

[0134] In the formula, f y (x, y) represents a yellow pixel;

[0135] If neither of the above two equations is satisfied, then it is a pixel point in the non-pressing plate area, and the gray value of the pixel point is Gray[f i (x, y)] = 0 output.

[0136] (4) After the first pixel is determined, the pixel points f are obtained from left to right and from top to bottom. i (x, y) order, repeat steps 1, 2, 3 to determine the pixel point f i (x, y), determine f iThe gray value of (x, y) is calculated until all pixels of the image are determined, and a binary image with a black background including the pressure plate and the sign is obtained. The initial segmentation of the pressure plate image is realized, and the segmentation result is as follows: Figure 6 .

[0137] Step 2.4: Image filtering

[0138] Due to the limited shooting conditions and the accuracy of obtaining the color threshold, the color extracted image contains a small number of interference points. These special pixels will cause small internal gaps and external isolated points in the initially segmented binary image. Therefore, it is necessary to perform morphological filtering on the initially segmented binary grayscale image. The filtering process includes corrosion processing, expansion processing, opening operation and closing operation. Figure 7 for Figure 6 The result after filtering.

[0139] Step 3: Use the eight-connected seed filling algorithm to extract the connected domain, perform morphological feature analysis on the connected domain, use the boundary information of the connected domain to remove the sign, and use the extreme erosion algorithm to separate the adhesion connected domain; specifically, step 3 includes the following steps:

[0140] Step 3.1: Extract image connected areas

[0141] After the pressure plate image is segmented by color features, the pressure plate area and the sign area are arranged in order, and each area is defined as a connected domain. The eight-connected seed filling algorithm is used to extract the connected domain of the image. Finally, all the connected domains in the image are obtained, and digital labels are attached to all the connected domains in order. Figure 8 Extract schematics for connected domains.

[0142] Step 3.2: Signboard removal based on boundary information analysis

[0143] By comparing and analyzing the morphological characteristics of the pressure plate and the signboard, it is found that a single pressure plate presents a vertical rectangle, while the signboard presents a horizontal rectangle. Therefore, the ratio of the boundary length to the boundary width of each connected domain is classified into two categories to realize the screening of the connected domains of the pressure plate and the signboard. The following formula is the formula for screening the pressure plate and the signboard. Fig. 9 This is the result image of signboard culling.

[0144]

[0145] In the formula, x i末 ,y i末 represents the horizontal and vertical coordinates of the last pixel in the i-th connected domain; x i首 ,y i首 Represents the horizontal and vertical coordinates of the first pixel in the i-th connected component.

[0146] Step 3.3: Initial determination of adhesion connected domains

[0147] Due to the complexity of the actual shooting environment, there may be adhesion connected domains with an area significantly larger than the average area in the segmented image, that is, there may be adhesion between the connected domains of the pressure plates and the pressure plates or between the pressure plates and the signboard.

[0148] The area of ​​the adhesion connected domain is generally twice the area of ​​the normal single platen connected domain. Therefore, by analyzing the relationship between the area of ​​each connected domain in the image and the average area of ​​all connected domains, the adhesion connected domains in the image can be preliminarily screened out. The screening formula for adhesion connected domains is as follows.

[0149]

[0150] In the formula, S i is the area of ​​the i-th connected domain; k represents the total number of connected domains in the image after removing the interfering light spots; S (re-max3)j The area of ​​the jth connected domain after excluding the three largest connected domains is to avoid the influence of the maximum area of ​​the adhesion connected domain on the average area; multiplying by 150% takes into account that there are certain area differences in the connected domains of a single pressure plate in a complex environment.

[0151] Step 3.4: Determine the separation adhesion connected domain

[0152] The extreme erosion algorithm was originally used to segment adhesion cell images in the field of biological cytology. The area feature analysis of the connected domain in the platen image can only preliminarily determine the larger connected domain in the image, and then use the extreme erosion algorithm to re-determine the larger connected domain to determine whether it is an adhesion connected domain. The purpose of the initial determination is to reduce the amount of data for re-determination and improve the determination efficiency. Fig.10 This is a flow chart for the treatment of adhesion platen, the specific steps are as follows:

[0153] (1) Obtain the initially determined larger connected domain L and continuously erode L using a circular structure element with a radius of 2.

