CN106951863B - Method for detecting change of infrared image of substation equipment based on random forest - Google Patents
Method for detecting change of infrared image of substation equipment based on random forest Download PDFInfo
- Publication number
- CN106951863B CN106951863B CN201710166168.3A CN201710166168A CN106951863B CN 106951863 B CN106951863 B CN 106951863B CN 201710166168 A CN201710166168 A CN 201710166168A CN 106951863 B CN106951863 B CN 106951863B
- Authority
- CN
- China
- Prior art keywords
- image
- gray level
- differential
- change detection
- detection result
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
-
- 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
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/56—Extraction of image or video features relating to colour
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/60—Extraction of image or video features relating to illumination properties, e.g. using a reflectance or lighting model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10024—Color image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10048—Infrared image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20212—Image combination
- G06T2207/20224—Image subtraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/467—Encoded features or binary features, e.g. local binary patterns [LBP]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a method for detecting the change of infrared images of substation equipment based on random forests, which comprises the following steps that step 1, registered infrared images acquired by the same equipment at two different times are acquired as original infrared images; step 2, extracting a differential color image and a differential gray level image from the two original images in the step 1; step 3, extracting features by utilizing the differential color image and the differential gray level image generated in the step 2; step 4, manufacturing a sample set to form a training sample and a test sample; step 5, training and testing, and performing cross verification to obtain a preliminary infrared image change detection result image; step 6, morphological filtering is carried out on the infrared image change detection result obtained in the step 5, and a final change detection result image is obtained; the method solves the technical problems of high subjectivity, poor judgment accuracy, low accuracy rate, high error rate and the like of manual inspection in the prior art due to the fact that the image processing technology adopts the processing technology of remote sensing images.
Description
Technical Field
The invention belongs to an infrared image detection technology, and particularly relates to a substation equipment infrared image change detection method based on a random forest.
Background
In the transformer substation, the target equipment needs to be inspected so as to discover possible equipment damage in time and remove faults in time, thereby avoiding unnecessary economic loss and serious accidents caused by the unnecessary economic loss. In the manual inspection process, a large amount of time and manpower resources are consumed, and the inspection cost is increased. Moreover, the subjectivity of judgment exists in manual inspection, different inspection staff can make different judgment on the equipment state, and along with the increase of inspection intensity, the judgment accuracy of the inspection staff on the equipment state is drastically reduced. In this context, in order to release human resources, many enterprises or substations have begun to study inspection work of the electric power transmission device by using robots instead of manual work.
As a patrol robot for a substation to patrol, the patrol robot needs to use sensors to sense environment and equipment information, wherein the final sensing equipment is a visual equipment, namely a visible light camera and an infrared camera. The optical camera is mainly used for reading the number and the liquid level of the instrument, the infrared camera is mainly a thermal infrared imager, and infrared rays radiated by the environment and the target equipment are converted into infrared images to be output through conversion, so that an image change detection technology is needed; most of the existing image or scene change detection technologies are developed aiming at remote sensing images, and can be used for monitoring land and soil resources or forest coverage and farmland area transition. And can be used for monitoring changes in military bases. Thus, some classical change detection methods, such as differential imaging (Image Differencing), ratio imaging (Image ranging), change vector analysis CVA (Change Vector Analysis), and principal component analysis PCA (Principal Component Analysis), are also left. The differential image method is the simplest, two images acquired at different time are directly subtracted, and then an appropriate change region segmentation threshold value is found by adopting automatic threshold processing to predict a change region. The ratio image method carries out quotient formation on the two images to obtain a quotient image, wherein the quotient is close to 1 and the quotient is unchanged, and the quotient is deviated from 1 and the quotient is changed. The change vector analysis method mainly obtains a change vector from multi-band (multi-channel) values of two images through correlation conversion (such as difference), and obtains the degree and the type of change by solving the amplitude and the phase angle of the change vector. The principal component analysis method is that firstly, multi-band images of two-phase images are subjected to PCA transformation and mapped to a feature vector space, then, first principal components of the multi-band images are taken as differences to obtain a first principal component difference image, and a change area is determined through a threshold value.
However, the prior art is used for remote sensing image detection processing, and is not suitable for a change detection method of substation inspection infrared images, and the problems of low accuracy, high error rate and the like exist by directly applying the image detection method.
The invention comprises the following steps:
the invention aims to solve the technical problems that: the method for detecting the change of the infrared image of the substation equipment based on the random forest is provided, and the technical problems that in the prior art, the subjectivity is high when the substation inspection is performed manually, the judgment accuracy of inspection personnel on the equipment state is poor, the accuracy is low when the robot infrared inspection is performed by adopting a remote sensing image processing technology, the error rate is high and the like are solved.
The technical scheme of the invention is as follows:
a method for detecting the change of infrared images of substation equipment based on random forests comprises the following steps:
step 1, acquiring registered infrared images acquired by the same equipment at two different times as original infrared images;
step 2, extracting a differential color image and a differential gray level image from the two original images in the step 1;
step 3, extracting features by utilizing the differential color image and the differential gray level image generated in the step 2;
step 4, manufacturing a sample set to form a training sample and a test sample;
step 5, training and testing, and performing cross verification to obtain a preliminary infrared image change detection result image;
and 6, morphological filtering is carried out on the infrared image change detection result obtained in the step 5, and a final change detection result image is obtained.
It also includes:
and 7, evaluating the detection result through the performance evaluation indexes CSI (Card Similarity Index) and AE (Area Error) of the detection result.
The process of extracting the differential color image and the differential gray image in the step 2 is as follows: subtracting the pixel points of the two-phase original infrared image obtained in the step 1 to obtain a differential color image; converting the two-phase original infrared image obtained in the step into a gray level image, and then subtracting pixel points of the two gray level images from pixel points of the two gray level images to obtain a differential gray level image.
The feature extraction in step 3 includes: gray scale features, RGB features, gray scale average features, LBP features; gray histogram features and texture features.
The feature extraction method comprises the following steps:
directly taking the gray intensity of each pixel point as the gray characteristic of the pixel point by utilizing the differential gray image obtained in the step 2;
the 3 channel values of the differential color image obtained in the step 2 at each pixel point are connected in series to form a 3-dimensional vector which is used as the RGB feature of the pixel point;
taking each pixel point of the differential gray level image obtained in the step 2 as a center, opening a 3*3 window, carrying out weighted average on all pixel points of the window, wherein a weighted matrix is 3*3 and is subjected to Gaussian distribution, and the obtained weighted average is used as gray level average characteristics of the pixel points;
taking each pixel point of the differential gray image obtained in the step 2 as a center, opening a window of 3*3, comparing the sizes of 8 neighborhood pixel points and the center pixel point, if the neighborhood pixel is large, assigning 1 to the corresponding neighborhood pixel, otherwise assigning 0, stringing the 0 and 1 obtained in the 8 neighborhood together to obtain an unsigned binary number, performing cyclic shift operation on the binary number, and converting the minimum unsigned binary number into a decimal number to be used as the LBP characteristic of the pixel point;
converting the differential gray level image obtained in the step 2 into a gray level range of 16 gray levels from 0 to 15, then opening a 3*3 window by taking each pixel point as the center, counting the occurrence times of each gray level in the window, and forming a 16-dimensional gray level histogram characteristic by parallel-serial connection;
normalizing the differential gray level diagram obtained in the step 2 to 16 gray levels in total, taking each pixel point as a center, opening a 5*5 window, calculating a gray level co-occurrence matrix with the size of 16 x 16 and the step length of 1 of the window matrix in four directions, respectively calculating the values of 5 attributes of energy, entropy, correlation, moment of inertia and inverse moment of the 4 gray level co-occurrence matrices, and respectively taking the average value and standard deviation of the 5 attributes as two characteristic values, so that a 10-dimensional texture feature can be obtained by threading; fusing the 6 features into a 32-dimensional feature vector;
the method for preparing the sample set and forming the training sample and the test sample in the step 4 is as follows: fusing the features extracted in the step 3 together to extract 32-dimensional features for each pixel point to form a sample mode; calibrating a reference change image according to priori knowledge, and extracting sample labels from all pixel points according to the reference change image; a sample set is formed by the sample mode and the sample label; one part of the sample set is randomly selected as a training sample, and the other part is selected as a test sample.
Training and testing, and performing cross verification to obtain a preliminary infrared image change detection result image, wherein the method comprises the following steps: taking 3 groups of training samples each time to train the random forest classifier, putting the rest 3 groups of samples as test samples into the trained random forest classifier for change detection, performing 20 groups of cross validation in total, and obtaining 3 change detection results of 3 test samples each time, thus obtaining 60 detection results finally.
The detection result is evaluated by the performance evaluation indexes CSI (Card Similarity Index) and AE (Area Error) of the detection result, and the formula is as follows:
the larger the CSI is, the better the detection result is; the smaller the AE is, the better the detection effect is; v (V) A Representing the prospect of the change detection result, V R Representing a reference change image foreground.
The invention has the beneficial effects that:
the invention mainly comprises the following steps: (1) The image preprocessing is mainly used for extracting gray differential images of two infrared images obtained at different time and preparing for feature extraction; (2) The feature extraction is used for representing each pixel point in the feature space; (3) Cross validation, which is used for training a random forest classifier, and then predicting a change area by using the classifier; (4) Morphological filtering, namely performing morphological processing on the detection result, and removing noise points and filling holes; (5) Performance index design, adding some indexes for evaluating detection results on the basis of the prior art indexes; compared with the prior art, the method has great superiority, and solves the technical problems that the subjectivity is high, the judgment accuracy of equipment states by patrol staff is poor, the accuracy is low, the error rate is high and the like in the prior art when the substation patrol is performed manually by adopting a remote sensing image processing technology through the robot infrared patrol.
The invention is characterized in that:
(1) The invention provides visual support for intelligent inspection of the inspection robot.
(2) The invention eliminates the equipment state judgment deviation caused by subjective judgment and experience difference in manual inspection of the transformer substation, and greatly improves the accuracy and stability of equipment fault identification and early warning.
(3) According to the invention, human resources are released from substation inspection, and the production efficiency is improved.
(4) The machine learning method is used for substation inspection, and contributes to the intellectualization of substation inspection.
(5) The present invention finds a feature descriptor suitable for describing a change in the state of a device.
Drawings
FIG. 1 is an overall flow chart of the present invention;
FIG. 2 is a schematic diagram of a set of original samples and test results and comparison method results according to an embodiment of the present invention;
FIG. 3 is data of performance evaluation of results obtained from experiments of the method of the present invention and prior art comparative methods;
FIG. 4 is a graph of the average, maximum, and minimum values of the performance metrics of the method cross-validation 60 sets of results.
The specific embodiment is as follows:
a method for detecting the change of infrared images of substation equipment based on random forests comprises the following steps:
step 1, acquiring registered infrared images acquired by the same equipment at two different times as original infrared images;
step 2, extracting a differential color image and a differential gray level image from the two original images in the step 1;
step 3, extracting features by utilizing the differential color image and the differential gray level image generated in the step 2;
step 4, manufacturing a sample set to form a training sample and a test sample;
step 5, training and testing, and performing cross verification to obtain a preliminary infrared image change detection result image;
and 6, morphological filtering is carried out on the infrared image change detection result obtained in the step 5, and a final change detection result image is obtained.
It also includes:
and 7, evaluating the detection result through the performance evaluation indexes CSI (Card Similarity Index) and AE (Area Error) of the detection result.
The process of extracting the differential color image and the differential gray image in the step 2 is as follows: subtracting the pixel points of the two-phase original infrared image obtained in the step 1 to obtain a differential color image; converting the two-phase original infrared image obtained in the step into a gray level image, and then subtracting pixel points of the two gray level images from pixel points of the two gray level images to obtain a differential gray level image.
The feature extraction in step 3 includes: gray scale features, RGB features, gray scale average features, LBP features; gray histogram features and texture features.
The feature extraction method comprises the following steps:
directly taking the gray intensity of each pixel point as the gray characteristic of the pixel point by utilizing the differential gray image obtained in the step 2;
the 3 channel values of the differential color image obtained in the step 2 at each pixel point are connected in series to form a 3-dimensional vector which is used as the RGB feature of the pixel point;
taking each pixel point of the differential gray level image obtained in the step 2 as a center, opening a 3*3 window, carrying out weighted average on all pixel points of the window, wherein a weighted matrix is 3*3 and is subjected to Gaussian distribution, and the obtained weighted average is used as gray level average characteristics of the pixel points;
taking each pixel point of the differential gray image obtained in the step 2 as a center, opening a window of 3*3, comparing the sizes of 8 neighborhood pixel points and the center pixel point, if the neighborhood pixel is large, assigning 1 to the corresponding neighborhood pixel, otherwise assigning 0, stringing the 0 and 1 obtained in the 8 neighborhood together to obtain an unsigned binary number, performing cyclic shift operation on the binary number, and converting the minimum unsigned binary number into a decimal number to be used as the LBP characteristic of the pixel point;
converting the differential gray level image obtained in the step 2 into a gray level range of 16 gray levels from 0 to 15, then opening a 3*3 window by taking each pixel point as the center, counting the occurrence times of each gray level in the window, and forming a 16-dimensional gray level histogram characteristic by parallel-serial connection;
normalizing the differential gray level diagram obtained in the step 2 to 16 gray levels in total, taking each pixel point as a center, opening a 5*5 window, calculating a gray level co-occurrence matrix with the size of 16 x 16 and the step length of 1 of the window matrix in four directions, respectively calculating the values of 5 attributes of energy, entropy, correlation, moment of inertia and inverse moment of the 4 gray level co-occurrence matrices, and respectively taking the average value and standard deviation of the 5 attributes as two characteristic values, so that a 10-dimensional texture feature can be obtained by threading; fusing the 6 features into a 32-dimensional feature vector;
the method for preparing the sample set and forming the training sample and the test sample in the step 4 is as follows: fusing the features extracted in the step 3 together to extract 32-dimensional features for each pixel point to form a sample mode; calibrating a reference change image according to priori knowledge, and extracting sample labels from all pixel points according to the reference change image; a sample set is formed by the sample mode and the sample label; one part of the sample set is randomly selected as a training sample, and the other part is selected as a test sample.
Training and testing, and performing cross verification to obtain a preliminary infrared image change detection result image, wherein the method comprises the following steps: taking 3 groups of training samples each time to train the random forest classifier, putting the rest 3 groups of samples as test samples into the trained random forest classifier for change detection, performing 20 groups of cross validation in total, and obtaining 3 change detection results of 3 test samples each time, thus obtaining 60 detection results finally.
The detection result is evaluated by the performance evaluation indexes CSI (Card Similarity Index) and AE (Area Error) of the detection result, and the formula is as follows:
the larger the CSI is, the better the detection result is; the smaller the AE is, the better the detection effect is; v (V) A Representing the prospect of the change detection result, V R Representing a reference change image foreground.
The process according to the invention is described in detail below with reference to the accompanying drawings.
As shown in fig. 1, first a calibrated raw infrared image is acquired of the same target device acquired at two different times. The two images are then respectively converted into grayscale images.
And then extracting the differential color image and the differential gray level image, wherein the extraction process is to subtract the two color images and the two gray level images pixel by pixel to obtain the differential gray level image and the differential color image.
After the differential gray level image and the differential color image are obtained, feature extraction can be performed on the two differential images. There are 6 categories of features that need to be extracted. The method comprises the steps of extracting 1 category of RGB features through a differential color image, wherein 5 types of features can be extracted on the differential gray image: gray scale features, gray scale average features, gray scale histogram features, LBP features, and texture features based on gray scale co-occurrence matrices.
The characteristic extraction process of the 6 categories is as follows:
directly taking the gray intensity of each pixel point as the gray characteristic of the pixel point by utilizing the differential gray image obtained in the step 2;
the 3 channel values of the differential color image obtained in the step 2 at each pixel point are connected in series to form a 3-dimensional vector which is used as the RGB feature of the pixel point;
taking each pixel point of the differential gray level image obtained in the step 2 as a center, opening a 3*3 window, carrying out weighted average on all pixel points of the window, wherein a weighted matrix is 3*3 and is subjected to Gaussian distribution, and the obtained weighted average is used as gray level average characteristics of the pixel points;
taking each pixel point of the differential gray image obtained in the step 2 as a center, opening a window of 3*3, comparing the sizes of 8 neighborhood pixel points and the center pixel point, if the neighborhood pixel is larger, assigning 1 to the corresponding neighborhood pixel, otherwise assigning 0, stringing the 0 and 1 obtained in the 8 neighborhood together to obtain an unsigned binary number, performing cyclic shift operation on the binary number, and converting the minimum unsigned binary number into decimal number to be used as the LBP characteristic of the pixel point;
converting the differential gray level image obtained in the step 2 into a gray level range of 16 gray levels from 0 to 15, then opening a 3*3 window by taking each pixel point as the center, counting the occurrence times of each gray level in the window, and forming a 16-dimensional gray level histogram characteristic by parallel-serial connection;
normalizing the differential gray level diagram obtained in the step 2 to 16 gray levels in total, taking each pixel point as a center, opening a 5*5 window, calculating a gray level co-occurrence matrix with the size of 16 x 16 and the step length of 1 of the window matrix in four directions, respectively calculating the values of 5 attributes of energy, entropy, correlation, moment of inertia and inverse moment of the 4 gray level co-occurrence matrices, and respectively taking the average value and standard deviation of the 5 attributes as two characteristic values, so that a 10-dimensional texture feature can be obtained by threading;
after the feature extraction is completed, the 6 types of features are connected in series and fused into a 32-dimensional feature vector, the whole image is traversed, and each pixel point can obtain a 32-dimensional feature vector. Each pixel point is considered to be one sample.
And extracting labels from the reference change image for each sample pixel point according to the prior knowledge to the reference change image.
The sample vectors and corresponding labels together form a sample set.
A portion is randomly selected from the sample set as a training sample and the remaining portion is used as a test sample. And sending the data into a random forest classifier for cross verification.
The cross verification process is as follows: 3 groups of training samples are taken each time to train the random forest classifier, the rest 3 groups of samples are taken as test samples to be put into the trained random forest classifier for change detection, 20 groups of cross validation can be performed in total, 3 change detection results of 3 test samples can be obtained through each cross validation, and therefore 60 detection results are finally obtained.
And after the cross verification is finished, obtaining a preliminary substation equipment infrared image change detection result.
Morphological filtering is carried out on the obtained preliminary change detection result so as to fill holes and isolated noise points generated by the result in the prediction process.
Morphological filtering is the operation of performing expansion corrosion, open operation and close operation on the preliminary result image. And after the morphological filtering of the preliminary change detection result is completed, a final substation equipment infrared image change detection result can be obtained.
Claims (7)
1. A method for detecting the change of infrared images of substation equipment based on random forests comprises the following steps:
step 1, acquiring registered infrared images acquired by the same equipment at two different times as original infrared images;
step 2, extracting a differential color image and a differential gray level image from the two original images in the step 1;
step 3, extracting features by utilizing the differential color image and the differential gray level image generated in the step 2; the feature extraction in step 3 includes: gray scale features, RGB features, gray scale average features, LBP features, gray scale histogram features, and texture features;
step 4, manufacturing a sample set to form a training sample and a test sample;
step 5, training and testing, and performing cross verification to obtain a preliminary infrared image change detection result image;
and 6, morphological filtering is carried out on the infrared image change detection result obtained in the step 5, and a final change detection result image is obtained.
2. The random forest based substation equipment infrared image change detection method according to claim 1, wherein: it also includes:
and 7, evaluating the detection result through the performance evaluation indexes CSI (Card Similarity Index) and AE (Area Error) of the detection result.
3. The random forest based substation equipment infrared image change detection method according to claim 1, wherein: the process of extracting the differential color image and the differential gray image in the step 2 is as follows: subtracting the pixel points of the two-phase original infrared image obtained in the step 1 to obtain a differential color image; converting the two-phase original infrared image obtained in the step into a gray level image, and then subtracting pixel points of the two gray level images from pixel points of the two gray level images to obtain a differential gray level image.
4. The random forest based substation equipment infrared image change detection method according to claim 1, wherein: the feature extraction method comprises the following steps:
directly taking the gray intensity of each pixel point as the gray characteristic of the pixel point by utilizing the differential gray image obtained in the step 2;
the 3 channel values of the differential color image obtained in the step 2 at each pixel point are connected in series to form a 3-dimensional vector which is used as the RGB feature of the pixel point;
taking each pixel point of the differential gray level image obtained in the step 2 as a center, opening a 3*3 window, carrying out weighted average on all pixel points of the window, wherein a weighted matrix is 3*3 and is subjected to Gaussian distribution, and the obtained weighted average is used as gray level average characteristics of the pixel points;
taking each pixel point of the differential gray image obtained in the step 2 as a center, opening a window of 3*3, comparing the sizes of 8 neighborhood pixel points and the center pixel point, if the neighborhood pixel is large, assigning 1 to the corresponding neighborhood pixel, otherwise assigning 0, stringing the 0 and 1 obtained in the 8 neighborhood together to obtain an unsigned binary number, performing cyclic shift operation on the binary number, and converting the minimum unsigned binary number into a decimal number to be used as the LBP characteristic of the pixel point;
converting the differential gray level image obtained in the step 2 into a gray level range of 16 gray levels from 0 to 15, then opening a 3*3 window by taking each pixel point as the center, counting the occurrence times of each gray level in the window, and forming a 16-dimensional gray level histogram characteristic by parallel-serial connection;
normalizing the differential gray level diagram obtained in the step 2 to 16 gray levels in total, taking each pixel point as a center, opening a 5*5 window, calculating a gray level co-occurrence matrix with the size of 16 x 16 and the step length of 1 of the window matrix in four directions, respectively calculating the values of 5 attributes of energy, entropy, correlation, moment of inertia and inverse moment of the 4 gray level co-occurrence matrices, and respectively taking the average value and standard deviation of the 5 attributes as two characteristic values, so that a 10-dimensional texture feature can be obtained by threading; the 6 features are fused into a 32-dimensional feature vector.
5. The random forest based substation equipment infrared image change detection method according to claim 1, wherein: the method for preparing the sample set and forming the training sample and the test sample in the step 4 is as follows:
fusing the features extracted in the step 3 together to extract 32-dimensional features for each pixel point to form a sample mode; calibrating a reference change image according to priori knowledge, and extracting sample labels from all pixel points according to the reference change image; a sample set is formed by the sample mode and the sample label; one part of the sample set is randomly selected as a training sample, and the other part is selected as a test sample.
6. The random forest based substation equipment infrared image change detection method according to claim 1, wherein: training and testing, and performing cross verification to obtain a preliminary infrared image change detection result image, wherein the method comprises the following steps: taking 3 groups of training samples each time to train the random forest classifier, putting the rest 3 groups of samples as test samples into the trained random forest classifier for change detection, performing 20 groups of cross validation in total, and obtaining 3 change detection results of 3 test samples each time, thus obtaining 60 detection results finally.
7. The random forest based substation equipment infrared image change detection method according to claim 2, wherein: the detection result is evaluated by the performance evaluation indexes CSI (Card Similarity Index) and AE (Area Error) of the detection result, and the formula is as follows:
the larger the CSI is, the better the detection result is; the smaller the AE is, the better the detection effect is; v (V) A Representing the prospect of the change detection result, V R Representing a reference change image foreground.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710166168.3A CN106951863B (en) | 2017-03-20 | 2017-03-20 | Method for detecting change of infrared image of substation equipment based on random forest |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710166168.3A CN106951863B (en) | 2017-03-20 | 2017-03-20 | Method for detecting change of infrared image of substation equipment based on random forest |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106951863A CN106951863A (en) | 2017-07-14 |
| CN106951863B true CN106951863B (en) | 2023-09-26 |
Family
ID=59472016
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710166168.3A Active CN106951863B (en) | 2017-03-20 | 2017-03-20 | Method for detecting change of infrared image of substation equipment based on random forest |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106951863B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109086801A (en) * | 2018-07-06 | 2018-12-25 | 湖北工业大学 | A kind of image classification method based on improvement LBP feature extraction |
| CN108535618A (en) * | 2018-07-11 | 2018-09-14 | 云南电网有限责任公司电力科学研究院 | A kind of GIS method for detecting insulation defect |
| CN110263608B (en) * | 2019-01-25 | 2023-07-07 | 天津职业技术师范大学(中国职业培训指导教师进修中心) | Automatic electronic component identification method based on image feature space variable threshold measurement |
| CN110119710A (en) * | 2019-05-13 | 2019-08-13 | 广州锟元方青医疗科技有限公司 | Cell sorting method, device, computer equipment and storage medium |
| CN113503970A (en) * | 2021-07-08 | 2021-10-15 | 江苏高速公路信息工程有限公司 | Fire monitoring method and system |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1885346A (en) * | 2006-06-01 | 2006-12-27 | 电子科技大学 | Detection method for moving target in infrared image sequence under complex background |
| CN102499649A (en) * | 2011-11-19 | 2012-06-20 | 江苏大学 | Pig body temperature detection method based on machine vision |
| CN104331521A (en) * | 2014-11-27 | 2015-02-04 | 国家电网公司 | Automatic power transformation equipment abnormity identification method based on image processing |
| CN104899558A (en) * | 2015-05-25 | 2015-09-09 | 东华大学 | Scene recognition and colorization processing method for vehicle-mounted infrared image |
| CN105023269A (en) * | 2015-04-27 | 2015-11-04 | 东华大学 | Vehicle-mounted infrared image colorization method |
| CN105300482A (en) * | 2015-09-28 | 2016-02-03 | 北京石油化工学院 | Water meter calibration method, apparatus and system based on image processing |
| CN105469098A (en) * | 2015-11-20 | 2016-04-06 | 中北大学 | Precise LINDAR data ground object classification method based on adaptive characteristic weight synthesis |
| WO2016200406A1 (en) * | 2015-06-12 | 2016-12-15 | Intergraph Corporation | Image change detection |
| CN106295636A (en) * | 2016-07-21 | 2017-01-04 | 重庆大学 | Passageway for fire apparatus based on multiple features fusion cascade classifier vehicle checking method |
| CN111869291A (en) * | 2018-02-02 | 2020-10-30 | 康奈尔大学 | Channel mapping in a wireless system |
-
2017
- 2017-03-20 CN CN201710166168.3A patent/CN106951863B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1885346A (en) * | 2006-06-01 | 2006-12-27 | 电子科技大学 | Detection method for moving target in infrared image sequence under complex background |
| CN102499649A (en) * | 2011-11-19 | 2012-06-20 | 江苏大学 | Pig body temperature detection method based on machine vision |
| CN104331521A (en) * | 2014-11-27 | 2015-02-04 | 国家电网公司 | Automatic power transformation equipment abnormity identification method based on image processing |
| CN105023269A (en) * | 2015-04-27 | 2015-11-04 | 东华大学 | Vehicle-mounted infrared image colorization method |
| CN104899558A (en) * | 2015-05-25 | 2015-09-09 | 东华大学 | Scene recognition and colorization processing method for vehicle-mounted infrared image |
| WO2016200406A1 (en) * | 2015-06-12 | 2016-12-15 | Intergraph Corporation | Image change detection |
| CN105300482A (en) * | 2015-09-28 | 2016-02-03 | 北京石油化工学院 | Water meter calibration method, apparatus and system based on image processing |
| CN105469098A (en) * | 2015-11-20 | 2016-04-06 | 中北大学 | Precise LINDAR data ground object classification method based on adaptive characteristic weight synthesis |
| CN106295636A (en) * | 2016-07-21 | 2017-01-04 | 重庆大学 | Passageway for fire apparatus based on multiple features fusion cascade classifier vehicle checking method |
| CN111869291A (en) * | 2018-02-02 | 2020-10-30 | 康奈尔大学 | Channel mapping in a wireless system |
Non-Patent Citations (1)
| Title |
|---|
| 红外序列图像地面目标检测及特征提取技术研究;江小明;《中国优秀硕士学位论文全文数据库 信息科技辑》;I138-695 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106951863A (en) | 2017-07-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112380952B (en) | Power equipment infrared image real-time detection and identification method based on artificial intelligence | |
| CN106951863B (en) | Method for detecting change of infrared image of substation equipment based on random forest | |
| CN109118479B (en) | Capsule network-based insulator defect identification and positioning device and method | |
| CN107944396B (en) | Knife switch state identification method based on improved deep learning | |
| CN104966085B (en) | A kind of remote sensing images region of interest area detecting method based on the fusion of more notable features | |
| CN111784633B (en) | Insulator defect automatic detection algorithm for electric power inspection video | |
| CN108428231B (en) | Multi-parameter part surface roughness learning method based on random forest | |
| CN111639587B (en) | Hyperspectral image classification method based on multi-scale spectrum space convolution neural network | |
| CN111985499B (en) | High-precision bridge apparent disease identification method based on computer vision | |
| CN110598613B (en) | Expressway agglomerate fog monitoring method | |
| CN106557740B (en) | The recognition methods of oil depot target in a kind of remote sensing images | |
| CN114494185B (en) | Electrical equipment fault detection method based on RGB-T multi-scale feature fusion | |
| CN107358259A (en) | Covering ice for insulator detection method based on GLOH descriptions and GVF Snake models | |
| CN109784205B (en) | Intelligent weed identification method based on multispectral inspection image | |
| CN114841972A (en) | Power transmission line defect identification method based on saliency map and semantic embedded feature pyramid | |
| CN103824092A (en) | Image classification method for monitoring state of electric transmission and transformation equipment on line | |
| CN113255590A (en) | Defect detection model training method, defect detection method, device and system | |
| CN113065484A (en) | Insulator contamination state assessment method based on ultraviolet spectrum | |
| CN116681979A (en) | Power equipment target detection method under complex environment | |
| CN112084860A (en) | Target object detection method and device and thermal power plant detection method and device | |
| CN108596244A (en) | A kind of high spectrum image label noise detecting method based on spectrum angle density peaks | |
| Tao et al. | Electric insulator detection of UAV images based on depth learning | |
| CN116519710A (en) | Method and system for detecting surface pollution state of composite insulator | |
| CN116051539A (en) | Diagnosis method for heating fault of power transformation equipment | |
| CN112508946B (en) | Cable tunnel anomaly detection method based on antagonistic neural network |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |