A substation equipment inspection image automatic naming method based on intelligent identification and rapid matching

Abstract

The application is suitable for the substation inspection technical field, and discloses a substation equipment inspection image automatic naming method based on intelligent identification and rapid matching, and the process is as follows: receiving a substation equipment image training set, using an image recognition algorithm to establish image feature point description information of the preprocessed image, and using machine learning to train the preprocessed substation equipment image training set; receiving a substation equipment image to be identified, preprocessing the image, extracting feature points, performing image search based on a SIFT algorithm, and identifying the substation equipment category; using a maximum cross-correlation matching method to perform image similarity analysis, calculating the position deviation of the image recognition target frame and the template picture, obtaining the power equipment category, and naming.

Description

Technical Field

[0001] This invention belongs to the technical field of power systems, and particularly relates to an automatic naming method for substation equipment inspection images based on intelligent recognition and rapid matching. Background Technology

[0002] In recent years, some beneficial explorations and gratifying achievements have been made in the application of image processing and recognition technology in power systems. However, at present, it still relies on on-duty personnel to directly observe and analyze the collected images to judge the operating status of power equipment. There is a lack of automatic identification and analysis functions for power equipment in substations. This situation lags behind other industries. The reason for this is that the research on the analysis of complex substation images and the methods for judging the operating faults of power equipment is still immature.

[0003] CN115062178A discloses a method, apparatus, computer equipment, and storage medium for processing images from unmanned aerial vehicle (UAV) equipment inspections. The method includes: acquiring equipment inspection images collected during UAV inspections of power transmission equipment and determining the UAV's inspection route; then acquiring the image capture points of the equipment inspection images; matching these image capture points with each inspection capture point along the inspection route to determine the target inspection capture point corresponding to the equipment inspection image; identifying the power transmission equipment located at the target inspection capture point; and naming the equipment inspection image according to the identifier associated with the power transmission equipment, resulting in an image name including the identifier associated with the power transmission equipment. By matching the equipment inspection image capture points with each inspection capture point to determine the power transmission equipment located at the target inspection capture point, and automatically naming the equipment inspection image using the identifier associated with the power transmission equipment, the efficiency of UAV inspection image processing is improved. However, there is still room for improvement. Summary of the Invention

[0004] To address the problems existing in the prior art, this invention proposes an automatic naming method for substation equipment inspection images based on intelligent recognition and rapid matching. The specific steps of this method are as follows:

[0005] Step 1: Receive the training set of images of substation equipment, use image recognition algorithms to establish image feature point description information for the preprocessed images, and use machine learning to train the training set of images of substation equipment.

[0006] Step 2: Receive the image of the substation equipment to be identified, preprocess it, extract feature points, and perform image search based on the SIFT algorithm for feature point detection;

[0007] Step 3: Perform image similarity analysis using the maximum cross-correlation matching method, calculate the positional deviation between the image within the target bounding box and the template image, and name the device in the current image when the obtained deviation is the smallest.

[0008] Secondly, in step one, the specific steps of the preprocessing in this method include:

[0009] Upon receiving the valid location identification command, mark the valid locations that can be used for identification in the image of the power equipment;

[0010] Furthermore, in this method, the images of the power equipment with marked valid positions are corrected and skewed.

[0011] Furthermore, in this method, image feature point description information is established using an image recognition algorithm on the preprocessed image, and the device is named.

[0012] Furthermore, in step two, the image of the substation equipment to be identified is received, preprocessed, feature points are extracted, and image search is performed based on the SIFT algorithm for feature point detection.

[0013] Furthermore, in step three, image similarity analysis is performed using the maximum cross-correlation matching method to calculate the positional deviation between the image within the target bounding box and the template image. When the obtained deviation is minimized, the device in the current image is named. The specific steps are as follows:

[0014] make It is a picture with dimensions of The image (denoted as W). It is a picture with dimensions of Given an image (denoted as M), use correlation matching to find the parts in W that are similar to M. W represents A. The sub-block of W with the same size as M at its top-left corner also represents the matrix corresponding to that sub-block, i.e.:

[0015] ,

[0016] ,

[0017] In the formula, represent The correlation coefficient with M, where E is the variance of M; for Covariance with M.

[0018] ,

[0019] ,

[0020] ,

[0021] In the formula, M and M represent images respectively. The gray mean of M, yes The grayscale value of the pixel in the i-th row and j-th column. It is the grayscale value of the pixel in the i-th row and j-th column of M.

[0022] If σ(x, y) is large or close to 1, it indicates that image M matches image W at point (x, y). For a given image W, it is necessary to scan from the top left corner to the bottom right corner, recording a total of (Aa) × (Bb) σ(x, y). The point (x, y) corresponding to the maximum value is the point that matches image W. The final result is the deviation value between the two images in the horizontal and vertical directions, respectively. The method is as follows: Take two images, W is the current image, M is a small part of the center of the template image, and S... x，y Let S be any region in W that is exactly the same size as M. Scan image M using the method described above to determine the region S corresponding to the largest σ(x,y). x，y The center coordinates of the image are subtracted from the center coordinates of the image W to obtain the deviation (c) between the current image and the template image. x c y When the deviation is minimized, the device in the template image is identified, and the device in the current image is named according to the device name in the template image.

[0023] Compared with the prior art, the beneficial effects of the present invention include:

[0024] This invention provides an automatic naming method for substation equipment inspection images based on intelligent recognition and fast matching. It has high target recognition accuracy and fast matching speed in substation scenarios, making it a practical method for automatic naming of substation equipment inspection images. Attached Figure Description

[0025] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0026] Figure 1 This is a flowchart of an automatic naming method for substation equipment inspection images based on intelligent recognition and rapid matching.

[0027] Figure 2 This is a flowchart illustrating a method for automatically naming substation equipment inspection images based on intelligent recognition and rapid matching.

[0028] Figure 3 Images of transformer equipment before automatic naming.

[0029] Figure 4 Images of power equipment after automatic naming. Detailed Implementation

[0030] Step 1: Receive the training set of images of substation equipment, use image recognition algorithms to establish image feature point description information for the preprocessed images, and use machine learning to train the training set of images of substation equipment.

[0031] Due to inherent defects in image acquisition equipment and environmental factors, images of power equipment input into a computer inevitably contain distortion and noise. This can severely interfere with subsequent image processing, feature extraction, and recognition analysis, and affect the accuracy of the processing results.

[0032] Image preprocessing: Marking valid locations in the image: Receive a valid location identification instruction and mark the valid locations in the power equipment image; by inputting a large number of real object photographs, developers use computers to pre-mark valid locations in the image (such as nameplates, equipment panels, wiring layouts, etc.). Further, in this method, the power equipment image with marked valid locations undergoes correction and skew-correction processing.

[0033] Feature point analysis: Substation images contain important information about the operation of power equipment. The key to recognizing power equipment images in substations lies in analyzing the various features of the power equipment. After preprocessing, image features that can distinguish the categories of power equipment are selected as input vectors for power equipment recognition. Pattern recognition methods are applied to classify and recognize substation power equipment, and the results of the recognition experiment are summarized and analyzed.

[0034] Feature point extraction: A feature point is a local extremum with directional information detected in images at different scales. A feature point has three characteristics: scale, orientation, and size.

[0035] Images of a normal size can be calculated. ~ The system has a number of eigenvalues, each of which is a 128-dimensional floating-point vector. It uses the spatial information of the vectors to construct a kd-tree, i.e., a k-dimensional vector tree, where k = 128.

[0036] A. Generating the Difference-of-Gaussian Pyramid (DOG Pyramid), scale-space construction

[0037] By scaling the original image, a multi-scale spatial representation sequence of the image is obtained. The main contour of the scale space is extracted from these sequences, and the main contour is used as a feature vector to achieve edge and corner detection and key point extraction at different resolutions.

[0038] B. Spatial extreme point detection (preliminary exploration of key points).

[0039] C. Precise positioning of key stability points.

[0040] D. Stable key point orientation information allocation.

[0041] E. Key point description.

[0042] Describing key points is a crucial step in subsequent matching; the description is essentially a mathematical process of defining key points. A descriptor includes not only the key point itself, but also the neighboring points that contribute to it.

[0043] Image search: Image search involves calculating the Euclidean distance between vectors and finding the sample vector with the smallest distance. The smaller the Euclidean distance, the higher the similarity. When the Euclidean distance is less than a set threshold, a match can be considered successful.

[0044] To eliminate unmatched keypoints caused by image occlusion and background clutter, a SIFT matching method comparing nearest neighbor and second nearest neighbor distances is used: A SIFT keypoint in one image is selected, and its two closest Euclidean distances in another image are found. Between these two keypoints, if the ratio (the distance between the nearest and second nearest neighbors) is less than a certain threshold T, the pair of matching points is accepted. This is because for false matches, due to the high dimensionality of the feature space, similar distances may result in numerous other false matches, leading to a high ratio value. Clearly, lowering this ratio threshold T reduces the number of SIFT matching points but makes the system more stable, and vice versa. The experimentally determined ratio values ​​are shown in the table below:

[0045]

[0046] The approach described is to divide the pixel region around the key point into blocks, calculate the gradient histogram within the block, and generate a unique vector that is an abstract representation of the image information of that region.

[0047] The image recognition algorithm of OpenCV is used to correct and rectify the marked areas in these photos, and to establish a database of image feature point description information and device naming information.

[0048] Step 2: Receive the image of the substation equipment to be identified, preprocess it, extract feature points, and perform image search based on the SIFT algorithm for feature point detection;

[0049] 1. Extract feature points from the image to be recognized;

[0050] 2. Match the image to be recognized with the database. If a match is successful, return the ID of the corresponding image.

[0051] 3. Using the obtained ID, the associated power equipment can be found.

[0052] Image training and matching:

[0053] The training set images need to be consistent with the actual scene environment to be recognized;

[0054] Considering the various possibilities in real-world application scenarios, the images for each label need to cover all possible scenarios in the real-world scenario, such as changes in shooting angle and lighting. The more scenarios the training set covers, the stronger the model's generalization ability will be.

[0055] Of course, during real-time recognition, the object is not exactly the same as the one in the initial training. There may be differences in viewpoint, model, color, size, etc. The algorithm will sort the results according to the magnitude of these differences, placing the smallest differences first and the larger differences later. This is the sorting of the results by confidence level. Finally, a confidence threshold is determined based on the actual recognition performance. Recognitions above this threshold are considered valid and usable. Furthermore, in this method, during image search and matching of the image of the power equipment to be recognized, the confidence level is calculated, sorted, and a confidence threshold is set. Recognitions with a confidence level greater than the threshold are considered valid.

[0056] Step 3: Perform image similarity analysis using the maximum cross-correlation matching method, calculate the positional deviation between the image within the target bounding box and the template image, and name the device in the current image when the obtained deviation is the smallest.

[0057] To automatically name unnamed devices, this paper uses the maximum cross-correlation matching method for image similarity analysis to calculate whether the device in the current image is the same as the device in the template image. The maximum cross-correlation matching method model is as follows:

[0058] make It is a picture with dimensions of The image (denoted as W). It is a picture with dimensions of Given an image (denoted as M), use correlation matching to find the parts in W that are similar to M. W represents A. The sub-block of W with the same size as M at its top-left corner also represents the matrix corresponding to that sub-block, i.e.:

[0059] ,

[0060] ,

[0061] In the formula, represent The correlation coefficient with M, where E is the variance of M; for Covariance with M.

[0062] ,

[0063] ,

[0064] ,

[0065] In the formula, M and M represent images respectively. The gray mean of M, yes The grayscale value of the pixel in the i-th row and j-th column. It is the grayscale value of the pixel in the i-th row and j-th column of M.

[0066] If σ(x, y) is large or close to 1, it indicates that image M matches image W at point (x, y). For a given image W, it is necessary to scan from the top left corner to the bottom right corner, recording a total of (Aa) × (Bb) σ(x, y). The point (x, y) corresponding to the maximum value is the point that matches image W. The final result is the deviation value between the two images in the horizontal and vertical directions. The method is as follows: In the two images, W is the current image, M is a small part of the center of the template image, and S... x,y Let S be any region in W that is exactly the same size as M. Scan image M using the method described above to determine the region S corresponding to the largest σ(x, y). x,y The center coordinates of the image are subtracted from the center coordinates of the image W to obtain the deviation (c) between the current image and the template image. x c y When the deviation is minimized, the device in the template image is identified, and the device in the current image is named according to the device name in the template image.

[0067] This invention provides an automatic naming method for substation equipment inspection images based on intelligent recognition and fast matching. It has high target recognition accuracy and fast matching speed in substation scenarios, making it a practical method for automatic naming of substation equipment inspection images.

[0068] The above embodiments are merely preferred technical solutions of the present invention and should not be considered as limitations on the present invention. The scope of protection of the present invention should be limited to the technical solutions described in the claims, including equivalent substitutions of the technical features described in the claims. That is, equivalent substitutions and improvements within this scope are also within the scope of protection of the present invention.

[0069] Example: Please refer to Figure 1-2 An automatic naming method for substation equipment inspection images based on intelligent recognition and rapid matching includes the following specific steps:

[0070] 1. Receive the training set of images of substation equipment, establish image feature point description information using image recognition algorithms on the preprocessed images, and train the preprocessed training set of substation equipment images using machine learning.

[0071] 2. Receive the image of the substation equipment to be identified, preprocess it, extract feature points, perform image search based on the SIFT feature point detection algorithm, and identify the substation equipment.

[0072] Third, image similarity analysis is performed using the maximum cross-correlation matching method. The positional deviation between the image within the target bounding box and the template image is calculated. When the obtained deviation is the smallest, the device in the current image is named.

[0073] See Figure 3-4 This paper presents an automatic naming method for substation equipment inspection images based on intelligent recognition and rapid matching. In practical application, substation equipment images are unnamed before operation and automatically named after operation. It demonstrates high accuracy in target recognition and fast matching speed in substation scenarios, making it a suitable method for automatic naming of substation equipment inspection images in real-world engineering projects.

[0074] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A substation equipment inspection image automatic naming method based on intelligent identification and rapid matching, characterized in that, The specific steps are: Step one, receiving a power equipment image training set, preprocessing the image, using image recognition algorithm to establish image feature point description information, using machine learning to train the preprocessed power equipment image training set; Step two, receiving a power equipment image to be identified, preprocessing the image, extracting feature points, and detecting SIFT algorithm based on feature points to search the image and identify the power equipment; Step three, using the maximum cross correlation matching method to analyze the image similarity, calculating the position deviation of the image recognition target frame picture and the template picture, and naming the equipment in the current picture when the obtained deviation is the smallest; In step one, the specific steps of preprocessing include: receiving a power equipment image training set, preprocessing the power equipment image, using image recognition algorithm to establish image feature point description information and naming the equipment after preprocessing the image; In step three, the maximum cross correlation matching method is used to analyze the image similarity, and the position deviation of the image recognition target frame picture and the template picture is calculated, and the equipment in the current picture is named when the obtained deviation is the smallest; In order to realize the automatic naming of unnamed equipment, the maximum cross correlation matching method is used to analyze the image similarity to calculate whether the equipment in the current picture and the equipment in the template picture are the same equipment, and the maximum cross correlation matching method model is as follows: Let be a picture of size denoted by W, be a picture of size denoted by M, find the similar part of M in W by correlation matching, let represent the sub-block of W with the same size as M with the top-left corner at and also represent the matrix corresponding to the sub-block, i.e. ， ， wherein represents the correlation coefficient with M, E is the variance of M; is the covariance with M, ， ， wherein and denote the image and the mean gray value of M, is the gray value of the pixel in the i-th row and j-th column of M, is the gray value of the pixel in the i-th row and j-th column of M. If is very large or close to 1, it indicates that the image M matches the image W at point, for a given image W, it is necessary to scan from the top left corner to the bottom right corner, a total of , the maximum value of which corresponds to point that matches the image W, the final result is the deviation of the two images in the horizontal and vertical directions, the method is as follows: W is the current image, M is the small part of the center of the template image, is any one of the same size as M in W, according to the above method, the image M is scanned to determine the maximum corresponding to center coordinates, the center coordinates are subtracted from the center coordinates of the image W, and the deviation of the current image and the template image is obtained , when the obtained deviation is the smallest, it is the device in the template image, and the device in the current image is named according to the device name in the template image.​ 2. The method according to claim 1, characterized in that, In step two, receiving a power equipment image to be identified, preprocessing the power equipment image to be identified, extracting feature points in the preprocessed power equipment image, and detecting SIFT algorithm based on feature points to search the image, which includes: in the image search matching of the power equipment image to be identified, calculating the confidence, sorting the confidence, setting the confidence threshold, and the calculated confidence greater than the confidence threshold is effective recognition.

Images