[0154] (2) If a connected domain is split during the continuous erosion of the image, the two connected domains L that are split are saved. 1 , L 2 If no connected domain is cracked during the corrosion process and a seed point is eventually formed, the connected domain L is determined to be a connected domain formed by a single pressure plate.

[0155] (3) For two connected domains L that have undergone splitting 1 , L 2 , continue to perform extreme corrosion on it. During the corrosion process, the connected domain with a smaller area will be corroded into a seed point g first 1 (x 1 ,y1 ), the pixel coordinates of the first seed point are saved and no further erosion is performed on this seed point.

[0156] (4) Continue to corrode the remaining connected domain until the second seed point g is formed after corrosion. 2 (x 2 ,y 2 ) and save the pixel coordinates of this seed point.

[0157] (5) Analyze the pixel coordinate position information of the two seed points. If (x 1 -x 2 ) 2 -(y 1 -y 2 ) 2 >0, indicating that the two seed points are distributed left and right, and the adhesion connected domain is determined to be formed by the adhesion of two adjacent pressure plates, so the two connected domains L generated after cleavage are retained. 1 , L 2 ; If (x 1 -x 2 ) 2 -(y 1 -y 2 ) 2 <0, indicating that the two seed points are distributed up and down, and it is judged that the adhesion connected domain is formed by the adhesion of the pressure plate and the signboard below it.

[0158] (6) For the connected domain determined to be the adhesion between the pressure plate and the signboard, if y 1 >y 2 , determine L 1 For the identification plate, L 2 As the pressure plate, let L 1 The gray value of each pixel in the connected domain is 1, and it is set as the background area; if y 1 <y 2 , determine L 1 For the pressure plate, L 2 For identification.

[0159] (7) By re-determining the connected domain information obtained, the adhesion connected domains are separated, the signboard connected domain is eliminated, the pressure plate connected domain is left, and the connected domain is renumbered.

[0160] Fig.11 This is the result of the adhesion treatment between the pressure plate and the nameplate.

[0161] Step 4: Use the circumscribed rectangle to locate the position and number of the pressure plate, and identify the pressure plate state based on the fusion of the simple circumscribed rectangle boundary feature and the deflection angle feature of the area circumscribed rectangle. Specifically, step 4 includes the following steps:

[0162] Step 4.1: Positioning the bounding rectangle

[0163] The boundary of the pressure plate connected domain extracted by the seed filling algorithm is relatively rough, which is not conducive to representing the overall morphological characteristics of the connected domain through a few specific pixel points. Therefore, it is necessary to perform minimum circumscribed rectangle processing on the segmented pressure plate image.

[0164] The principle of the simple minimum enclosing rectangle is to use two parallel X-axis and two parallel Y-axis boundary lines to enclose the minimum rectangular box of the connected domain. Fig.12 (b) is a simple minimum bounding rectangle processing diagram; the minimum bounding rectangle refers to the minimum area bounding rectangle frame that surrounds the connected domain. This bounding rectangle frame reflects the deflection angle of the pressure plate. Fig.12 (c) is the processing diagram of the minimum enclosing rectangle.

[0165] Step 4.2: Identification of normal platen status

[0166] The connected domain that has not been processed by the extreme erosion algorithm is called a normal segmented pressure plate, and a simple minimum circumscribed rectangle is used to locate it. Fig.13 The advantage of the simple minimum bounding rectangle is that it is less sensitive to the slight deflection angle of the pressure plate. Analysis of its characteristics shows that the rectangular box in the "open" and "closed" states has unique boundary characteristics. When the pressure plate is in the "open" state, the ratio of length to width is close to 1; when it is in the "closed" state, the ratio of length to width is generally less than 0.5.

[0167] By analyzing the length-to-width ratio range of a large number of rectangular frames surrounding the pressure plate, it is found that when the length-to-width ratio B i When the threshold is set to 0.6, the binary classification result is better. The state classification of the i-th connected domain can be achieved according to the following formula. Fig.14 for Fig.12 Normal platen status recognition result:

[0168]

[0169] Where x j1 ,y j1 Indicates the horizontal and vertical coordinates of the upper left corner of the circumscribed rectangular box; x j4 ,y j4 Indicates the horizontal and vertical coordinates of the lower right corner of the bounding rectangle.

[0170] Step 4.3: Identification of the state of the sticking plate

[0171] Because the connected domain separated from the adhesion domain has a large distortion after the extreme corrosion treatment, there is a high probability of misidentification using only the aspect ratio method of the circumscribed rectangle. The left and right adhesion platens separated by extreme corrosion still contain a part of the horizontally derived area, which makes the boundary length of the simple circumscribed rectangular frame larger, and it is easy to judge the closed state platen as the open state, and then make a wrong judgment; the upper and lower adhesion platens, whose connected domains contain a part of the vertically derived area, make the boundary width of the simple minimum circumscribed rectangular frame larger, and it is easy to judge the "open" state platen as the "closed" state, and then make a wrong judgment. Therefore, for the adhesion connected domain, the minimum area circumscribed rectangular frame with lower sensitivity to the boundary size can be used to judge its state.

[0172] like Fig.15 It is the characteristic diagram of the minimum area circumscribed rectangle. When the pressure plate is in the "open" state, the deflection angle of the minimum area circumscribed rectangle is about 45°; when the pressure plate is in the "closed" state, the deflection angle of the minimum area circumscribed rectangle is about 0°.

[0173] By analyzing a large number of platen deflection angle ranges, when the deflection angle a i When the threshold is set to 10°, the binary classification result is better. The state classification of the left and right adhesion pressure plates can be realized according to the following formula.

[0174]

[0175] In the formula, x m1 ,y m1 Indicates the horizontal and vertical coordinates of the upper left corner of the smallest circumscribed rectangular box; x m3 ,y m3 Represents the horizontal and vertical coordinates of the lower left corner of the minimum circumscribed rectangular box. Fig.16 for Fig.12 The circumscribed rectangle feature statistics and state recognition results of the No. 16 pressure plate.

[0176] Step 5: Using the adhesion platen image, the deflection angle and the aspect ratio recognition method are compared. In order to consider the performance of the platen recognition method of the present invention, the RGB clustering segmentation recognition method is compared with the HSI segmentation recognition method proposed in the present invention; the aspect ratio recognition method is compared with the recognition method proposed in the present invention. Specifically, step 5 includes the following steps:

[0177] Step 5.1: Left and right adhesion platen status identification test

[0178] For the normal segmented pressure plate image, the circumscribed rectangle aspect ratio method has a higher recognition rate, but for the specially processed adhesion connected domain, the boundary information is greatly affected by the derived area and is prone to recognition errors. This paper obtains some adhesion pressure plate images and conducts a comparative experiment between the deflection angle and aspect ratio recognition methods. Fig.17(a) is a simple minimum bounding rectangle diagram of a certain adhesive platen. Fig.17 (b) is the minimum area circumscribed rectangle diagram, in which the No. 5 and No. 6 pressure plates are stuck together. Fig.18 In order to use the aspect ratio method to identify the platen status results, only the first 6 platen identification results are shown. Fig.19 It is the result of identifying the status of the No. 5 and No. 6 pressure plates by the deflection angle.

[0179] contrast Fig.17 and Fig.18 From the recognition results of No. 5 and No. 6 pressure plates, it can be seen that the recognition results of No. 6 pressure plate are the same, but the recognition results of No. 5 pressure plate are different. The pressure plate is manually observed to be in the "open" state, and the deflection angle recognition method is correct. This is because after the left and right adhesion pressure plates are corroded and cracked, there are still a large number of pixels that deviate from the corresponding pressure plate, resulting in a large boundary length of the simple circumscribed rectangle, and then recognition errors; the offset angle of the circumscribed rectangle of the pressure plate area is mainly affected by the overall offset of the pressure plate, and a slight offset of the pixel points has a lower effect on its deflection angle, so the deflection angle recognition method is better for the recognition of poor quality adhesion pressure plates.

[0180] Step 5.2: Recognition rate and recognition efficiency test

[0181] In order to consider the performance of the pressure plate recognition method of the present invention, the existing RGB clustering segmentation recognition method is compared with the HSI segmentation recognition method of the present invention; the aspect ratio recognition method is compared with the recognition method of the present invention. The hardware platform adopted by the present invention is Inter Core i5-12400CPU, and the software platform is Python3.8. 2000 pressure plate image samples were obtained from the State Grid Huzhou Power Supply Company, and all samples were processed and recognized by a protection pressure plate operation status recognition method based on an OTSU threshold segmentation optimization method and an image fusion-based recognition method with the introduction of an improved RANSAC algorithm, a color extraction method + aspect ratio classification recognition method of the present invention, and a color extraction + fusion recognition method of the present invention, and the like. Fig. 20 The following table shows the recognition results of each method.

[0182] Fig. 20 Where Tp represents the number of pressure plates whose real state is also "open" among the pressure plates whose recognition result is "open"; FP represents the number of pressure plates whose real state is "closed" among the pressure plates whose recognition result is "open"; FN represents the number of pressure plates whose real state is "open" among the pressure plates whose recognition result is "closed"; TN represents the number of pressure plates whose real state is also "closed" among the pressure plates whose recognition result is "closed".

[0183] The recognition rate PAC of each method can be calculated as follows, which indicates the proportion of samples that are correctly identified. The recognition rate is an important indicator for judging the performance of different recognition methods. At the same time, the time module function is introduced to obtain the time required for the three methods to identify all the pressure plates, reflecting the recognition efficiency of the three methods. Fig.21 This is a comparison table of recognition rate and recognition efficiency of the three methods.

[0184]

[0185] analyze Fig.21 The recognition rate and recognition efficiency of the three methods show that: in terms of the recognition rate of the pressure plate, the recognition accuracy of the method proposed by the present invention is as high as 98.09%; the recognition accuracy of the protection pressure plate operation status identification method based on image fusion based on the OTSU threshold segmentation optimization method and the introduction of the improved RANSAC algorithm is only 94.7%; based on the HSI color space segmentation method of the pressure plate image in this paper, the accuracy of the aspect ratio recognition method is as high as 98.01%. Comparing the recognition rates of the protection pressure plate operation status identification method based on image fusion and the method of the present invention, it can be seen that the method of the present invention has obvious recognition advantages, mainly because the segmentation method of the present invention is more advantageous, avoiding the influence of illumination factors, and the segmented image effect is better; from the recognition results of the aspect ratio method and the method of the present invention, it can be seen that the fusion recognition method of the present invention also has a slight advantage, and the reason why its advantage is not obvious is mainly because the number of adhesion connected domains in the sample is small, and the impact on the overall recognition rate is limited.

[0186] In terms of recognition efficiency, the method proposed in the present invention takes 77.74 seconds to recognize 2,000 platen images with a total of 37,842 platens, while the protection platen operating status recognition method based on image fusion takes 82.59 seconds, and the aspect ratio method takes 77.67 seconds. The method of the present invention takes less time than the cluster segmentation recognition method, because the pixel color extraction and segmentation method used in the present invention is simpler than the RGB cluster segmentation method, and the present invention eliminates the connected domain of the sign after segmentation, which will greatly reduce the amount of data for subsequent image processing and improve efficiency; the aspect ratio method and the method proposed in the present invention are basically the same in recognition efficiency, because the two only have slight differences in the platen status judgment method, which has little effect on the recognition efficiency.

[0187] In the process of segmenting the acquired RGB true color platen image, the illumination difference has a great influence on it. The color extraction technology based on the HSI color space in the embodiment of the present invention can separate the brightness information from the color information. Compared with the traditional RGB color extraction method, it can avoid the influence of illumination factors on image segmentation and has stronger adaptability to complex lighting environments.

[0188] Image segmentation is also a key step in image processing. The embodiment of the present invention uses color feature separation technology to segment the pressure plate, which can directly extract the effective pressure plate and eliminate the spare pressure plate, greatly reducing the workload of pressure plate identification and improving the efficiency of pressure plate identification.

[0189] Image feature extraction and matching play a vital role in the accuracy of image recognition. The embodiment of the present invention utilizes the area features of the pressure plate and the identification plate to remove the connected domain of the identification plate in the image, thereby simplifying the information in the image and improving the recognition efficiency.

[0190] The problem of adhesion connected domains is prone to occur when segmenting fuzzy images. The embodiment of the present invention can segment fuzzy images using the extreme erosion algorithm. The number of seed points obtained by the extreme erosion algorithm is used to determine whether it is an adhesion connected domain, and the type of adhesion connected domain is determined based on the pixel coordinates of the seed points. The adhesion connected domain is segmented, and the segmentation of the adhesion connected domain is completed, thereby achieving the segmentation of the fuzzy image and improving the recognition rate of the fuzzy image pressure plate state. In order to improve the recognition accuracy of the adhesion pressure plate, different recognition methods are used for the normal pressure plate and the adhesion pressure plate, and the fusion of the two recognition methods can improve the recognition rate.

[0191] From the above analysis results, it can be seen that the HSI color segmentation method used in the present invention has more advantages than the conventional RGB color segmentation method in extracting the connected domain of the pressure plate, and it takes less time and is more efficient. Compared with the single aspect ratio method, the aspect ratio and deflection angle fusion recognition method used in the present invention has a higher recognition rate when processing special adhesion pressure plates.

[0192] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for locating and identifying a hard pressure plate in an intelligent substation, characterized in that: The steps include: Step 1, using Hough transform line detection and perspective transform to pre-process the collected pressure plate image for distortion correction; Step 2: Use HSI color feature extraction technology to extract color features from the image after distortion correction preprocessing to achieve segmentation of the pressure plate area; Step 3: Perform morphological feature analysis on the connected domains of the pressure plate area and the signboard area obtained after the pressure plate image is segmented by color features, and use the morphological features of the pressure plate and the signboard to perform binary classification on the ratio of the boundary length to the boundary width of each connected domain to achieve the screening of the connected domains of the pressure plate and the signboard; use the boundary information of the connected domain to remove the signboard, use the extreme erosion algorithm to select the initially determined larger connected domain, and use the circular structure element of the specified radius to continuously erode the connected domain. If the connected domain is cracked during the continuous erosion of the image, the two cracked connected domains are saved, thereby achieving the separation of the sticky connected domains; Step 4, using the circumscribed rectangle method to process the binary grayscale image of the pressure plate obtained in step 3 to locate the position and number of the pressure plate, matching the connected domain number with its corresponding pressure plate, and identifying the pressure plate state based on the fusion of the simple circumscribed rectangle boundary feature and the deflection angle feature of the area circumscribed rectangle; The step 3 specifically includes the following steps: Step 3.1: Extract image connected areas After the pressure plate image is segmented by color features, the pressure plate area and the sign area arranged in sequence are obtained. Each area is defined as a connected domain. The eight-connected seed filling algorithm is used to extract the connected domains of the image. Finally, all the connected domains in the image are obtained and digital labels are attached to all the connected domains in sequence. Step 3.2: Signboard removal based on boundary information analysis By comparing and analyzing the morphological characteristics of the pressure plate and the signboard, it is found that a single pressure plate presents a vertical rectangle, while the signboard presents a horizontal rectangle. The ratio of the boundary length to the boundary width of each connected domain is classified into two categories to realize the screening of the connected domains of the pressure plate and the signboard. The following formula is the formula for screening the pressure plate and the signboard: In the formula, x i末 ,y i末 represents the horizontal and vertical coordinates of the last pixel in the i-th connected domain; x i首 ,y i首 Represents the horizontal and vertical coordinates of the first pixel in the i-th connected domain; Step 3.3: Initial determination of adhesion connected domains The area of ​​the adhesion connected domain is generally twice the area of ​​the normal single platen connected domain. The relationship between the area of ​​each connected domain in the image and the average area of ​​all connected domains is analyzed to preliminarily screen out the adhesion connected domains in the image. The screening formula for adhesion connected domains is as follows: In the formula, S i is the area of ​​the i-th connected domain; k represents the total number of connected domains in the image after removing the interfering light spots; S (re-max3)j is the area of ​​the jth connected domain after excluding the three largest connected domains; Step 3.4: Determine the separation adhesion connected domain (1) Obtain the initially determined larger connected domain L and continuously erode L using a circular structure element with a radius of 2; (2) If the connected domain is split during the continuous erosion of the image, the two connected domains L1 and L2 are saved; if the connected domain is not split during the erosion process and a seed point is finally formed, the connected domain L is determined to be a connected domain formed by a single pressing plate; (3) For the two connected domains L1 and L2 that have been split, continue to perform extreme corrosion on them. During the corrosion process, the connected domain with a smaller area will be corroded into a seed point g1(x1, y1) first. At this time, the pixel coordinates of the first seed point are saved and the seed point is no longer corroded; (4) Continue to erode the remaining connected domain until a second seed point g2(x2, y2) is formed after corrosion, and save the pixel coordinates of this seed point; (5) Analyze the pixel coordinate position information of the two seed points. If (x1-x2) 2 -(y1-y2) 2 >0, indicating that the two seed points are distributed left and right, and the adhesion connected domain is determined to be formed by the adhesion of two adjacent pressure plates, so the two connected domains L1 and L2 generated after cleavage are retained; if (x1-x2) 2 -(y1-y2) 2 <0, indicating that the two seed points are distributed up and down, and it is judged that the adhesion connected domain is formed by the adhesion of the pressure plate and the signboard below it; (6) For the connected domain where the pressure plate and the sign are adhered, if y1>y2, L1 is determined to be the sign and L2 is the pressure plate. At this time, the gray value of each pixel point in the connected domain where L1 is located is set to 1, and it is set to the background area; if y1<y2, L1 is determined to be the pressure plate and L2 is the sign; (7) By re-determining the connected domain information obtained, the adhesion connected domains are separated, the signboard connected domain is eliminated, the pressure plate connected domain is left, and the connected domain is renumbered.

2. The method for locating and identifying a hard pressure plate of an intelligent substation according to claim 1, characterized in that: The step 2 specifically includes: the step 1 specifically includes the following steps: Step 1.1: Use Hough transform line detection to obtain the intersection coordinates of the lines The horizontal gaps in the detection image are distributed at the top and bottom of all the horizontal lines. By comparing the sizes of the ordinates of the intersections of all the horizontal lines and the gap lines, the lines with the smallest and largest ordinates are the upper and lower lines where the horizontal gaps are located. Finally, the four coordinates of the intersections of the two horizontal gaps and the two vertical gaps are obtained, namely P1 (x1, y1), P2 (x2, y2), P3 (x3, y3), and P4 (x4, y4), which are arranged from left to right and from top to bottom. Step 1.2: Use image perspective transformation to correct the image and obtain the corrected image The basic principle of perspective transformation is to project the tilted image onto a new corrected plane. The projection between planes involves point-to-point mapping. The formula for perspective transformation is as follows: The calculation formula of the corrected horizontal coordinate is as follows: The calculation formula of the corrected ordinate is as follows: Where x, y are the coordinates (x, y) of a pixel in the tilted image, x', y' are the new pixel coordinates (x', y') after the image is corrected; a 11 、a 12 、a 13 、a 21 、a 22 、a 23 、a 31 、a 32 、a 33 There are 9 parameters to be determined; 8 equations need to be constructed to solve the 9 parameters. First, the coordinates of the 4 intersection points can be obtained by straight line detection. By calculating the geometric distances of these 4 coordinates, the mapping coordinates P1′, P2′, P3′, and P4′ after image correction can be approximately estimated. The 4 intersection coordinates and the corresponding 4 mapping coordinates are substituted into the perspective transformation matrix to obtain 8 parameter equations. The perspective transformation matrix parameters are solved, and finally the corrected image is obtained.

3. The method for locating and identifying a hard pressure plate of an intelligent substation according to claim 1, characterized in that: The step 2 specifically includes: Step 2.1: Convert RGB model to HSI model In the HSI color model, H represents the hue component, S represents the saturation component, and I represents the brightness component. This color model is used to separate the color information, brightness information, and saturation information of the image. Different light intensities will only affect the size of the I value, but have no effect on the H value and the S value. Therefore, thresholds are set for the H component and the S component in the HSI color model to achieve image color extraction; the mapping relationship from the RGB model to the HSI model is as follows: Where: R, G, B represent the three components of the RGB color model; H, S, I represent the three components of the HSI; Step 2.2: Set the platen color feature threshold to achieve image segmentation The protection function pressure plate is marked in yellow, the trip pressure plate is marked in red, and the standby pressure plate is marked in light camel. By extracting the red and yellow components of the image in the HSI color model, a binary image containing only the pressure plate and the identification plate can be obtained. According to the H and S value ranges of red and yellow, the binary classification threshold is obtained. By setting thresholds for the H and S values, the red and yellow pixels in the pressure plate image can be screened out to achieve image segmentation. Step 2.3: Platen image color extraction Based on the thresholds of H and S values, each pixel in the pressure plate image is screened, and the pixels that meet the conditions are retained and their grayscale values ​​are set to 255, and the pixels that do not meet the conditions are set to 0, and finally a binary grayscale image containing the pressure plate and the sign is obtained; Step 2.4: Image filtering Morphological filtering is performed on the initially segmented binary grayscale image, and the filtering process includes corrosion processing, expansion processing, opening operation and closing operation.

4. The method for locating and identifying a hard pressure plate of an intelligent substation according to claim 3, characterized in that: Step 2.3 The specific steps are as follows: (1) Input the RGB color plate image and obtain the values ​​of the R, G, and B components of the first pixel point f1(x, y); (2) Using the color model conversion formula, obtain the H component and S component values ​​of the pixel point f1(x, y) in the HSI color model; (3) If the H and S values ​​of the pixel point f1(x, y) satisfy the following formula, then the pixel point is red, and its gray value is Gray[f i (x, y)] = 255 Output: In the formula, f r (x, y) represents a red pixel; If the H and S values ​​do not satisfy the above formula, but satisfy the following formula, the pixel is yellow, and its gray value is Gray[f i (x, y)] = 255 Output: In the formula, f y (x, y) represents a yellow pixel; If neither of the above two equations is satisfied, then it is a pixel point in the non-pressing plate area, and the gray value of the pixel point is Gray[f i (x, y)] = 0 output; (4) After the first pixel is determined, the pixel points f are obtained from left to right and from top to bottom. i (x, y) order, repeat steps 1, 2, 3 to determine the pixel point f i (x, y), determine f i The gray value of (x, y) is calculated until all the pixels of the image are determined, and a binary image with a black background including the pressure plate and the identification plate is obtained, thereby realizing the initial segmentation of the pressure plate image.

5. The method for locating and identifying a hard pressure plate of an intelligent substation according to claim 1, characterized in that: The step 4 specifically includes the following steps: Step 4.1: Positioning the bounding rectangle The minimum rectangular frame enclosing the connected domain is formed by two boundary lines parallel to the X-axis and two boundary lines parallel to the Y-axis. The minimum circumscribed rectangle refers to the minimum circumscribed rectangular frame enclosing the connected domain, and this circumscribed rectangular frame reflects the deflection angle of the pressing plate. Step 4.2: Identification of normal platen status The connected domain that has not been processed by the extreme corrosion algorithm is called a normally segmented pressure plate. A simple minimum circumscribed rectangle is used to locate it. The rectangular frame in the "open" and "closed" states has unique boundary characteristics. When the pressure plate is in the "open" state, the ratio of length to width is close to 1; when it is in the "closed" state, the ratio of length to width is generally less than 0.

5. The state classification of the i-th connected domain can be realized as follows: Where x j1 ,y j1 Indicates the horizontal and vertical coordinates of the upper left corner of the circumscribed rectangular box; x j4 ,y j4 Indicates the horizontal and vertical coordinates of the lower right corner of the circumscribed rectangular box; Step 4.3: Identification of the sticking platen status For the adhesion connected domain, the smallest circumscribed rectangle is used to determine its state. When the pressure plate is in the "open" state, the deflection angle of the smallest circumscribed rectangle is about 45°; when the pressure plate is in the "closed" state, the deflection angle of the smallest circumscribed rectangle is about 0°. The following formula can be used to classify the status of the left and right adhesion pressure plates: In the formula, x m1 ,y m1 Indicates the horizontal and vertical coordinates of the upper left corner of the smallest circumscribed rectangular box; x m3 ,y m3 Represents the horizontal and vertical coordinates of the lower left corner of the minimum circumscribed rectangular box.

Images