CN112116573A - High-precision infrared image anomaly detection method and system - Google Patents
High-precision infrared image anomaly detection method and system Download PDFInfo
- Publication number
- CN112116573A CN112116573A CN202010974406.5A CN202010974406A CN112116573A CN 112116573 A CN112116573 A CN 112116573A CN 202010974406 A CN202010974406 A CN 202010974406A CN 112116573 A CN112116573 A CN 112116573A
- Authority
- CN
- China
- Prior art keywords
- image
- training
- different
- power equipment
- neural network
- 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.)
- Granted
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 94
- 238000012549 training Methods 0.000 claims abstract description 88
- 238000003062 neural network model Methods 0.000 claims abstract description 45
- 230000002159 abnormal effect Effects 0.000 claims abstract description 29
- 238000000034 method Methods 0.000 claims abstract description 20
- 238000010438 heat treatment Methods 0.000 claims abstract description 17
- 230000000694 effects Effects 0.000 claims abstract description 10
- 238000007781 pre-processing Methods 0.000 claims abstract description 9
- 238000011156 evaluation Methods 0.000 claims description 13
- 238000006243 chemical reaction Methods 0.000 claims description 11
- 230000005856 abnormality Effects 0.000 claims description 9
- 238000004458 analytical method Methods 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 8
- 230000007613 environmental effect Effects 0.000 claims description 7
- 238000002372 labelling Methods 0.000 claims description 7
- 239000000463 material Substances 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 6
- 238000004422 calculation algorithm Methods 0.000 claims description 5
- 230000001629 suppression Effects 0.000 claims description 5
- 238000012546 transfer Methods 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000005070 sampling Methods 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims 2
- 238000013508 migration Methods 0.000 abstract description 5
- 230000005012 migration Effects 0.000 abstract description 5
- 230000005855 radiation Effects 0.000 description 5
- 238000012360 testing method Methods 0.000 description 4
- 238000012937 correction Methods 0.000 description 3
- 239000013598 vector Substances 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 230000001788 irregular Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000003990 capacitor Substances 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 239000012212 insulator Substances 0.000 description 1
- 230000003647 oxidation Effects 0.000 description 1
- 238000007254 oxidation reaction Methods 0.000 description 1
- 238000011897 real-time detection Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- 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
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
- G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
-
- 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
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- 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
- G06T2207/20112—Image segmentation details
- G06T2207/20132—Image cropping
-
- 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
- G06T2207/20172—Image enhancement details
- G06T2207/20182—Noise reduction or smoothing in the temporal domain; Spatio-temporal filtering
-
- 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)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- Quality & Reliability (AREA)
- Radiation Pyrometers (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to the technical field of detection, in particular to a high-precision infrared image abnormity detection method and a high-precision infrared image abnormity detection system, wherein the method comprises the following steps: s1, collecting a plurality of infrared abnormal heating image samples of the power equipment, expanding the image samples and constructing a data set; s2, constructing a network model, and training an SSD basic model by using the data set to obtain a power equipment detection neural network model; and S3, detecting the image to be detected by using the power equipment detection neural network model. According to the invention, the efficiency and high precision of infrared image or infrared video detection are greatly improved through more detailed sample classification; the image samples are expanded through the migration of the image styles, so that the training sample amount can be increased, and the influence of complex weather on the image samples can be solved; by preprocessing the image sample, the training complexity can be increased, the training effect is improved, and the accuracy of model detection is improved.
Description
Technical Field
The invention relates to the technical field of detection, in particular to a high-precision infrared image anomaly detection method and system.
Background
The power equipment is a basic composition module of the transformer substation, the state of the power equipment is detected in time, and the equipment can be efficiently managed, so that the transformer substation can operate more efficiently, safely and reliably. The abnormal heating of most of the power equipment in the transformer substation is caused by the abnormal heating of partial areas of the power equipment due to the reasons of poor contact of key parts, loose connecting parts, overload, oxidation of joints of old equipment, leakage current and the like. The infrared thermal imager has the advantages of non-contact uninterrupted power detection of power equipment and the like, is widely used in each power unit, and along with the application and popularization of intelligent equipment such as intelligent inspection and fixed-point monitoring of a transformer substation, the infrared image data is continuously increased, massive inspection infrared image data is retrieved and analyzed manually, and the time, labor and efficiency are low.
The traditional infrared detection method usually needs to manually set an infrared temperature threshold or manually observe to judge abnormal heating, is time-consuming and labor-consuming, and cannot perform continuous monitoring. The method for recognizing abnormal heating of the electrical equipment by applying the traditional image processing technology is often poor in recognition accuracy rate due to the fact that training samples are too few, sample classification is not detailed, and the influence of environmental factors such as regions, seasons, time periods, temperatures and weather on the electrical equipment is not considered, and is also a big problem due to the fact that the training speed is slow due to the fact that the training samples are too large in amount.
Content of application
In view of the defects of the prior art, the invention aims to provide a high-precision infrared image abnormality detection method for detecting and analyzing infrared image abnormality of electric equipment with high precision.
In order to achieve the above purpose, the embodiment of the invention is realized by the following technical scheme:
a high-precision infrared image anomaly detection method comprises the following steps:
s1, collecting a plurality of infrared abnormal heating image samples of the power equipment, expanding the image samples and constructing a data set;
s2, constructing a network model, and training an SSD basic model by using the data set to obtain a power equipment detection neural network model;
and S3, detecting the image to be detected by using the power equipment detection neural network model.
Further, the sample collection method of step S1 includes:
s11, collecting image samples of red heat abnormity and fault equipment of the power equipment under the conditions of different time intervals, different regions, different seasons, different weather, different temperatures and the like;
s12, converting the image sample into images of different domains by using an image style converter, realizing the transfer of the image style and expanding the image sample;
s13, each image sample labels different labels for different acquisition time periods, different environmental factors such as regions, different seasons, different weather and different temperatures, different red heat abnormal areas and different fault equipment images;
s14, carrying out batch denoising and stereo matching preprocessing on the labeled image samples, and carrying out one or more of rotation, mirror image, down sampling, blurring or enhancement processing, so as to increase the training complexity, improve the training effect and uniformly cut the image samples;
s15, a voc2007 standardized data set is created using the cut infrared abnormal heating image as a material for creating the data set.
Further, the method for constructing a model network in step S2 includes:
s21, loading a data set, and converting the data format of the data set to obtain a training file;
s22, constructing an SSD basic model, and training the SSD basic model by using the training file;
s23, obtaining a power equipment detection neural network model by adopting a RMSprop optimizer in training;
and S24, evaluating the power equipment detection neural network model and obtaining an index value according to the analysis and calculation of the training evaluation index.
Further, the neural network model evaluation method of step S24 includes:
taking the full-class average precision mAP as an evaluation index, and adjusting the parameters of the MobileNet basic network to retrain when the mAP value is less than 50%; and when the mAP value is larger than 50%, the power equipment detection neural network model is saved.
Further, the step S3 specifically includes the following steps:
s31, inputting an image to be detected into the power equipment detection neural network model;
s32, the power equipment detection neural network model randomly selects a plurality of areas of the image to be detected;
s33, scoring each area by using an SSD algorithm;
s34, setting a similarity threshold, comparing the score value of each region with the similarity threshold, judging the region with the score value higher than the similarity threshold as a positive example, and labeling the region;
judging the area with the score value lower than the similarity threshold value as a negative example;
s35, until all the areas judged as positive examples on the image to be detected are labeled;
s36, extracting the area determined as the positive example, and representing the area by a binary value graph;
and S37, searching all the contours in the binary image, and performing non-maximum suppression on the binary image to obtain a final binary image as a detection result.
Correspondingly, the invention also provides a high-precision infrared image anomaly detection system, which comprises:
the data set module is used for acquiring a plurality of infrared abnormal heating image samples of the power equipment, expanding the image samples and constructing a data set;
the data format conversion module is used for converting the data set into an xml format to obtain a training file;
a model building module for building an SSD base model;
the model training module is used for loading a data set and training the SSD basic model to obtain a power equipment detection neural network model;
and the detection analysis module is used for detecting the image to be detected through the power equipment detection neural network model.
Further, the specific process of the data set module for constructing the data set is as follows:
collecting image samples of red heat abnormity and fault equipment of the power equipment under the conditions of different time periods, different regions, different seasons, different weather, different temperatures and the like;
converting the image sample into images of different domains by using an image style converter, realizing the transfer of the image style and expanding the image sample;
each image sample labels different labels for different acquisition time periods, regions, seasons, weather, temperature and other environmental factors, different red heat abnormal areas and different fault equipment images;
carrying out batch denoising and stereo matching pretreatment on the marked image samples, and carrying out one or more of rotation, mirroring, downsampling, blurring or enhancement treatment, so that the training complexity is increased, the training effect is improved, and the image samples are uniformly cut;
and (4) taking the cut infrared abnormal heating image as a material for preparing a data set to prepare a voc2007 standardized data set.
Further, the specific process of the model training module training the electrical equipment detection neural network model is as follows:
loading a data set, and converting the data format of the data set to obtain a training file;
training the SSD base model by using the training file;
the optimizer in training adopts an RMSprop optimizer to obtain a power equipment detection neural network model;
and evaluating the power equipment detection neural network model and analyzing and calculating according to the training evaluation index to obtain an index value.
Further, the evaluation index is the full-class average precision mAP, and when the mAP value is less than 50%, the parameters of the MobileNet basic network are adjusted for retraining; and when the mAP value is larger than 50%, the power equipment detection neural network model is saved.
Further, the specific process of the detection analysis module obtaining the detection result is as follows:
inputting an image to be detected into the power equipment detection neural network model;
the power equipment detection neural network model randomly selects a plurality of areas of the image to be detected;
scoring each of said regions using an SSD algorithm;
setting a similarity threshold, comparing the score value of each region with the similarity threshold, judging the region with the score value higher than the similarity threshold as a positive example, and labeling the region;
judging the area with the score value lower than the similarity threshold value as a negative example;
until all the areas judged as positive examples on the image to be detected are marked;
extracting the regions determined as positive examples, and representing the regions by a binary image;
and searching all the contours in the binary image, and performing non-maximum suppression on the binary image to obtain a final binary image as a detection result.
The technical scheme of the embodiment of the invention at least has the following advantages and beneficial effects: the method and the system train the parameters of the network through a large amount of collected and expanded data, the training sample data volume is large, and the detection accuracy is high; meanwhile, the sample classification is more detailed, and the training speed can be improved when the sample classification participates in training.
Compared with the traditional infrared image anomaly detection method for the power equipment, the infrared image anomaly detection method and the infrared video anomaly detection system for the power equipment have the advantages that the efficiency and the high precision of infrared image or infrared video detection are greatly improved by carrying out more detailed sample classification on the collected infrared image or infrared video; the image samples are expanded through the migration of the image styles, so that the training sample amount can be increased, and the influence of complex weather on the image samples can be solved; by preprocessing the image sample, the training complexity can be increased, the training effect is improved, and the accuracy of model detection is improved.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present invention and therefore should not be considered as limiting the scope, and for those skilled in the art, other related drawings can be obtained according to the drawings without inventive efforts.
FIG. 1 is a flowchart of a method provided in example 1 of the present invention;
fig. 2 is a system block diagram provided in embodiment 2 of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. The components of embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations.
Thus, the following detailed description of the embodiments of the present invention, presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Example 1
The invention will be further described with reference to the accompanying drawings and the detailed description below:
referring to fig. 1, the method for detecting an abnormality in a high-precision infrared image according to the present embodiment includes the following steps:
and S1, collecting a plurality of infrared abnormal heating image samples of the power equipment, expanding the image samples and constructing a data set.
S11, collecting image samples of red heat abnormity and fault equipment of the power equipment under the conditions of different time intervals, different regions, different seasons, different weather, different temperatures and the like; the method comprises the steps that infrared high-definition cameras are arranged in power equipment areas in different regions, image samples of power equipment under different seasons, different weather and different temperature conditions and in multiple time periods are collected, and the influences of factors such as the regions, the seasons, the weather, the temperatures, the time periods and the like on the collected image samples of the infrared high-definition cameras and the power equipment are fully considered, for example, the power consumption in the south is large, the power consumption in winter and summer is large, the power consumption in the daytime is large, the power consumption in high and low temperatures are different, and the weather such as rain, snow, haze, wind power, thunder, frosting and the like can all cause remarkable influences on the power equipment, so that the classification of the samples is more detailed; besides, the electric power equipment images under the same scene can be collected on the public data set and the material website through a network approach, and the sample size is increased.
S12, converting the image sample into images of different domains by using an image style converter, realizing the transfer of the image style and expanding the image sample;
aiming at the power equipment in special scenes (such as rain, snow, haze, wind, thunder, frost and the like), the fault occurrence rate is higher than that in ordinary scenes, so that the training sample amount needs to be increased to achieve higher detection precision,
external factors affecting the infrared device's acquisition of the infrared radiation of the subject matter include the attenuation of the infrared radiation of the subject matter by the atmosphere, for example, in a rainy day scene, the air humidity thereof increases; the air humidity in winter or snow days is relatively reduced; lightning radiation may have an effect on the infrared device; wind power changes the air flow rate to cause infrared radiation refraction of the target object; haze affects infrared image precision and the like, and the atmospheric coefficients all affect infrared radiation of a collection target collected by infrared equipment; namely, the infrared image collected in normal weather and the infrared image collected in special scene have obvious difference;
on the premise that the image sample is acquired in step S11, the expansion of the special scene image sample can be realized by image style migration,
specifically, a training basic model is established, a style conversion training model is obtained after training, an image is input to the model to extract a feature region in the image, the image is compressed into a plurality of feature vectors with the same size, the dissimilar features of the image are combined, and the feature vectors of the image in a DA domain are converted into the feature vectors in a DB domain. The method can expand image samples through image style migration, and can also solve the influence of complex weather on the precision of the image samples, for example, the image collected in the haze weather is often blurred due to the influence of haze, so that the influence of the haze weather on the image can be reduced, but the influence of different atmospheric coefficients received by different images cannot be solved.
Aiming at the problem that the influence of the atmospheric coefficient cannot be solved in the steps, a style conversion training basic model is established, the attenuation rate of the atmosphere to the heat value of the comparison image is calculated according to the atmospheric information (rain, snow, haze, wind power, thunder and frost) and the distance information, wherein x is a distance, y is an atmospheric coefficient, the air coefficient is an integer in a range of 1-100 given by a meteorological station system, and 100 is the optimal air quality, and the atmospheric attenuation coefficient is as follows: k ═ f (x, y);
inputting an image sample to be converted, and carrying out characteristic region labeling on an image heat value of the image sample;
extracting the atmospheric information of the image sample to be converted and the distance between the atmospheric information and a target object, calculating the atmospheric attenuation coefficient of the image to be converted according to the calculation formula, obtaining the difference value of the atmospheric attenuation coefficients between the two images, performing compensation operation on the heat value characteristic area of the image sample to be converted according to the difference value to obtain a converted image with the same atmospheric attenuation coefficient as the comparison image, and correspondingly replacing labels such as the atmospheric coefficient corresponding to the image; and repeating the steps to obtain the trained style conversion training model.
Secondly, in the embodiment, aiming at the condition that the image style possibly does not meet the sample standard after being converted, the image is used for data training, so that the aim of estimating and expanding the image training sample amount to improve the detection precision cannot be fulfilled, and the detection precision is easy to reduce; therefore, an image checking process needs to be added after the image expansion,
the method comprises the following specific steps:
firstly, obtaining a converted image by using a trained style conversion training model, and obtaining the converted image, an atmospheric scene corresponding to the converted image, an original image, an image of the same atmospheric scene and an image of the same equipment type;
the converted image is inversely operated according to the atmospheric attenuation rate, the result of the inverse operation is compared with the original image, if the result of the inverse operation is different from the original image, the converted image is withdrawn, if the result of the inverse operation is the same as the original image, the checking result is scored, and the next checking step is executed,
comparing the converted image with the image of the same atmospheric scene, removing the characteristic region in the converted image and the image of the same atmospheric scene, reserving the non-characteristic region, comparing the non-characteristic regions of the converted image and the image of the same atmospheric scene, withdrawing the converted image if the similarity is too low, scoring the check result if the similarity is higher, and executing the next check step,
correcting and comparing the converted image with infrared images of the same type of equipment, extracting characteristic regions of the converted image and the infrared images of the same type of equipment, eliminating non-characteristic regions, converting the characteristic regions of the converted image and the infrared images of the same type of equipment into binary images, if comparing 1 infrared image of the same type of equipment, calculating the similarity, scoring the similarity, if the similarity is too low, withdrawing the converted image,
in order to ensure the checking accuracy, the converted image is compared with a plurality of images at the same time, and the average similarity of the converted image is calculated so as to avoid the error withdrawal;
and (4) descending the sequence of the checked and scored converted images according to the similarity scores of the converted images, extracting the converted images according to the proportion, performing withdrawal on the rest of the converted images, and converting the rest of the converted images into the rest of the atmosphere scene styles, so that the precision is lower after some image samples are converted into a certain atmosphere scene, and if the conversion of the original atmosphere scene style is still performed after the withdrawal, the converted images are always not in accordance with the sample standard.
The withdrawing in the checking step is specifically as follows:
the method for converting the image comprises the steps of converting the image into a converted image, and performing conversion on the converted image to obtain a converted image, wherein the converted image is restored to a state before the style conversion of the style conversion training model, and the converted image is required to be manufactured by the aid of a script which is not necessarily executed.
S13, each image sample labels different labels for different acquisition time periods, different environmental factors such as regions, different seasons, different weather and temperature, different red-hot abnormal areas and different fault equipment images; the infrared image is influenced by different atmospheric scenes, the difference of power consumption is obvious in different time periods, the load states of power equipment are different, and the proportion of industrial power consumption and household power consumption is changed; different regions, such as northwest, have thin air and low atmospheric attenuation rate; the common red heat abnormal regions include dots, short strips, short discs, lumps, strips, discs and large-area irregular shapes. Optionally, the point, short strip and short disc types are classified into 1 type, which is a general emergency type, further, the general emergency type can be further classified into three types according to the situation, the bulk, long strip and disc types are classified into 2 types, which are emergency types, and further, the emergency type can be further classified into three types according to the situation; large area irregular shapes locate class 3, which is a special anomaly class. Through classifying and grading the red heat abnormal area, the sample classification is more detailed, and the accuracy rate of the model for positioning the abnormal area can be improved.
The common electronic devices which are easy to be abnormal are respectively: the transformer, insulators, current transformers, voltage transformers, bolts, coupling capacitors, lightning arresters, circuit breakers, isolating switches, reactors, wave traps, sleeves, lead cables and the like. Through classifying above-mentioned electronic equipment, can improve the rate of accuracy of the unusual equipment of model identification, its operating condition of above-mentioned electronic equipment of different periods is different wherein, for example, arrester and the like under the thunder and lightning weather, and the period is industrial electricity utilization ratio decline when evening, and family's power consumption ratio rises etc. all can cause the influence to electronic equipment to influence the unusual classification of infrared image.
Each image sample labels target frames with different labels for the corresponding abnormal area type, the abnormal equipment type and the environmental factors during acquisition, wherein each target frame contains an abnormal heating area of a single known abnormal equipment infrared image.
S14, carrying out batch denoising and stereo matching preprocessing on the marked image samples, and carrying out one or more of rotation, mirror image, down sampling, blurring or enhancement processing, so as to increase the training complexity, improve the training effect, and uniformly cut the image samples, wherein optionally, the image samples are uniformly cut to 480 × 480 px; the method comprises the steps that part of collected image sources are collected at different periods, the image sources are often unclear and have more noise points, the establishment of a unified typical fault power equipment infrared image data set with better quality is not facilitated, further preprocessing needs to be carried out on original image samples, and the preprocessing comprises one or more of batch denoising, stereo matching preprocessing, rotation or correction, mirroring, downsampling, blurring or enhancement processing, wherein the blurring processing in the rotation or correction processing is used for increasing the training complexity, the correction and enhancement processing can be lower in training complexity degree, the relative training speed is higher, and the training effect can be improved to the extent by increasing the training complexity, so that the accuracy of the training model detection images can be improved.
Further, after the image sample is processed, image verification is performed, and verification contents include but are not limited to image integrity and corresponding class labels.
S15, a voc2007 standardized data set is created using the cut infrared abnormal heating image as a material for creating the data set.
Further, images in the data set are randomly divided into training sets and testing sets in proportion, and each training set and each testing set comprises one image and a corresponding label.
S2, constructing a network model, and training an SSD basic model by using the data set to obtain a power equipment detection neural network model; optionally, the SSD basic model adopts a lightweight MobileNets structure instead of the original VGG basic network, thereby further improving the detection speed of the model.
S21, loading a data set, and converting the data format of the data set, optionally converting the data format into an xml format, to obtain a training file;
s22, constructing an SSD basic model, and performing model pre-training by using the divided training sample set;
and resizing the dimensions of the trained data set to 224 × 3;
carrying out model training by utilizing the training set with well-divided dimensionality,
further, a contrast experiment model is built, and the same operation is carried out on the training of the contrast experiment to obtain a pre-training model;
and inputting the model test into the test set to obtain a final model.
S23, obtaining a power equipment detection neural network model by adopting a RMSprop optimizer in training; optionally, stopping model training when the loss value is smaller than a certain threshold value after network iteration, and storing the power equipment detection neural network model;
and analyzing the experimental result aiming at the model prediction result.
And S24, evaluating the power equipment detection neural network model and obtaining an index value according to the analysis and calculation of the training evaluation index.
The neural network model evaluation method of step S24 includes:
taking the full-class average precision mAP as an evaluation index, and adjusting the parameters of the MobileNet basic network to retrain when the mAP value is less than 50%; and when the mAP value is larger than 50%, storing the result to obtain a final weight and a trained power equipment detection neural network model.
And S3, detecting the image to be detected by using the power equipment detection neural network model.
S31, inputting an image to be detected into the power equipment detection neural network model;
s32, the power equipment detection neural network model randomly selects a plurality of areas of the image to be detected;
s33, scoring each area by using an SSD algorithm;
s34, setting a similarity threshold, comparing the score value of each region with the similarity threshold, judging the region with the score value higher than the similarity threshold as a positive example, and labeling the region;
judging the area with the score value lower than the similarity threshold value as a negative example;
when the SSD basic model is used for image detection, a real-time detection real-time judgment and labeling mode is adopted, and the detection precision is improved.
S35, until all the areas judged as positive examples on the image to be detected are labeled;
s36, extracting the area determined as the positive example, and representing the area by a binary value graph;
and S37, searching all the contours in the binary image, and performing non-maximum suppression on the binary image to obtain a final binary image as a detection result.
Example 2
Referring to fig. 2, the high-precision infrared image anomaly detection system provided in this embodiment includes: the data set module is used for acquiring a plurality of infrared abnormal heating image samples of the power equipment, expanding the image samples and constructing a data set;
the data format conversion module is used for converting the data set into an xml format to obtain a training file;
a model building module for building an SSD base model;
the model training module is used for loading a data set and training the SSD basic model to obtain a power equipment detection neural network model;
and the detection analysis module is used for detecting the image to be detected through the power equipment detection neural network model.
Since the working principle of each module is the same as the flow principle of the method, it is not described in detail in this embodiment.
Therefore, the efficiency and the high precision of infrared image or infrared video detection are greatly improved by carrying out more detailed sample classification on the collected infrared image or infrared video; the image samples are expanded through the migration of the image styles, so that the training sample amount can be increased, and the influence of complex weather on the image samples can be solved; by preprocessing the image sample, the training complexity can be increased, the training effect is improved, and the accuracy of model detection is improved.
The above is only a preferred embodiment of the present invention, and is not intended to limit the present invention, and various modifications and changes will occur to those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (10)
1. A high-precision infrared image anomaly detection method is characterized by comprising the following steps:
s1, collecting a plurality of infrared abnormal heating image samples of the power equipment, expanding the image samples and constructing a data set;
s2, constructing a network model, and training an SSD basic model by using the data set to obtain a power equipment detection neural network model;
and S3, detecting the image to be detected by using the power equipment detection neural network model.
2. The infrared image abnormality detection method according to claim 1, characterized in that the sample collection method of step S1 includes:
s11, collecting image samples of red heat abnormity and fault equipment of the power equipment under the conditions of different time intervals, different regions, different seasons, different weather, different temperatures and the like;
s12, converting the image sample into images of different domains by using an image style converter, realizing the transfer of the image style and expanding the image sample;
s13, each image sample labels different labels for different acquisition time periods, different environmental factors such as regions, different seasons, different weather and different temperatures, different red heat abnormal areas and different fault equipment images;
s14, carrying out batch denoising and stereo matching preprocessing on the labeled image samples, and carrying out one or more of rotation, mirror image, down sampling, blurring or enhancement processing, so as to increase the training complexity, improve the training effect and uniformly cut the image samples;
s15, a voc2007 standardized data set is created using the cut infrared abnormal heating image as a material for creating the data set.
3. The infrared image abnormality detection method according to claim 2, characterized in that the model network construction method of step S2 includes:
s21, loading a data set, and converting the data format of the data set to obtain a training file;
s22, constructing an SSD basic model, and training the SSD basic model by using the training file;
s23, obtaining a power equipment detection neural network model by adopting a RMSprop optimizer in training;
and S24, evaluating the power equipment detection neural network model and obtaining an index value according to the analysis and calculation of the training evaluation index.
4. The infrared image abnormality detection method according to claim 3, characterized in that the neural network model evaluation method of step S24 includes:
taking the full-class average precision mAP as an evaluation index, and adjusting the parameters of the MobileNet basic network to retrain when the mAP value is less than 50%; and when the mAP value is larger than 50%, the power equipment detection neural network model is saved.
5. The infrared image abnormality detection method according to claim 3, characterized in that step S3 specifically includes the steps of:
s31, inputting an image to be detected into the power equipment detection neural network model;
s32, the power equipment detection neural network model randomly selects a plurality of areas of the image to be detected;
s33, scoring each area by using an SSD algorithm;
s34, setting a similarity threshold, comparing the score value of each region with the similarity threshold, judging the region with the score value higher than the similarity threshold as a positive example, and labeling the region;
judging the area with the score value lower than the similarity threshold value as a negative example;
s35, until all the areas judged as positive examples on the image to be detected are labeled;
s36, extracting the area determined as the positive example, and representing the area by a binary value graph;
and S37, searching all the contours in the binary image, and performing non-maximum suppression on the binary image to obtain a final binary image as a detection result.
6. An infrared image abnormality detection system of high precision, characterized by comprising:
the data set module is used for acquiring a plurality of infrared abnormal heating image samples of the power equipment, expanding the image samples and constructing a data set;
the data format conversion module is used for converting the data set into an xml format to obtain a training file;
a model building module for building an SSD base model;
the model training module is used for loading a data set and training the SSD basic model to obtain a power equipment detection neural network model;
and the detection analysis module is used for detecting the image to be detected through the power equipment detection neural network model.
7. The infrared image anomaly detection system according to claim 6, wherein said dataset module constructs a dataset by:
collecting image samples of red heat abnormity and fault equipment of the power equipment under the conditions of different time periods, different regions, different seasons, different weather, different temperatures and the like;
converting the image sample into images of different domains by using an image style converter, realizing the transfer of the image style and expanding the image sample;
each image sample labels different labels for different acquisition time periods, regions, seasons, weather, temperature and other environmental factors, different red heat abnormal areas and different fault equipment images;
carrying out batch denoising and stereo matching pretreatment on the marked image samples, and carrying out one or more of rotation, mirroring, downsampling, blurring or enhancement treatment, so that the training complexity is increased, the training effect is improved, and the image samples are uniformly cut;
and (4) taking the cut infrared abnormal heating image as a material for preparing a data set to prepare a voc2007 standardized data set.
8. The infrared image anomaly detection system according to claim 7, wherein the specific process of the model construction module for constructing the electrical equipment detection neural network model is as follows:
loading a data set, and converting the data format of the data set to obtain a training file;
constructing an SSD basic model, and training the SSD basic model by using the training file;
the optimizer in training adopts an RMSprop optimizer to obtain a power equipment detection neural network model;
and evaluating the power equipment detection neural network model and analyzing and calculating according to the training evaluation index to obtain an index value.
9. The infrared image abnormality detection system according to claim 8,
the evaluation index is the full-class average precision mAP, and when the mAP value is less than 50%, the basic network parameters of the mobileNet are adjusted for retraining; and when the mAP value is larger than 50%, the power equipment detection neural network model is saved.
10. The infrared image anomaly detection system according to claim 9, wherein the specific process of obtaining the detection result by the detection analysis module is as follows:
inputting an image to be detected into the power equipment detection neural network model;
the power equipment detection neural network model randomly selects a plurality of areas of the image to be detected;
scoring each of said regions using an SSD algorithm;
setting a similarity threshold, comparing the score value of each region with the similarity threshold, judging the region with the score value higher than the similarity threshold as a positive example, and labeling the region;
judging the area with the score value lower than the similarity threshold value as a negative example;
until all the areas judged as positive examples on the image to be detected are marked;
extracting the regions determined as positive examples, and representing the regions by a binary image;
and searching all the contours in the binary image, and performing non-maximum suppression on the binary image to obtain a final binary image as a detection result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010974406.5A CN112116573B (en) | 2020-09-16 | 2020-09-16 | High-precision infrared image anomaly detection method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010974406.5A CN112116573B (en) | 2020-09-16 | 2020-09-16 | High-precision infrared image anomaly detection method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112116573A true CN112116573A (en) | 2020-12-22 |
| CN112116573B CN112116573B (en) | 2024-08-16 |
Family
ID=73803085
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010974406.5A Active CN112116573B (en) | 2020-09-16 | 2020-09-16 | High-precision infrared image anomaly detection method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112116573B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112672052A (en) * | 2020-12-24 | 2021-04-16 | 华中光电技术研究所（中国船舶重工集团公司第七一七研究所） | Image data enhancement method and system, electronic equipment and storage medium |
| CN112991200A (en) * | 2021-02-08 | 2021-06-18 | 中国科学院西安光学精密机械研究所 | Method and device for adaptively enhancing infrared image |
| CN113052865A (en) * | 2021-04-16 | 2021-06-29 | 南通大学 | Power transmission line small sample temperature image amplification method based on image similarity |
| CN114199381A (en) * | 2021-12-13 | 2022-03-18 | 国网陕西省电力公司延安供电公司 | Electrical equipment fault detection method for improving infrared detection model |
| CN117309195A (en) * | 2023-11-30 | 2023-12-29 | 中控技术股份有限公司 | Detection method and device of temperature measuring instrument and electronic equipment |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108595558A (en) * | 2018-04-12 | 2018-09-28 | 福建工程学院 | A kind of image labeling method of data balancing strategy and multiple features fusion |
| 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 |
| CN109816695A (en) * | 2019-01-31 | 2019-05-28 | 中国人民解放军国防科技大学 | Target detection and tracking method for infrared small unmanned aerial vehicle under complex background |
| CN110706197A (en) * | 2019-08-06 | 2020-01-17 | 北方工业大学 | Railway foreign matter intrusion detection method based on transfer learning in special scene |
| CN110782461A (en) * | 2019-10-28 | 2020-02-11 | 华北电力大学 | Mask-RCNN-based infrared image segmentation method for electric power equipment |
| CN111091101A (en) * | 2019-12-23 | 2020-05-01 | 中国科学院自动化研究所 | High-precision pedestrian detection method, system and device based on one-step method |
| CN111179249A (en) * | 2019-12-30 | 2020-05-19 | 南京南瑞信息通信科技有限公司 | Power equipment detection method and device based on deep convolutional neural network |
-
2020
- 2020-09-16 CN CN202010974406.5A patent/CN112116573B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108595558A (en) * | 2018-04-12 | 2018-09-28 | 福建工程学院 | A kind of image labeling method of data balancing strategy and multiple features fusion |
| 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 |
| CN109816695A (en) * | 2019-01-31 | 2019-05-28 | 中国人民解放军国防科技大学 | Target detection and tracking method for infrared small unmanned aerial vehicle under complex background |
| CN110706197A (en) * | 2019-08-06 | 2020-01-17 | 北方工业大学 | Railway foreign matter intrusion detection method based on transfer learning in special scene |
| CN110782461A (en) * | 2019-10-28 | 2020-02-11 | 华北电力大学 | Mask-RCNN-based infrared image segmentation method for electric power equipment |
| CN111091101A (en) * | 2019-12-23 | 2020-05-01 | 中国科学院自动化研究所 | High-precision pedestrian detection method, system and device based on one-step method |
| CN111179249A (en) * | 2019-12-30 | 2020-05-19 | 南京南瑞信息通信科技有限公司 | Power equipment detection method and device based on deep convolutional neural network |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112672052A (en) * | 2020-12-24 | 2021-04-16 | 华中光电技术研究所（中国船舶重工集团公司第七一七研究所） | Image data enhancement method and system, electronic equipment and storage medium |
| CN112991200A (en) * | 2021-02-08 | 2021-06-18 | 中国科学院西安光学精密机械研究所 | Method and device for adaptively enhancing infrared image |
| CN112991200B (en) * | 2021-02-08 | 2024-02-02 | 中国科学院西安光学精密机械研究所 | Method and device for adaptively enhancing infrared image |
| CN113052865A (en) * | 2021-04-16 | 2021-06-29 | 南通大学 | Power transmission line small sample temperature image amplification method based on image similarity |
| CN113052865B (en) * | 2021-04-16 | 2023-12-19 | 南通大学 | Power transmission line small sample temperature image amplification method based on image similarity |
| CN114199381A (en) * | 2021-12-13 | 2022-03-18 | 国网陕西省电力公司延安供电公司 | Electrical equipment fault detection method for improving infrared detection model |
| CN117309195A (en) * | 2023-11-30 | 2023-12-29 | 中控技术股份有限公司 | Detection method and device of temperature measuring instrument and electronic equipment |
| CN117309195B (en) * | 2023-11-30 | 2024-04-12 | 中控技术股份有限公司 | Detection method and device of temperature measuring instrument and electronic equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112116573B (en) | 2024-08-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112116573A (en) | High-precision infrared image anomaly detection method and system | |
| CN108647716B (en) | Photovoltaic array fault diagnosis method based on composite information | |
| CN112379231B (en) | Equipment detection method and device based on multispectral image | |
| CN112381784A (en) | Equipment detecting system based on multispectral image | |
| CN111784692B (en) | Method and device for detecting defects of insulators in power system and electronic equipment | |
| CN111798412A (en) | Intelligent diagnosis method and system for substation equipment defects based on infrared images | |
| CN107506768A (en) | A kind of stranded recognition methods of transmission line wire based on full convolutional neural networks | |
| CN113011763B (en) | Bridge damage identification method based on space-time diagram convolution attention | |
| CN114973032B (en) | Deep convolutional neural network-based photovoltaic panel hot spot detection method and device | |
| CN103063166A (en) | Detection method and device for wind deflection angle of suspension type composite insulator chain | |
| CN109617526A (en) | A method of photovoltaic power generation array fault diagnosis and classification based on wavelet multiresolution analysis and SVM | |
| CN115588021A (en) | Ceramic insulator and hardware unmanned aerial vehicle infrared inspection image fusion and segmentation method | |
| CN115049607B (en) | Insulation board defect identification method based on YOLOx _s enhanced target feature detection | |
| CN111161236A (en) | Method, device, medium and equipment for detecting hydrophobicity of composite insulator | |
| CN114694130A (en) | Method and device for detecting telegraph poles and pole numbers along railway based on deep learning | |
| CN116485802B (en) | Insulator flashover defect detection method, device, equipment and storage medium | |
| CN106355187A (en) | Application of visual information to electrical equipment monitoring | |
| CN112683916A (en) | Method and device for identifying missing or mounting error of small hardware fittings of current collecting line tower | |
| CN110555460B (en) | Image slice-based bird detection method for power transmission line at mobile terminal | |
| CN111767777B (en) | Transformer substation disconnecting switch state analysis method based on image processing algorithm | |
| CN115546191A (en) | Insulator defect detection method and device based on improved RetinaNet | |
| CN113657310A (en) | Photovoltaic module infrared hot spot fault identification method based on multilayer convolutional neural network | |
| CN112884720A (en) | Distribution line pollution flashover insulator detection method and system | |
| CN113033489A (en) | Power transmission line insulator identification and positioning method based on lightweight deep learning algorithm | |
| Chen et al. | Development of Management System for Insulator UV Imaging Detection |
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 |