CN-121982358-A - State checking and analyzing method for relay protection pressing plate

CN121982358ACN 121982358 ACN121982358 ACN 121982358ACN-121982358-A

Abstract

The invention discloses a state checking and analyzing method of a relay protection pressing plate, which comprises the following analysis steps of S1, collecting relay protection pressing plate images by adopting image collecting equipment and processing original pressing plate images, S2, constructing a pressing plate state recognition algorithm based on a YOLOv model to obtain a boundary frame for calibrating pressing plate states and pressing plate double-name label coordinates, S3, correcting an inclined text image calibrated by the boundary frame based on perspective transformation, S4, constructing a relay protection pressing plate corpus and recognizing pressing plate double names based on corrected text images, S5, comparing the pressing plate states obtained by recognition with a pressing plate state library which is required to be input based on the recognized pressing plate states and pressing plate double names, and realizing pressing plate state checking.

Inventors

DU LINKUN
PENG SHUAI
BI LIANG
XU PENGYUE
ZHANG HENGYU
ZHOU LILI
LI TIANFENG
HU ZHIWEI
JIANG NAN

Assignees

国网河南省电力公司许昌供电公司

Dates

Publication Date: 20260505
Application Date: 20251208

Claims (9)

1. A state checking and analyzing method of a relay protection pressing plate is characterized by comprising the following analyzing steps: s1, acquiring relay protection pressing plate images by adopting image acquisition equipment, and processing original pressing plate images; S2, constructing a pressing plate state recognition algorithm based on a YOLOv model to obtain a bounding box for calibrating the pressing plate state and the pressing plate double name label coordinates; S3, correcting the inclination angle of the inclination text image marked by the boundary frame based on perspective transformation; S4, constructing a relay protection pressing plate corpus, and identifying double names of pressing plates based on corrected text images; s5, comparing the pressing plate state and the pressing plate double name obtained based on the identification with a pressing plate state library to be put into, and checking the pressing plate state.
2. The method for checking and analyzing the state of the relay protection pressing plate according to claim 1, wherein in the step S1, a CCD industrial camera is specifically adopted for the original pressing plate image.
3. The method for checking and analyzing the state of the relay protection pressing plate is characterized in that in the step S1, the original pressing plate image is processed, the method comprises graying, stretching, binarization and filtering, the graying is conducted on the original pressing plate image by means of weighted average, the stretching is conducted on a graying image after the graying, a sigmoid function is used as a stretching transformation function to map a minimum gray value and a maximum gray value in the graying image to an ideal interval, the binarization is conducted on the graying image by means of a law method to divide a background and a foreground in the graying image, and the filtering is conducted on the image by means of adaptive median filtering to smooth the image based on the processed graying image.
4. The method for checking and analyzing the state of the relay protection pressing plate according to claim 1 is characterized in that in the step S2, a pressing plate state recognition algorithm is built based on a YOLOv model, a YOLOv model structure comprises an input end, a backbone network, a neck network layer and a detection head, wherein the input end enhances a rich data set through Mosaic data, a K-means self-adaptive anchor frame algorithm is adopted to calculate and obtain the optimal anchor frame size and self-adaptive scaling picture size, the backbone network comprises a C3 module and an SPPF module and is responsible for extracting feature pictures from pictures of the input end, the neck network comprises a feature pyramid network FPN and a multi-path information fusion network PAN and is used for fusing feature information extracted by the backbone network, and the detection head analyzes feature pictures of different scales output by the neck network and trains or predicts through a loss function and NMS non-maximum suppression.
5. The method for checking and analyzing the state of the relay protection pressing plate according to claim 1, wherein the step S3 is characterized in that the inclined text image is subjected to inclination angle correction, and specifically comprises the steps of determining edges and straight lines based on a Canny edge detection algorithm and a Hough transformation straight line detection algorithm respectively, calculating four coordinate points according to a straight line set, obtaining parameters of a perspective transformation matrix, and performing perspective transformation after the perspective transformation matrix is generated to obtain a horizontal text image.
6. The method for checking and analyzing the state of the relay protection pressing plate according to claim 5, wherein the determining the edge based on the Canny edge detection algorithm in the step S3 specifically includes the steps of ⑴ calculating the image by using a finite difference Sobel operator of a3×3 neighborhood first order bias Gradient amplitude of (2) And gradient direction , wherein, , , , , 、 Convolution templates in the X and Y directions respectively, , Performing non-maximum suppression on ⑵, traversing the image, and performing gradient amplitude value on the pixel points Corresponding to the gradient direction in the 3 multiplied by 3 neighborhood The two elements most closely related are compared, if not greater in magnitude than they are I.e. the pixel is not an edge, otherwise let ⑶ Setting high and low thresholds of gradient amplitude, ① calculating image gradient amplitude 128-Level histogram of (2) And find non-zero starting points for histograms And non-zero endpoint ② Computing the slave in the histogram To the point of Is used to determine the slope of the (c) for the (c), ③ Find the slope maximum If the point of (a) is used as a bump of the histogram Will be As a new starting point And returning to the value step ② to continue searching for new bumps, otherwise, turning to the next step ④ to connect all found bumps with non-zero starting points and non-zero ending points to obtain the minimum convex polygon enveloping the histogram And calculates a notch residual of the histogram The high and low thresholds are respectively: , And ⑷, determining edge points, wherein the pixel points with the gradient amplitude value larger than the high threshold value are edge points, namely the gray value is 255, and the pixel points with the gradient amplitude value smaller than the low threshold value are non-edge pixel points, namely the gray value is 0, and the quasi-edge pixel points between the two can be marked as the edge points only when the determined edge points exist in the 3 multiplied by 3 neighborhood.
7. The method for checking and analyzing the state of the relay protection pressing plate according to claim 5, wherein the determining the straight line based on the Hough transform straight line detection algorithm in the step S3 specifically comprises the following steps of, for any point in the image space, the functional relationship is: Wherein: Representing the distance from the origin to the straight line, And carrying out the transformation on n points positioned on the same straight line L, correspondingly obtaining n sine curves in a parameter space by n points of the original image space, intersecting the sine curves at the same point, voting the points, setting a vote threshold value, and judging the detected straight line exceeding the threshold value to obtain a Hough transformation result.
8. The method for checking and analyzing the state of the relay protection pressing plate according to claim 1, wherein in the step S4, the double names of the pressing plate are identified by adopting optical character recognition, and particularly, character recognition is carried out by adopting TESSERACT optical character recognition engines.
9. The method for checking and analyzing the state of the relay protection pressing plate according to claim 1, wherein the construction process of the relay protection pressing plate corpus in the step S4 specifically comprises the following steps of ① collecting text images with double names of the relay protection pressing plate and generating character frames, ② judging whether the character frames are correct or not, if not, manually correcting the character frames until the character frames are correct, and if so, performing the next step, ③ generating character frame files, ④ generating character configuration files, training files, character set files and clustering character files, ⑤ merging all files, and ⑥ pressing plate double-name professional corpus.

Description

State checking and analyzing method for relay protection pressing plate Technical Field The invention belongs to the technical field of relay protection, and particularly relates to a state checking and analyzing method of a relay protection pressing plate. Background Along with intelligent upgrading of the power system, the number of relay protection equipment of the transformer substation is increased, the corresponding pressing plate verification workload is increased, the number of pressing plates in a large transformer substation can reach hundreds, in the prior art, the state verification of the relay protection pressing plates mainly depends on manual detection of operation and maintenance personnel, 2-3 hours are required for single comprehensive verification, the efficiency is low and the relay protection pressing plates are easily influenced by fatigue of the personnel, and according to statistics, the false judgment rate of manual verification is up to 15%, and the failure or false operation of a protection device can be caused. Therefore, in order to solve the above-mentioned problems, it is necessary to develop a status checking and analyzing method for the relay protection pressing plate. Disclosure of Invention The invention aims to overcome the defects of the prior art, and provides a state checking and analyzing method for a relay protection pressing plate, which improves the checking efficiency of the relay protection pressing plate and reduces the risk of state misjudgment. The invention aims to realize the state checking and analyzing method of the relay protection pressing plate, which comprises the following analyzing steps: s1, acquiring relay protection pressing plate images by adopting image acquisition equipment, and processing original pressing plate images; S2, constructing a pressing plate state recognition algorithm based on a YOLOv model to obtain a bounding box for calibrating the pressing plate state and the pressing plate double name label coordinates; S3, correcting the inclination angle of the inclination text image marked by the boundary frame based on perspective transformation; S4, constructing a relay protection pressing plate corpus, and identifying double names of pressing plates based on corrected text images; s5, comparing the pressing plate state and the pressing plate double name obtained based on the identification with a pressing plate state library to be put into, and checking the pressing plate state. Further, in the step S1, a CCD industrial camera is specifically used for the original platen image. The method comprises the steps of S1, namely, carrying out processing on an original pressing plate image, namely, carrying out gray processing on the original pressing plate image by a weighted average method, stretching the original pressing plate image, namely, carrying out stretching processing on the gray image subjected to gray processing by a sigmoid function serving as a stretching transformation function, namely, mapping a minimum gray value and a maximum gray value in the gray image to an ideal interval, carrying out binarization processing on the gray image by a discriminant method in the binarization processing, namely, dividing a background from a foreground in the gray image, and carrying out smoothing processing on the image based on the processed gray image by a self-adaptive median filtering. Further, in the step S2, in the step of constructing a platen state recognition algorithm based on the YOLOv model, the YOLOv model structure comprises an input end, a backbone network, a neck network layer and a detection head, wherein the input end is used for enhancing a rich data set through Mosaic data, calculating and obtaining the optimal anchor frame size by adopting a Kmeans self-adaptive anchor frame algorithm and self-adaptively scaling the picture size, the backbone network comprises a C3 module and an SPPF module and is responsible for extracting a characteristic picture from the picture of the input end, the neck network comprises a characteristic pyramid network FPN and a multi-path information fusion network PAN and is used for fusing characteristic information extracted by the backbone network, and the detection head is used for analyzing the characteristic pictures of different scales output by the neck network and training or predicting through a loss function and NMS non-maximum suppression. Further, in the step S3, the inclination angle correction is carried out on the inclined text image, and the method specifically comprises the following steps of determining edges and straight lines based on a Canny edge detection algorithm and a Hough transformation straight line detection algorithm respectively, calculating four coordinate points according to a straight line set, obtaining parameters of a perspective transformation matrix, and carrying out perspective transformation after the perspective transformation matrix is generated, so as to