CN111797925B - Visual image classification method and device for power system - Google Patents
Visual image classification method and device for power system Download PDFInfo
- Publication number
- CN111797925B CN111797925B CN202010631654.XA CN202010631654A CN111797925B CN 111797925 B CN111797925 B CN 111797925B CN 202010631654 A CN202010631654 A CN 202010631654A CN 111797925 B CN111797925 B CN 111797925B
- Authority
- CN
- China
- Prior art keywords
- points
- image
- point
- network model
- power system
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 40
- 230000000007 visual effect Effects 0.000 title claims abstract description 23
- 238000012549 training Methods 0.000 claims abstract description 21
- 230000004044 response Effects 0.000 claims abstract description 14
- 238000004364 calculation method Methods 0.000 claims abstract description 7
- 238000001514 detection method Methods 0.000 claims abstract description 6
- 239000011159 matrix material Substances 0.000 claims description 9
- 230000001629 suppression Effects 0.000 claims description 9
- 238000000605 extraction Methods 0.000 claims description 6
- 238000005316 response function Methods 0.000 claims description 6
- 230000011218 segmentation Effects 0.000 claims description 6
- 239000013598 vector Substances 0.000 claims description 6
- 238000013527 convolutional neural network Methods 0.000 claims description 5
- 238000005457 optimization Methods 0.000 claims description 3
- 238000004519 manufacturing process Methods 0.000 abstract description 2
- 238000002360 preparation method Methods 0.000 abstract description 2
- 230000008569 process Effects 0.000 description 5
- 238000012360 testing method Methods 0.000 description 5
- 238000004422 calculation algorithm Methods 0.000 description 4
- 238000006243 chemical reaction Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000009826 distribution Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 238000004590 computer program Methods 0.000 description 3
- 238000003745 diagnosis Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 238000012795 verification Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 229910018503 SF6 Inorganic materials 0.000 description 1
- 230000005856 abnormality Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000006866 deterioration Effects 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 239000007789 gas Substances 0.000 description 1
- 230000005764 inhibitory process Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 238000004393 prognosis Methods 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
- SFZCNBIFKDRMGX-UHFFFAOYSA-N sulfur hexafluoride Chemical compound FS(F)(F)(F)(F)F SFZCNBIFKDRMGX-UHFFFAOYSA-N 0.000 description 1
- 229960000909 sulfur hexafluoride Drugs 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000011426 transformation method Methods 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a visual image classification method and a visual image classification device for an electric power system, which are characterized in that an original data set is established by collecting images of various equipment of the electric power system, and the original data set is classified; establishing an image classification network model, and training and optimizing the image classification network model by using the classified original data set to obtain an optimized image classification network model; the visual images of the power system are classified by utilizing the optimized image classification network model, the visual image recognition and classification of the power system are carried out by utilizing the image classification network model trained by various equipment images of the power system, the comprehensive judgment of relevant pictures of the power system can be rapidly realized according to the classification result, the classification result is accurate, the detection efficiency is greatly improved, and the response speed and the safety production of the power system are improved; the ORB method is adopted to extract and match the characteristics of the targets in the images, so that early preparation is made for realizing classification, the method is relatively stable, the robustness is high, the calculation speed is high, and the method has good rotation invariance and noise resistance.
Description
Technical Field
The invention belongs to the field of image processing, and particularly relates to a visual image classification method and device for an electric power system.
Background
In recent years, in order to improve the computing power of a computer, things can be recognized, so that judgment and deep learning can be performed, computer vision technology methods and applications are rapidly developed, and the global computer vision market is rapidly developing. At present, due to the continuous improvement of national life and economical strength, the demand of society for electric energy is huge, the traditional operation mode is more difficult to fully meet the requirements of safe and stable operation of a power system, and in the background, the automation of the power system becomes a new development trend, and the multimedia technology occupies an important position. The technology has the irreplaceable advantages of other technologies in the aspects of information input, output, transmission and the like, and mainly processes based on image information. Therefore, in order to gradually increase the degree of automation of the power system, development of image processing techniques is highly necessary.
In power systems, conditions such as transformer discharge faults, line faults, leakage of insulating gases (e.g., sulfur hexafluoride) and the like often occur. Currently, there is an increasing demand for electricity, and the power system is more and more complex, and the occurrence of various faults is a not insignificant challenge for monitoring personnel. The fault diagnosis system of the power system can give decision references of accident conditions, so that the application of the fault diagnosis system of the power system can reduce the pressure of monitoring personnel to a great extent, and simultaneously, the fault diagnosis system of the power system has higher speed and the capability of identifying and positioning faults more accurately.
In the conventional manner, in order to acquire information contained in an image, it is mainly through observation of human eyes and understanding of brain. However, human judgment is subjective as compared with a computer, and at the same time, long-time viewing of video or images is also prone to fatigue. If an anomaly occurs in the power system, these negative effects may cause a decrease in the accuracy of the prognosis and processing. In addition, since the power system is a system that changes in real time, it is necessary to quickly perform comprehensive judgment on the relevant screen, and thus, a high response speed is required for the power system with respect to the change of energy and information. Since the transient process of the power system is short, the operations such as opening and reclosing of the circuit are almost instantaneously completed, and if an abnormality occurs, the circuit must be cut off in a short time to prevent further deterioration of the situation.
Disclosure of Invention
The invention aims to provide a visual image classification method and device for an electric power system, which are used for overcoming the defects in the prior art.
In order to achieve the above purpose, the invention adopts the following technical scheme:
A visual image classification method of a power system comprises the following steps:
step 1), collecting images of various devices of a power system, establishing an original data set, and classifying the original data set;
Step 2), establishing an image classification network model, and training and optimizing the image classification network model by using the classified original data set to obtain an optimized image classification network model;
And 3) classifying the visual images of the power system by using the optimized image classification network model.
Further, the image classification network model adopts a convolutional neural network.
Further, training and optimizing the image classification network model, specifically, adjusting the image size of the image by using the classified original data set, performing format conversion, then graying the image after format conversion, extracting feature points of the image after graying by using an ORB method, and setting matcher parameters by using the extracted features to complete the training and optimizing of the image classification network model.
Further, the BGR format image is specifically converted into an RGB format image.
Further, in a circular neighborhood taking a pixel point of the image after graying as a center, detecting a gray value of the pixel point on the circumference, setting a gray difference value between the pixel point in a radius set around the neighborhood around the center point and the center point to be larger than a set threshold, taking the center point as a candidate feature point, calculating a direction of each candidate feature point by using a gray centroid method, calculating the direction of the candidate feature point by using a vector from the candidate feature point to the neighborhood centroid, and performing rotation operation on BRIEF descriptors according to the direction of the candidate feature point and a corresponding rotation matrix to finish feature extraction.
Further, the radius of the neighborhood around the central point is 3, threshold segmentation is performed on 16 points on the circumference radius, and if the difference between the gray values of the continuous N pixel points on the circumference radius and the central point is larger than the threshold value, the central point is the candidate feature point.
Further, during detection, a lower threshold is set firstly to obtain candidate points with the number exceeding N, then response values of the candidate points are calculated according to the Harris corner response function and are ordered, and the first N points with large response values are taken as final candidate feature points.
Further, score calculation and non-maximum suppression are carried out on the extracted characteristic points: calculating the score of the feature point by adopting a dichotomy, and reserving the feature point with the score of 0-255; and comparing the scores of the characteristic points with the scores of 8 adjacent points around, and reserving the points as maximum value points in a 3X 3 window taking the characteristic points as the center if the score of the characteristic points is maximum, otherwise, discarding the points.
Further, non-maximum suppression is performed on the feature points, the feature points are taken as the center, the kernel size is 5, the side length of the square window is (2×5) +1=11, if and only if the score of the feature points is the maximum value in 121 points of the 11×11 window, the feature points are reserved, otherwise, the feature points are not reserved.
Compared with the prior art, the invention has the following beneficial technical effects:
The invention relates to a visual image classification method of an electric power system, which is characterized in that an original data set is established by collecting various equipment images of the electric power system, and the original data set is classified; establishing an image classification network model, and training and optimizing the image classification network model by using the classified original data set to obtain an optimized image classification network model; the visual images of the power system are classified by utilizing the optimized image classification network model, the visual image recognition and classification of the power system are carried out by utilizing the image classification network model trained by various equipment images of the power system, the comprehensive judgment of the relevant pictures of the power system can be rapidly realized according to the classification result, the classification result is accurate, the detection efficiency can be greatly improved, the response speed of the power system is improved, and the safety production is improved.
Furthermore, the ORB method is adopted to extract and match the characteristics of the targets in the images, so that early preparation is made for realizing classification, the method is relatively stable, the robustness is strong, the calculation speed is high, and the method has good rotation invariance and noise resistance.
Furthermore, through two times of non-maximum value inhibition, feature points which do not meet the requirements are further filtered, and accuracy is improved.
Drawings
Fig. 1 is a schematic diagram of a system structure according to an embodiment of the present invention.
FIG. 2 is a schematic diagram of training optimization in an embodiment of the present invention.
Fig. 3 is a schematic diagram of feature point extraction in an embodiment of the invention.
FIG. 4 is a schematic diagram of homography matrix computation in an embodiment of the present invention.
Detailed Description
The invention is described in further detail below with reference to the attached drawing figures:
As shown in fig. 1, a visual image classification method of a power system includes the following steps:
step 1), collecting images of various devices of a power system, establishing an original data set, and classifying the original data set;
specifically, various equipment images of the power system are collected, and the electrical equipment images and the non-electrical images are marked manually according to the category of the images by different classification methods; then dividing the image into a training set, a testing set and a verification set according to a certain proportion, wherein the specific segmentation proportion of the training set, the testing set and the verification set is 6:2:2. by manual labeling and segmentation, an original dataset is generated. Non-power image samples were derived from the ImageNet image library, and power image samples were collected by professionals.
Step 2), establishing an image classification network model, and training and optimizing the image classification network model by using the classified original data set to obtain an optimized image classification network model;
And 3) classifying the visual images of the power system by using the optimized image classification network model.
Specifically, the image classification network model adopts a convolutional neural network, and can process pixel areas of small blocks on the picture, so that the neural network can see the picture, the capability of a computer for identifying the image, understanding and learning is further improved, the method has remarkable advantages, and the whole process from feature extraction to classification can be completed by training the convolutional neural network, so that various classification tasks are realized. The study herein is based on improvements in pre-trained CNN models to achieve the final goal.
A visual image classification device of an electric power system comprises a data classification module, a network model optimization module and a classifier module;
The data classification module is used for classifying the collected original data sets of various equipment images of the power system; the network model optimizing module trains and optimizes the image classifying network model by using the classified original data set to obtain an optimized image classifying network model, and the classifier module classifies visual images of the power system by using the optimized image classifying network model.
The application also comprises a terminal device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, the processor implementing the steps of the above method when executing the computer program. And comprises a computer readable storage medium storing a computer program which, when executed by a processor, implements the steps of the above method.
As shown in fig. 2, specifically, training and optimizing the image classification network model, specifically, adjusting the image size of the image by using the classified original data set, performing format conversion, specifically, converting the BGR format into the RGB format, then graying the image after the format conversion, extracting feature points of the image after the graying by using the ORB method, and setting matcher parameters by using the extracted features to complete the training and optimizing of the image classification network model.
Specifically, as shown in fig. 3, feature points are extracted by using a FAST method: detecting the gray value of a pixel point on the circumference in a circular adjacent area taking one pixel point of the image after graying as the center, and considering the center point as a candidate feature point if the gray difference value between the pixel point in a preset radius of the adjacent area around the center point and the center point is larger than a preset threshold value; taking 3 of the radius of the neighborhood around the central point, and comparing 16 peripheral pixel points on the circumference; after the neighborhood is determined, carrying out threshold segmentation on 16 points on the circumference, and if the difference between gray values of continuous N pixel points and the central point on the circumference is larger than a threshold value, considering the central point as a candidate feature point; FAST-9 of the present application, i.e., n=9;
Calculating a response value R for each candidate feature point according to the corner response function; the pixel points are divided into three types according to the size of the response value: r is positive and the corner point is the corner point when the value is large; r is negative and the absolute value is large, the edge is the edge; the absolute value of R is small and is a flat area. In order to obtain N characteristic points, during detection, a lower threshold is firstly set to obtain candidate points with the number exceeding N, then response values of the candidate points are calculated according to a Harris angular point response function and are ordered, and the first N points with large response values are taken as final candidate characteristic points; in the neighborhood of each characteristic point, 256 pixel point pairs are selected according to Gaussian distribution, and the gray value of two points in each pixel point pair is compared; for the ith point pair, if the gray value of the first point is smaller than that of the second point, a value of 1 is allocated to the corresponding bit in the descriptor, otherwise, a value of 0 is allocated, and the value of the ith binary bit in the feature vector can be obtained. This procedure is repeated 256 times for the same feature point, and then goes to the next feature point;
And calculating the direction of each candidate feature point by using a gray level centroid method, calculating the direction of the feature point by using a vector from the feature point to the neighborhood centroid of the candidate feature point, and performing rotation operation on the BRIEF descriptor according to the direction of the feature point and the corresponding rotation matrix to finish feature extraction.
According to FAST-9, feature points in the image are extracted, score calculation and non-maximum suppression are carried out on the extracted feature points, and whether the two steps are carried out or not can be selected when a function is called. The score of the characteristic point is the maximum threshold value which enables the point to meet the condition of the characteristic point, and the value range of the pixel point of the gray level image is 0-255, so that the score of the characteristic point is also within the range, and the score is calculated by adopting a dichotomy; the non-maximum suppression is to compare the scores of the feature points with the scores of 8 adjacent points around, namely, in a3×3 window taking the feature points as the center, if the score of the feature points is the maximum, the feature points are reserved, otherwise, the feature points are discarded. On the basis of the FAST algorithm, non-maximum suppression is performed again, the feature points are taken as the center, the kernel size is 5, the side length of the square window is (2×5) +1=11, and if and only if the score of the feature points is the maximum value in 121 points of the 11×11 window, the feature points which do not meet the requirements are reserved, and therefore the feature points which do not meet the requirements are further filtered. Solving corner response values of all candidate feature points in the image according to the Harris corner response function; determining a direction for each feature point according to a gray centroid method; the space distribution table required by generating 256-bit binary feature vectors according to Gaussian distribution, namely the offset of 256 pairs of point sets in the neighborhood relative to the transverse coordinates and the longitudinal coordinates of the feature points is a 256×4 matrix, the data are derived from OpenCV, and the rotation BRIEF descriptor is generated by calculating the distribution of the 256 pairs of point sets; constructing image pyramids with different scales; dividing an original data set into a training image and a test image, matching characteristic points in the training image and the test image, and using K nearest neighbor search to obtain search results of nearest neighbors and next neighbors of each characteristic point according to a Hamming distance, wherein the final matching result must simultaneously satisfy the following three conditions: the hamming distance between the matching point pairs is not greater than 0.25; the ratio of hamming distances of nearest neighbor and next neighbor results is not greater than 0.98; the cross minimum check is satisfied. If a certain feature point A in the training image is B as a search result of the nearest neighbor in the test image, for B, the result C of the nearest neighbor search in the training image is equal to A, otherwise, the verification is not met;
As shown in fig. 4, a RANSAC algorithm is adopted to calculate a homography matrix, and the collineation condition of four points is judged; since the four randomly selected pairs of points in the RANSAC algorithm must be linearly uncorrelated, i.e., any three points (with four combinations) cannot be collinear, the function uses a method that calculates the area of a triangle formed by three points, which are collinear if the area is 0, or otherwise not collinear.
① Solving a homography matrix by a direct linear transformation method;
② Homography transformation is carried out on the coordinates, the reprojection error between the matched point pairs is calculated, and the points with the error smaller than the threshold value are added into the inner point set;
③ The iteration number of the RANSAC algorithm is dynamically updated.
As shown in FIG. 1, the image classification network model established by the invention is a VGG16 network model, the VGG16 network model comprises 5 convolution groups, each convolution group comprises 2, 3 and 3 convolution layers, 13 convolution layers are connected after each convolution group, 5 convolution layers are connected, and 3 full connection layers are arranged in the model. All the convolution layers adopt the same filter parameters, and simultaneously, in the convolution calculation process, the step length and the filling mode of the matrix are reasonably set so that the final result has the same dimension as the input. The depth multiplication, the height width halving of the image in the latter convolutional layer, compared to the former, is due to the effect of the pooling layer between blocks, all of which also employ the same filter parameters.
Claims (4)
1. The visual image classification method of the electric power system is characterized by comprising the following steps of:
step 1), collecting images of various devices of a power system, establishing an original data set, and classifying the original data set;
Step 2), establishing an image classification network model, and training and optimizing the image classification network model by using the classified original data set to obtain an optimized image classification network model;
step 3), classifying and identifying visual images of the power system by using the optimized image classification network model;
training and optimizing an image classification network model, namely adjusting the image size of the image by using a classified original data set, converting the format of the image, graying the image after converting the format, extracting characteristic points of the image after graying by using an ORB method, and setting matcher parameters by using the extracted characteristics to finish the training and optimizing of the image classification network model;
Detecting the gray value of a pixel point on the circumference in a circular neighborhood taking one pixel point of the image after graying as the center, wherein the gray difference value between the pixel point and the center point in a neighborhood set radius around the center point is larger than a set threshold, then the center point is a candidate feature point, calculating the direction of each candidate feature point by using a gray centroid method, calculating the direction of the candidate feature point by using a vector from the candidate feature point to the neighborhood centroid of the candidate feature point, and performing rotation operation on BRIEF descriptors according to the direction of the candidate feature point and a corresponding rotation matrix to finish feature extraction;
taking 3 of the radius of the neighborhood around the central point, carrying out threshold segmentation on 16 points on the circumference radius, and if the difference between the gray values of the continuous N pixel points on the circumference radius and the central point is larger than the threshold, taking the central point as a candidate feature point;
During detection, a lower threshold is firstly set to obtain candidate points with the number exceeding N, then response values of the candidate points are calculated according to a Harris angular point response function and are ordered, and the first N points with large response values are taken as final candidate feature points;
and (3) performing score calculation and non-maximum suppression on the extracted characteristic points: calculating the score of the feature point by adopting a dichotomy, and reserving the feature point with the score of 0-255; comparing the scores of the feature points with the scores of 8 adjacent points around, and reserving the feature points as maximum value points in a 3X 3 window taking the feature points as the center if the score of the feature points is maximum, otherwise, discarding the feature points;
and performing non-maximum suppression on the characteristic points, wherein the characteristic points are taken as the center, the kernel size is 5, the side length of the square window is (2×5) +1=11, and the square window is reserved if and only if the score of the characteristic points is the maximum value in 121 points of the 11×11 window, otherwise, the square window is not reserved.
2. The method for classifying visual images of a power system according to claim 1, wherein the image classification network model employs a convolutional neural network.
3. The method for classifying visual images of a power system according to claim 1, wherein the BGR format image is specifically converted into an RGB format image.
4. The visual image classification device of the electric power system is characterized by comprising a data classification module, a network model optimization module and a classifier module;
The data classification module is used for classifying the collected original data sets of various equipment images of the power system; the network model optimizing module trains and optimizes the image classifying network model by using the classified original data set to obtain an optimized image classifying network model, and the classifier module classifies visual images of the power system by using the optimized image classifying network model;
training and optimizing an image classification network model, namely adjusting the image size of the image by using a classified original data set, converting the format of the image, graying the image after converting the format, extracting characteristic points of the image after graying by using an ORB method, and setting matcher parameters by using the extracted characteristics to finish the training and optimizing of the image classification network model;
Detecting the gray value of a pixel point on the circumference in a circular neighborhood taking one pixel point of the image after graying as the center, wherein the gray difference value between the pixel point and the center point in a neighborhood set radius around the center point is larger than a set threshold, then the center point is a candidate feature point, calculating the direction of each candidate feature point by using a gray centroid method, calculating the direction of the candidate feature point by using a vector from the candidate feature point to the neighborhood centroid of the candidate feature point, and performing rotation operation on BRIEF descriptors according to the direction of the candidate feature point and a corresponding rotation matrix to finish feature extraction;
taking 3 of the radius of the neighborhood around the central point, carrying out threshold segmentation on 16 points on the circumference radius, and if the difference between the gray values of the continuous N pixel points on the circumference radius and the central point is larger than the threshold, taking the central point as a candidate feature point;
During detection, a lower threshold is firstly set to obtain candidate points with the number exceeding N, then response values of the candidate points are calculated according to a Harris angular point response function and are ordered, and the first N points with large response values are taken as final candidate feature points;
and (3) performing score calculation and non-maximum suppression on the extracted characteristic points: calculating the score of the feature point by adopting a dichotomy, and reserving the feature point with the score of 0-255; comparing the scores of the feature points with the scores of 8 adjacent points around, and reserving the feature points as maximum value points in a 3X 3 window taking the feature points as the center if the score of the feature points is maximum, otherwise, discarding the feature points;
and performing non-maximum suppression on the characteristic points, wherein the characteristic points are taken as the center, the kernel size is 5, the side length of the square window is (2×5) +1=11, and the square window is reserved if and only if the score of the characteristic points is the maximum value in 121 points of the 11×11 window, otherwise, the square window is not reserved.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010631654.XA CN111797925B (en) | 2020-07-03 | 2020-07-03 | Visual image classification method and device for power system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010631654.XA CN111797925B (en) | 2020-07-03 | 2020-07-03 | Visual image classification method and device for power system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111797925A CN111797925A (en) | 2020-10-20 |
| CN111797925B true CN111797925B (en) | 2024-04-30 |
Family
ID=72810903
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010631654.XA Active CN111797925B (en) | 2020-07-03 | 2020-07-03 | Visual image classification method and device for power system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111797925B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107067415A (en) * | 2017-03-21 | 2017-08-18 | 南京航空航天大学 | A kind of quick accurate positioning method of target based on images match |
| CN110097051A (en) * | 2019-04-04 | 2019-08-06 | 平安科技（深圳）有限公司 | Image classification method, device and computer readable storage medium |
| CN110738673A (en) * | 2019-10-21 | 2020-01-31 | 哈尔滨理工大学 | Visual SLAM method based on example segmentation |
| WO2020107687A1 (en) * | 2018-11-27 | 2020-06-04 | 邦鼓思电子科技（上海）有限公司 | Vision-based working area boundary detection system and method, and machine equipment |
-
2020
- 2020-07-03 CN CN202010631654.XA patent/CN111797925B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107067415A (en) * | 2017-03-21 | 2017-08-18 | 南京航空航天大学 | A kind of quick accurate positioning method of target based on images match |
| WO2020107687A1 (en) * | 2018-11-27 | 2020-06-04 | 邦鼓思电子科技（上海）有限公司 | Vision-based working area boundary detection system and method, and machine equipment |
| CN110097051A (en) * | 2019-04-04 | 2019-08-06 | 平安科技（深圳）有限公司 | Image classification method, device and computer readable storage medium |
| CN110738673A (en) * | 2019-10-21 | 2020-01-31 | 哈尔滨理工大学 | Visual SLAM method based on example segmentation |
Non-Patent Citations (2)
| Title |
|---|
| 基于双目图像多特征点融合匹配物体识别与定位研究;王霖郁;蒋强卫;李爽;;无线电工程(第08期);全文 * |
| 基于改进ORB的抗视角变换快速图像匹配算法;左川;庞春江;;传感技术学报(第11期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111797925A (en) | 2020-10-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109118479B (en) | Capsule network-based insulator defect identification and positioning device and method | |
| CN111428748B (en) | HOG feature and SVM-based infrared image insulator identification detection method | |
| CN113205063A (en) | Visual identification and positioning method for defects of power transmission conductor | |
| Ni et al. | An improved Faster R-CNN for defect recognition of key components of transmission line | |
| CN112750125B (en) | Glass insulator piece positioning method based on end-to-end key point detection | |
| Liao et al. | Fusion of infrared-visible images in UE-IoT for fault point detection based on GAN | |
| CN112884715A (en) | Composite insulator grading ring inclination fault detection method based on deep learning | |
| CN116612098A (en) | Insulator RTV spraying quality evaluation method and device based on image processing | |
| WO2022222036A1 (en) | Method and apparatus for determining parking space | |
| CN111797925B (en) | Visual image classification method and device for power system | |
| CN112613564A (en) | Target detection post-processing method for eliminating overlapped frames | |
| Zhang et al. | Defect detection of anti-vibration hammer based on improved faster R-CNN | |
| Fu et al. | Bearing surface defect detection based on improved convolutional neural network | |
| CN115661757A (en) | Automatic detection method for pantograph arcing | |
| Bi et al. | Segmentation-based decision networks for steel surface defect detection | |
| CN114596487A (en) | Switch on-off state identification method based on self-attention mechanism | |
| Heng et al. | Anti-vibration hammer detection in UAV image | |
| CN111402223B (en) | Transformer substation defect problem detection method using transformer substation video image | |
| Xu et al. | Adaptive remote sensing image attribute learning for active object detection | |
| Hao et al. | Based on Surf feature extraction and insulator damage identification for capsule networks | |
| Wang et al. | High-Voltage Transmission Line Foreign Object and Power Component Defect Detection Based on Improved YOLOv5 | |
| Chen et al. | Accurate object recognition for unmanned aerial vehicle electric power inspection using an improved yolov2 algorithm | |
| CN116485802B (en) | Insulator flashover defect detection method, device, equipment and storage medium | |
| CN112861952B (en) | Partial discharge image matching deep learning method | |
| CN117173385B (en) | Detection method, device, medium and equipment of transformer substation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| TA01 | Transfer of patent application right | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20240307 Address after: 518000 1002, Building A, Zhiyun Industrial Park, No. 13, Huaxing Road, Henglang Community, Longhua District, Shenzhen, Guangdong Province Applicant after: Shenzhen Wanzhida Technology Co.,Ltd. Country or region after: China Address before: 710064 middle section, south two ring road, Shaanxi, Xi'an Applicant before: CHANG'AN University Country or region before: China |
|
| TA01 | Transfer of patent application right | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20240325 Address after: 458000 International (Hebi) Science and Technology Entrepreneurship Park D20, Liupanshan Road and Xinanjiang Road, Urban Rural Integration Demonstration Zone, Hebi City, Henan Province Applicant after: Henan Huijia Intelligent Technology Co.,Ltd. Country or region after: China Address before: 518000 1002, Building A, Zhiyun Industrial Park, No. 13, Huaxing Road, Henglang Community, Longhua District, Shenzhen, Guangdong Province Applicant before: Shenzhen Wanzhida Technology Co.,Ltd. Country or region before: China |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |