CN114724041B - Power equipment infrared chart identification method and system based on deep learning - Google Patents
Power equipment infrared chart identification method and system based on deep learning Download PDFInfo
- Publication number
- CN114724041B CN114724041B CN202210618590.9A CN202210618590A CN114724041B CN 114724041 B CN114724041 B CN 114724041B CN 202210618590 A CN202210618590 A CN 202210618590A CN 114724041 B CN114724041 B CN 114724041B
- Authority
- CN
- China
- Prior art keywords
- image
- area
- infrared image
- radiation area
- real working
- 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
Images
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
-
- 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)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention relates to the technical field of artificial intelligence, in particular to a method and a system for identifying an infrared chart of electric equipment based on deep learning. Acquiring an infrared image and an RGB image of the power equipment, and marking the real working area and the energy radiation area in the infrared image according to the real working area and the energy radiation area in the RGB image; training an image anomaly detection network by using the marked infrared image; and carrying out anomaly detection on the infrared image through the trained image anomaly detection network, and carrying out anomaly early warning on the power equipment according to a detection result. The infrared images of different power equipment can be labeled by combining the gray level difference of the pixel points in the RGB images and the infrared images, the problem of insufficient network generalization capability is solved, and the network training speed and the network anomaly detection precision are improved.
Description
Technical Field
The invention relates to the technical field of artificial intelligence, in particular to a method and a system for identifying an infrared chart of electric equipment based on deep learning.
Background
With the continuous development of the modern society, the demand of daily life for electric power is greater, so that the demand for electric power equipment is higher and higher, and the abnormity detection and management for the electric power equipment are improved continuously. In the prior art, the abnormality detection of the power equipment is mostly carried out by using infrared images, namely, the infrared images are mainly trained by using a deep neural network, and the abnormality detection is carried out according to the conventional VGG16 and Resnet 50.
Those skilled in the art have found the following problems in the prior art: the existing neural network is used for detection, a large number of image samples are required to participate in training, the generalization capability of the network is improved, the training cost is higher, and meanwhile, the trained network identification precision is limited due to the quality problem or semantic information missing problem of the image samples participating in the training.
Disclosure of Invention
In order to solve the above technical problems, an object of the present invention is to provide a method and a system for identifying an infrared chart of an electrical device based on deep learning, wherein the adopted technical scheme is as follows:
in a first aspect, an embodiment of the present invention provides a method for identifying an infrared thermal map of an electrical device based on deep learning, where the method includes the following steps:
acquiring an infrared image and an RGB image of power equipment, and labeling a real working area and an energy radiation area in the infrared image according to the real working area and the energy radiation area in the RGB image;
training an image anomaly detection network by using the marked infrared image; and carrying out anomaly detection on the infrared image through the trained image anomaly detection network, and carrying out anomaly early warning on the power equipment according to a detection result.
Further, the method for labeling the real working area and the energy radiation area in the infrared image according to the real working area and the energy radiation area in the RGB image includes:
acquiring a real working area and an energy radiation area in the RGB image by utilizing a semantic segmentation network;
mapping the real working area and the energy radiation area in the infrared image to obtain an initial real working area and an initial energy radiation area;
acquiring a segmentation edge radiation area of the infrared image based on edge pixel points of the initial real working area in the infrared image;
and re-labeling the actual real working area and the actual energy radiation area of the infrared image according to the segmentation edge radiation area.
Further, the method for obtaining the segmented edge radiation region of the infrared image based on the edge pixel point of the initial real working region in the infrared image includes:
taking the pixel points on the edge of the initial real working area as initial growth seed points, and calculating gray level differences between the initial production seed points and the pixel points in eight neighborhoods of the initial production seed points;
when all the gray differences are smaller than the gray difference threshold, taking the pixel points outside the initial real working area in the pixel points in the eight neighborhoods as next growth seed points; when part of gray level difference is smaller than a gray level difference threshold, acquiring pixel points smaller than a gray level abnormal threshold in eight neighborhoods, selecting the pixel point with the largest gray level difference in the pixel points as a next growth seed point, and stopping seed growth until the gray level difference between the growth seed point and all the pixel points in the eight neighborhoods is larger than the gray level difference threshold;
and forming a segmentation edge radiation area by the pixel points corresponding to the participation of the seed growth.
Further, the method for re-labeling the actual real working area and the actual energy radiation area of the infrared image according to the segmentation edge radiation area includes:
and taking the area within the segmentation edge radiation area as the actual real working area in the infrared image and the area between the segmentation edge radiation area and the initial energy radiation area as the actual energy radiation area in the infrared image.
Further, the method for training the image anomaly detection network by using the labeled infrared image comprises the following steps:
inputting the marked infrared image as a training set into the image anomaly detection network, and outputting an anomaly confidence coefficient corresponding to the infrared image;
in the training process of the image anomaly detection network, adjusting the feature weight in the convolution kernel sliding window process according to the labeling position of the segmentation edge radiation area, and acquiring the feature vector of the image according to the adjusted feature weight;
and the loss function of the image anomaly detection network is a cross entropy loss function.
Further, the adjusted feature weight and the minimum distance between the convolution kernel sliding window and the labeling position are in a negative correlation relationship, and the adjusted feature weight and the number of convolution layers are in a positive correlation relationship.
In a second aspect, another embodiment of the present invention provides a system for identifying an infrared thermal map of an electrical device based on deep learning, the system including:
the image labeling unit is used for acquiring an infrared image and an RGB image of the power equipment and labeling a real working area and an energy radiation area in the infrared image according to the real working area and the energy radiation area in the RGB image;
the image detection unit is used for training the image anomaly detection network by using the marked infrared image; and carrying out anomaly detection on the infrared image through the trained image anomaly detection network, and carrying out anomaly early warning on the power equipment according to a detection result.
Further, the method for labeling the real working area and the energy radiation area in the infrared image according to the real working area and the energy radiation area in the RGB image in the image labeling unit includes:
acquiring a real working area and an energy radiation area in the RGB image by utilizing a semantic segmentation network;
mapping the real working area and the energy radiation area in the infrared image to obtain an initial real working area and an initial energy radiation area;
acquiring a segmentation edge radiation area of the infrared image based on edge pixel points of the initial real working area in the infrared image;
and re-labeling the actual real working area and the actual energy radiation area of the infrared image according to the segmentation edge radiation area.
Further, the method for obtaining the segmentation edge radiation region of the infrared image based on the edge pixel point of the initial real working region in the infrared image in the image labeling unit includes:
taking the pixel points on the edge of the initial real working area as initial growth seed points, and calculating gray level differences between the initial production seed points and the pixel points in eight neighborhoods of the initial production seed points;
when all the gray differences are smaller than the gray difference threshold, taking the pixel points outside the initial real working area in the pixel points in the eight neighborhoods as next growth seed points; when partial gray difference is smaller than a gray difference threshold, acquiring pixel points in the eight neighborhoods which are smaller than a gray abnormal threshold, selecting the pixel points with the largest gray difference as next growth seed points until the gray difference between the growth seed points and all the pixel points in the eight neighborhoods of the growth seed points is larger than the gray difference threshold, and stopping seed growth;
and forming a segmentation edge radiation area by the pixel points corresponding to the participation of the seed growth.
Further, the method for re-labeling the actual real working area and the actual energy radiation area of the infrared image according to the segmentation edge radiation area in the image labeling unit includes:
and taking the area within the segmentation edge radiation area as the actual real working area in the infrared image, and taking the area between the segmentation edge radiation area and the initial energy radiation area as the actual energy radiation area in the infrared image.
The embodiment of the invention at least has the following beneficial effects: based on the real working area and the energy radiation area of the RGB image of the power equipment, the real working area and the actual energy radiation area are marked according to the gray difference of pixel points in the infrared image, so that the infrared images of different power equipment can be marked, and the speed of network training and the accuracy of network anomaly detection are improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions and advantages of the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.
Fig. 1 is a flowchart illustrating steps of a method for identifying an infrared thermal map of an electrical device based on deep learning according to an embodiment of the present invention.
Detailed Description
To further illustrate the technical means and effects of the present invention adopted to achieve the predetermined objects, the following detailed description of the method and system for recognizing an infrared thermal map of an electrical device based on deep learning according to the present invention with reference to the accompanying drawings and preferred embodiments shows the following detailed implementation, structure, features and effects. In the following description, different "one embodiment" or "another embodiment" refers to not necessarily the same embodiment. Furthermore, the particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
The specific scheme of the method and the system for recognizing the infrared heatmap of the electrical equipment based on deep learning provided by the invention is specifically described below with reference to the accompanying drawings.
The embodiment of the invention aims at the following specific scenes: according to the power equipment infrared detection scene, the power equipment inspection robot is used for image acquisition, the RGB camera is installed on the inspection robot, and the convertible infrared filter is installed between the lens and the CCD, so that the camera can acquire infrared images and RGB images of the power equipment, the power equipment detection is completed by using the infrared images and the RGB images, and the influence of external environments such as a light source is not considered.
Referring to fig. 1, a flowchart illustrating steps of a method for identifying an infrared heatmap of an electrical device based on deep learning according to an embodiment of the present invention is shown, where the method includes the following steps:
and S001, acquiring an infrared image and an RGB image of the power equipment, and labeling the real working area and the energy radiation area in the infrared image according to the real working area and the energy radiation area in the RGB image.
Specifically, gather power equipment's infrared image and RGB image through the camera, power equipment includes: generators, motors, transformers, circuit breakers, disconnectors, fuses, etc. The RGB image is subjected to necessary denoising processing to eliminate the influence of noise points, the denoising processing adopts median filtering denoising, the median filtering denoising is a known technology, and the implementation process of the median filtering denoising is not described.
The electric power equipment can generate heat when working normally and abnormal, the area which causes heat generation is mainly the surface of the working area of the electric power equipment, meanwhile, radiation is generated due to metal conduction of the electric power equipment, and partial irrelevant areas also generate energy, so that pixel points of the irrelevant areas also have pixel values in an infrared image, but in actual infrared image analysis, the irrelevant areas are not the core areas of the electric power equipment which directly generates heat, and therefore the areas in the infrared image need to be divided. In order to achieve more accurate segmentation and avoid mistaken segmentation of part of core areas, the target detection area of the RGB image is used as auxiliary information to help the infrared image to complete optimal self-adaptive segmentation, and the specific process comprises the following steps:
(1) and acquiring a real working area and an energy radiation area in the RGB image by utilizing a semantic segmentation network.
Specifically, an RGB image of the power equipment is obtained, a semantic segmentation network is constructed to segment the RGB image, and the power equipment is divided into a real working area and an energy radiation area according to the core structure of the power equipment. The specific training process of the semantic segmentation network comprises the following steps: constructing a semantic segmentation network, wherein the network structure is as follows: an encoder-decoder; carrying out graying processing on the RGB image, and marking the pixel points belonging to a real working area as 1, the pixel points of an energy radiation area as 2 and the pixel points of other background areas as 0; and inputting the marked RGB image into a semantic segmentation network for training, extracting image feature tensors of different areas through an encoder, and outputting a real working area segmentation image and an energy radiation area segmentation image through a decoder.
(2) And mapping the real working area and the energy radiation area in the infrared image to obtain an initial real working area and an initial energy radiation area.
Specifically, since the sizes of the infrared image and the RGB image are the same, the same region is labeled on the infrared image according to the real working region and the energy radiation region in the RGB image, that is, the initial real working region in the infrared image is labeled as a region a, and the initial energy radiation region is labeled as a region B.
It should be noted that the infrared image is an infrared image of the power device in the normal operation state.
(3) Acquiring a segmentation edge radiation area of the infrared image based on edge pixel points of an initial real working area in the infrared image; and re-labeling the actual real working area and the actual energy radiation area of the infrared image according to the segmentation edge radiation area.
Specifically, because the infrared image is a single-channel grayscale image, and the pixel values of the pixels in the infrared image are the energy information of the current power equipment position, the actual real working area and the actual energy radiation area in the infrared image are re-labeled according to the grayscale difference between the pixels, that is, the energy fluctuation change, and the specific method includes: under normal conditions, due to the capability conduction phenomenon of the real working area, the gray level difference of the pixel points corresponding to the real working area is small, and the gray level difference of the pixel points corresponding to the energy radiation area is large, so that the infrared image is subjected to area re-division and labeling according to the edge pixel points in the initial real working area (area A).
As an example, taking pixel points on the edge of the area a as initial growth seed points, calculating gray level differences between the initial production seed points and pixel points in eight neighborhoods thereof, respectively, and when all gray level differences are smaller than a gray level difference threshold, taking pixel points outside an initial real working area in the pixel points in the eight neighborhoods thereof as next growth seed points until gray level differences between the growth seed points and all pixel points in the eight neighborhoods thereof are larger than the gray level difference threshold, so as to stop seed growth; and when part of gray differences are smaller than the gray difference threshold, acquiring pixel points smaller than the gray abnormal threshold in the eight neighborhoods, selecting the pixel points with the largest gray difference in the pixel points as the next growth seed points until the gray differences between the growth seed points and all the pixel points in the eight neighborhoods are larger than the gray difference threshold, stopping seed growth, forming a segmentation edge radiation area by the pixel points corresponding to the participation of seed growth, and classifying the pixel points in the segmentation edge radiation area as the pixel points of the segmentation edge radiation area.
Further, the area within the segmentation edge radiation area is used as an actual real working area in the infrared image, and the area between the segmentation edge radiation area and the initial energy radiation area is used as an actual energy radiation area in the infrared image.
It should be noted that the purpose of obtaining the divided edge radiation regions is to perform accurate region labeling on actual real working regions and energy radiation regions on infrared images of the electrical equipment in order that different electrical equipment are in normal working processes due to different structures, different materials and different generated energy regions of the electrical equipment, and further perform deep learning of a subsequent image anomaly detection network according to region labeling information of the infrared images, so that a training process can be converged more quickly, and false detection or false recognition caused by differences of the electrical equipment can be reduced.
S002, training the image anomaly detection network by using the marked infrared image; and carrying out anomaly detection on the infrared image through the trained image anomaly detection network, and carrying out anomaly early warning on the power equipment according to a detection result.
Specifically, an image anomaly detection network is constructed, the network structure of the network is an encoder-full connection layer, and the image anomaly detection network is trained by using the labeled infrared image, so that the specific training process is as follows:
(1) and taking the marked infrared image and the infrared image as input data of the image anomaly detection network, wherein the marked infrared image is the infrared image of normal power equipment and is taken as a training set, the infrared image is taken as the training set, and the image anomaly detection network outputs the anomaly confidence coefficient of the infrared image.
(2) In the network training process, the encoder adjusts the feature weight in the convolution kernel window sliding process according to the labeled position of the segmented edge radiation area in the image convolution process, and the adjusting method is as follows: and constructing a weight adjusting formula according to the distance between the sliding window position of the convolution kernel and the edge of the marked area, wherein the weight adjusting formula is as follows:
wherein, in the process,
representing the minimum distance from a sliding window of the convolution kernel to the marked area;
indicating the adjustment factor, which is generally a very small number in order to avoid a denominator of 0, empirical values are taken here
；
And expressing the number of layers of image convolution, setting the initial sliding window position of a default image convolution kernel as the upper left corner of the image, and sequentially performing image convolution rightward and downward to obtain the feature vector of the image.
The significance of the weight adjustment formula is that: the characteristic influence of an irrelevant area can be reduced through the marked area of the image, and the importance degree of the abnormal characteristic is considered to be higher as the default distance is closer to the marked area; considering that the feature weight is adjusted according to the number of the convolution layers of the image, mainly due to the problem of the receptive field, the larger the number of the network convolution layers is, the larger the receptive field of the convolution kernel is, and the more important the extracted features are.
It should be noted that the reason why the regions other than the actual real working region and the actual energy radiation region in the infrared image are not directly segmented and removed is that: by changing the weight of the position of the convolution kernel sliding window, the significance of participation of partial irrelevant area image features in feature comparison is reserved, and the overfitting condition existing in network training is avoided.
(3) The loss function of the image anomaly detection network is a cross entropy loss function.
And further, inputting the infrared image of the power equipment into a trained image anomaly detection network to obtain an anomaly confidence coefficient of the infrared image, judging whether the infrared image of the current power equipment is anomalous or not according to the anomaly confidence coefficient, storing the anomalous infrared image, and performing anomaly early warning on the current power equipment through a monitoring system.
The method comprises the steps of taking a generator device as an example, acquiring an infrared image of the generator device, training an image anomaly detection network by utilizing the infrared image of the generator device, inputting the infrared image of the generator device acquired in real time into the trained image anomaly detection network to obtain an anomaly confidence coefficient corresponding to the infrared image, setting an anomaly confidence coefficient threshold, confirming that the infrared image is abnormal when the anomaly confidence coefficient is greater than or equal to the anomaly confidence coefficient threshold, and meanwhile, indicating that the working state of the generator device is abnormal, sending an anomaly signal to the generator device through a monitoring system to remind a detector to check the generator device in time, otherwise, confirming that the infrared image is a normal image when the anomaly confidence coefficient is less than the anomaly confidence coefficient threshold, and meanwhile, indicating that the generator device is in a normal working state.
Preferably, in the embodiment of the present invention, the anomaly confidence threshold is an empirical value, and then the anomaly confidence threshold is 0.7, and an implementer may reset the anomaly confidence threshold according to a real-time scenario of the implementer.
In summary, the embodiment of the present invention provides a method for identifying an infrared thermal image of an electrical device based on deep learning, the method collects an infrared image and an RGB image of the electrical device, and labels a real working area and an energy radiation area in the infrared image according to the real working area and the energy radiation area in the RGB image; and training the image anomaly detection network by using the marked infrared image, performing anomaly detection on the infrared image through the trained image anomaly detection network, and performing anomaly early warning on the power equipment according to a detection result. Based on the real working area and the energy radiation area of the RGB image of the power equipment, the real working area and the actual energy radiation area are marked according to the gray difference of pixel points in the infrared image, so that the infrared images of different power equipment can be marked, and the speed of network training and the accuracy of network anomaly detection are improved.
Based on the same inventive concept as the method, an embodiment of the present invention further provides a system for identifying an infrared heatmap of an electrical device based on deep learning, including a memory, a processor, and a computer program stored in the memory and running on the processor, where the processor implements the steps of any one of the methods for identifying an infrared heatmap of an electrical device based on deep learning when executing the computer program.
It should be noted that: the precedence order of the above embodiments of the present invention is only for description, and does not represent the merits of the embodiments. And that specific embodiments have been described above. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims may be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing may also be possible or may be advantageous.
The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.
Claims (2)
1. A method for identifying an infrared heat map of electric equipment based on deep learning is characterized by comprising the following steps:
acquiring an infrared image and an RGB image of power equipment, and labeling a real working area and an energy radiation area in the infrared image according to the real working area and the energy radiation area in the RGB image;
training an image anomaly detection network by using the marked infrared image; carrying out anomaly detection on the infrared image through the trained image anomaly detection network, and carrying out anomaly early warning on the power equipment according to a detection result;
the method for labeling the real working area and the energy radiation area in the infrared image according to the real working area and the energy radiation area in the RGB image comprises the following steps:
acquiring a real working area and an energy radiation area in the RGB image by utilizing a semantic segmentation network;
mapping the real working area and the energy radiation area in the infrared image to obtain an initial real working area and an initial energy radiation area;
acquiring a segmentation edge radiation area of the infrared image based on edge pixel points of the initial real working area in the infrared image;
re-labeling the actual real working area and the actual energy radiation area of the infrared image according to the segmentation edge radiation area;
the method for obtaining the segmentation edge radiation area of the infrared image based on the edge pixel point of the initial real working area in the infrared image comprises the following steps:
taking the pixel points on the edge of the initial real working area as initial growth seed points, and calculating gray level differences between the initial growth seed points and the pixel points in eight neighborhoods of the initial growth seed points respectively;
when all the gray differences are smaller than the gray difference threshold, taking the pixel points outside the initial real working area in the pixel points in the eight neighborhoods as next growth seed points; when part of gray level difference is smaller than a gray level difference threshold, acquiring pixel points smaller than a gray level abnormal threshold in eight neighborhoods, selecting the pixel point with the largest gray level difference in the pixel points as a next growth seed point, and stopping seed growth until the gray level difference between the growth seed point and all the pixel points in the eight neighborhoods is larger than the gray level difference threshold;
forming a segmentation edge radiation area by pixel points corresponding to the participation of seed growth;
the method for re-labeling the actual real working area and the actual energy radiation area of the infrared image according to the segmentation edge radiation area comprises the following steps:
taking the area within the segmentation edge radiation area as the actual real working area in the infrared image, and taking the area between the segmentation edge radiation area and the initial energy radiation area as the actual energy radiation area in the infrared image;
the method for training the image anomaly detection network by using the labeled infrared image comprises the following steps:
inputting the marked infrared image as a training set into the image anomaly detection network, and outputting an anomaly confidence coefficient corresponding to the infrared image;
in the training process of the image anomaly detection network, adjusting the characteristic weight in the convolution kernel sliding window process according to the labeling position of the segmentation edge radiation area, and acquiring the characteristic vector of the image according to the adjusted characteristic weight;
the loss function of the image anomaly detection network is a cross entropy loss function;
the adjusted feature weight and the minimum distance between the convolution kernel sliding window and the marked position are in a negative correlation relationship, and the adjusted feature weight and the number of convolution layers are in a positive correlation relationship.
2. A system for identifying an infrared thermal map of an electrical device based on deep learning, the system comprising:
the image labeling unit is used for acquiring an infrared image and an RGB image of the power equipment and labeling a real working area and an energy radiation area in the infrared image according to the real working area and the energy radiation area in the RGB image;
the image detection unit is used for training the image anomaly detection network by using the marked infrared image; carrying out anomaly detection on the infrared image through the trained image anomaly detection network, and carrying out anomaly early warning on the power equipment according to a detection result;
the method for labeling the real working area and the energy radiation area in the infrared image according to the real working area and the energy radiation area in the RGB image in the image labeling unit comprises the following steps:
acquiring a real working area and an energy radiation area in the RGB image by utilizing a semantic segmentation network;
mapping the real working area and the energy radiation area in the infrared image to obtain an initial real working area and an initial energy radiation area;
acquiring a segmentation edge radiation area of the infrared image based on edge pixel points of the initial real working area in the infrared image;
re-labeling the actual real working area and the actual energy radiation area of the infrared image according to the segmentation edge radiation area;
the method for acquiring the segmentation edge radiation area of the infrared image based on the edge pixel point of the initial real working area in the infrared image in the image labeling unit comprises the following steps:
taking the pixel points on the edge of the initial real working area as initial growth seed points, and calculating gray level differences between the initial growth seed points and the pixel points in eight neighborhoods of the initial growth seed points respectively;
when all the gray differences are smaller than the gray difference threshold value, taking the pixel points outside the initial real working area in the pixel points in the eight neighborhoods as next growth seed points; when part of gray level difference is smaller than a gray level difference threshold, acquiring pixel points smaller than a gray level abnormal threshold in eight neighborhoods, selecting the pixel point with the largest gray level difference in the pixel points as a next growth seed point, and stopping seed growth until the gray level difference between the growth seed point and all the pixel points in the eight neighborhoods is larger than the gray level difference threshold;
forming a segmentation edge radiation area by pixel points corresponding to the growth of the seeds;
the method for re-labeling the actual real working area and the actual energy radiation area of the infrared image according to the segmentation edge radiation area in the image labeling unit comprises the following steps:
taking the area within the segmentation edge radiation area as the actual real working area in the infrared image and the area between the segmentation edge radiation area and the initial energy radiation area as the actual energy radiation area in the infrared image;
the method for training the image anomaly detection network by using the labeled infrared image comprises the following steps:
inputting the marked infrared image as a training set into the image anomaly detection network, and outputting an anomaly confidence coefficient corresponding to the infrared image;
in the training process of the image anomaly detection network, adjusting the characteristic weight in the convolution kernel sliding window process according to the labeling position of the segmentation edge radiation area, and acquiring the characteristic vector of the image according to the adjusted characteristic weight;
the loss function of the image anomaly detection network is a cross entropy loss function;
the adjusted feature weight and the minimum distance between the convolution kernel sliding window and the marked position are in a negative correlation relationship, and the adjusted feature weight and the number of convolution layers are in a positive correlation relationship.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210618590.9A CN114724041B (en) | 2022-06-02 | 2022-06-02 | Power equipment infrared chart identification method and system based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210618590.9A CN114724041B (en) | 2022-06-02 | 2022-06-02 | Power equipment infrared chart identification method and system based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114724041A CN114724041A (en) | 2022-07-08 |
| CN114724041B true CN114724041B (en) | 2022-09-06 |
Family
ID=82232768
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210618590.9A Active CN114724041B (en) | 2022-06-02 | 2022-06-02 | Power equipment infrared chart identification method and system based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114724041B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109271847A (en) * | 2018-08-01 | 2019-01-25 | 阿里巴巴集团控股有限公司 | Method for detecting abnormality, device and equipment in unmanned clearing scene |
| CN113989618A (en) * | 2021-11-03 | 2022-01-28 | 深圳黑蚂蚁环保科技有限公司 | Recyclable article classification and identification method |
Family Cites Families (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106909905B (en) * | 2017-03-02 | 2020-02-14 | 中科视拓（北京）科技有限公司 | Multi-mode face recognition method based on deep learning |
| CN109101906A (en) * | 2018-07-27 | 2018-12-28 | 中国南方电网有限责任公司超高压输电公司贵阳局 | A kind of converting station electric power equipment infrared image exception real-time detection method and device |
| CN109949316B (en) * | 2019-03-01 | 2020-10-27 | 东南大学 | A weakly supervised instance segmentation method for power grid equipment images based on RGB-T fusion |
| US11003945B2 (en) * | 2019-05-22 | 2021-05-11 | Zoox, Inc. | Localization using semantically segmented images |
| CN112434546A (en) * | 2019-08-26 | 2021-03-02 | 杭州魔点科技有限公司 | Face living body detection method and device, equipment and storage medium |
| CN110969610A (en) * | 2019-12-03 | 2020-04-07 | 杭州天铂云科光电科技有限公司 | Power equipment infrared chart identification method and system based on deep learning |
| CN112233073B (en) * | 2020-09-30 | 2024-11-26 | 国网山西省电力公司大同供电公司 | A real-time detection method for abnormal infrared thermal imaging of substation equipment |
| CN113935992B (en) * | 2021-12-15 | 2022-03-22 | 武汉和众成设备工贸有限公司 | Image processing-based oil pollution interference resistant gear crack detection method and system |
| CN114529779B (en) * | 2021-12-31 | 2023-04-07 | 扬州市恒泰人防设备有限公司 | System and method for detecting rust state of civil air defense door |
| CN114494185B (en) * | 2022-01-25 | 2022-09-30 | 国网吉林省电力有限公司电力科学研究院 | Electrical equipment fault detection method based on RGB-T multi-scale feature fusion |
-
2022
- 2022-06-02 CN CN202210618590.9A patent/CN114724041B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109271847A (en) * | 2018-08-01 | 2019-01-25 | 阿里巴巴集团控股有限公司 | Method for detecting abnormality, device and equipment in unmanned clearing scene |
| CN113989618A (en) * | 2021-11-03 | 2022-01-28 | 深圳黑蚂蚁环保科技有限公司 | Recyclable article classification and identification method |
Non-Patent Citations (2)
| Title |
|---|
| Deep Learning Based Fusion of RGB and Infrared Images for the Detection of Abnormal Condition of Fused Magnesium Furnace;Shaowen Lu 等;《2019 IEEE 15th International Conference on Control and Automation (ICCA)》;20190719;第987-993页 * |
| 基于VR技术的电力系统异常情况巡检系统设计;芦佳硕 等;《计算技术与自动化》;20210331;第169-173页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114724041A (en) | 2022-07-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113344475B (en) | Transformer bushing defect identification method and system based on sequence modal decomposition | |
| CN112766195B (en) | Electrified railway bow net arcing visual detection method | |
| CN117677969A (en) | Defect detection method and device | |
| CN113436184B (en) | Power equipment image defect discriminating method and system based on improved twin network | |
| CN116863253B (en) | Operation and maintenance risk early warning method based on big data analysis | |
| CN115239672A (en) | Defect detection method and device, equipment and storage medium | |
| CN110175650A (en) | A kind of power equipment automatic identifying method and device | |
| CN117253192A (en) | Intelligent system and method for silkworm breeding | |
| CN117351271A (en) | Fault monitoring method and system for high-voltage distribution line monitoring equipment and storage medium thereof | |
| JP2024516642A (en) | Behavior detection method, electronic device and computer-readable storage medium | |
| CN114724041B (en) | Power equipment infrared chart identification method and system based on deep learning | |
| CN113642461A (en) | Elevator door opening and closing state identification method and device based on deep learning | |
| CN118397488A (en) | Unmanned aerial vehicle inspection small target detection method and system based on improved YOLOv model | |
| CN118429312A (en) | Method, system, equipment and medium for identifying and positioning chip and pin | |
| CN116704526B (en) | Gongchipu scanning robot and method thereof | |
| CN118537313A (en) | Intelligent detection method and detection system for electric power fitting surface | |
| CN117593193A (en) | Sheet metal image enhancement method and system based on machine learning | |
| CN117994573A (en) | Infrared dim target detection method based on superpixel and deformable convolution | |
| CN117671587A (en) | Power equipment defect detection method and system based on self-supervision learning | |
| CN116994049A (en) | Full-automatic flat knitting machine and method thereof | |
| CN113689399B (en) | Remote sensing image processing method and system for power grid identification | |
| CN116908184A (en) | A ground wire crimp detection system and its detection method | |
| CN113034465B (en) | A thermal fault monitoring method, device and medium for power equipment based on infrared images | |
| CN117790353B (en) | EL detection system and EL detection method | |
| CN117079416B (en) | Multi-person 5D radar falling detection method and system based on artificial intelligence algorithm |
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 | ||
| CB03 | Change of inventor or designer information | ||
| CB03 | Change of inventor or designer information |
Inventor after: Liang Chuan Inventor after: Shen Linxiang Inventor after: Li Hongyan Inventor after: Chen Lingzi Inventor before: Shen Linxiang Inventor before: Li Hongyan |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |