CN112465797A - Intelligent diagnosis method and system for thermal state of cable terminal - Google Patents

Intelligent diagnosis method and system for thermal state of cable terminal Download PDF

Info

Publication number
CN112465797A
CN112465797A CN202011437678.8A CN202011437678A CN112465797A CN 112465797 A CN112465797 A CN 112465797A CN 202011437678 A CN202011437678 A CN 202011437678A CN 112465797 A CN112465797 A CN 112465797A
Authority
CN
China
Prior art keywords
diagnosis
target
cable terminal
candidate
candidate frame
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.)
Pending
Application number
CN202011437678.8A
Other languages
Chinese (zh)
Inventor
杨斌
徐小冰
郭浩然
周承科
周文俊
李�根
谢诚
艾永恒
孙长群
毛永彬
陈阳
李福明
高鸣
高新昀
严一涛
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Wuhan University WHU
Wuhan Power Supply Co of State Grid Hubei Electric Power Co Ltd
Original Assignee
Wuhan University WHU
Wuhan Power Supply Co of State Grid Hubei Electric Power Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Wuhan University WHU, Wuhan Power Supply Co of State Grid Hubei Electric Power Co Ltd filed Critical Wuhan University WHU
Priority to CN202011437678.8A priority Critical patent/CN112465797A/en
Publication of CN112465797A publication Critical patent/CN112465797A/en
Pending legal-status Critical Current

Links

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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/047—Probabilistic or stochastic networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/70—Determining position or orientation of objects or cameras
    • G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
    • 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
    • G06T2207/20076—Probabilistic image processing
    • 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
    • G06T2207/20084—Artificial neural networks [ANN]
    • 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
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30108—Industrial image inspection

Abstract

The invention relates to a cable terminal thermal state intelligent diagnosis method and a system, wherein the method comprises the steps of marking infrared image information of a cable terminal shot in the inspection process, and constructing a training sample; constructing a fast RCNN network, and training the fast RCNN network by adopting the training sample; inputting infrared image information of a cable terminal to be detected into the trained fast RCNN for identification and positioning to obtain the category and the position of a target diagnosis object; and carrying out thermal state diagnosis on the target diagnosis object after the identification and positioning processing according to a preset diagnosis standard. According to the invention, the fast RCNN is constructed, and the trained fast RCNN is utilized to automatically identify and position the infrared image information of the cable terminal to be detected, so that the thermal diagnosis is completed, the manual operation part in the current infrared diagnosis can be effectively replaced, the dependence on manual processing and analysis is reduced, the diagnosis efficiency is greatly improved, and the diagnosis accuracy is improved.

Description

Intelligent diagnosis method and system for thermal state of cable terminal
Technical Field
The invention relates to the technical field of cable monitoring and diagnosis, in particular to a cable terminal thermal state intelligent diagnosis method and system.
Background
The power cable has good electrical performance and mechanical performance, has the advantages of environmental friendliness and the like, and is widely applied to power systems at present. Cable termination refers to a device for connecting a cable to other electrical equipment. Due to the influence of production and manufacturing processes, field installation and the like, the cable terminal becomes a weak link of a cable system. Whether the cable terminal is in a normal operation state affects the stability of the cable system.
Theoretical analysis and actual detection results show that: the defective cable termination temperature is typically higher than the cable termination in the normal state. Therefore, measuring the temperature of the cable termination is an effective means of diagnosing its condition. At present, the infrared thermal imaging technology has the advantages of non-contact, intuition, no influence of high-voltage electromagnetic field, high efficiency and the like, and is widely applied to temperature measurement of cable inspection.
However, in the infrared inspection of the cable system, the inspection personnel still mainly analyze the infrared image and diagnose the state of the photographed cable terminal manually. The state diagnosis of the cable terminal under the infrared background is not intelligentized.
At present, the domestic infrared diagnosis process for the cable system comprises the following steps: firstly, regularly shooting infrared images of a cable terminal by a patrol worker according to a patrol cycle; uploading the image of the thermal infrared imager after the thermal infrared imager returns to the team, and then judging whether an abnormal heating area exists in the image according to self experience; if an abnormal heating area exists, manually selecting the area as a superheat area, and simultaneously selecting a part at the same position as the area in other phases as a reference area; and finally, realizing the state diagnosis of the cable terminal based on corresponding diagnosis standards according to the temperature information of the overheating area and the reference area. On one hand, the diagnosis mode has the disadvantages that the number of the infrared images of the cable terminal is huge, the infrared images are processed by manpower, and the analysis efficiency is low; on the other hand, the excessive dependence on manual experience may cause missed judgment or erroneous judgment.
Disclosure of Invention
The invention aims to solve the technical problem of providing a cable terminal thermal state intelligent diagnosis method and system aiming at the defects of the prior art.
The technical scheme for solving the technical problems is as follows: an intelligent diagnosis method for the thermal state of a cable terminal comprises the following steps:
s1: marking infrared image information of the cable terminal shot in the routing inspection process, and constructing a training sample;
s2: constructing a fast RCNN network, and training the fast RCNN network by adopting the training sample;
s3: inputting target infrared image information of a cable terminal to be detected into the trained fast RCNN for identification and positioning to obtain the category and the position of a target diagnosis object;
s4: and carrying out thermal state diagnosis on the target diagnosis object subjected to the identification and positioning treatment according to a preset diagnosis standard to obtain a thermal state diagnosis result of the cable terminal to be detected.
The invention has the beneficial effects that: according to the intelligent diagnosis method for the thermal state of the cable terminal, the fast RCNN is constructed, the training sample constructed by the acquired infrared image information of the cable terminal is used for training, the trained fast RCNN is used for automatically identifying and positioning the target infrared image information of the cable terminal to be detected, thermal diagnosis is completed, the part of manual operation in the current infrared diagnosis can be effectively replaced, the dependence on manual processing and analysis is reduced, the diagnosis efficiency is greatly improved, the diagnosis accuracy is improved, and the deviation caused by the manual diagnosis is effectively avoided.
On the basis of the technical scheme, the invention can be further improved as follows:
further: in step S1, the infrared image information of the cable terminal shot in the annotation inspection process specifically includes:
acquiring infrared image information of a cable terminal, marking a marking frame at the position of a diagnostic object in the infrared image information, and inputting the name of the marking frame;
wherein the diagnostic object is contained within the label box.
The beneficial effects of the further scheme are as follows: by marking the acquired infrared image information of the cable terminal, the subsequent Faster RCNN network can be conveniently trained according to the marking frame as a reference, so that the trained Faster RCNN network has the automatic identification and positioning functions of the diagnostic object in the infrared image of the cable terminal.
Further: in the step S2, the training of the Faster RCNN network using the training samples specifically includes the following steps:
s21: initializing the Faster RCNN, inputting the training sample into the convolutional neural network CNN, extracting one or more image characteristics of the infrared image information of the cable terminal by using the convolutional neural network CNN, and forming a characteristic diagram;
s22: automatically generating a plurality of candidate frames containing the diagnostic object on the feature map by using a regional suggestion network (RPN), adjusting the positions and the sizes of the candidate frames according to the deviation between the labeling frame and the candidate frames, mapping the adjusted candidate frames to the feature map of the Convolutional Neural Network (CNN) to obtain corresponding candidate features, and determining the position of the diagnostic object;
s23: dividing the candidate features into a plurality of regions by using a region-of-interest pooling layer RoI, and performing maximum pooling on each region;
s24: and inputting the candidate features into a target detection layer, on one hand, calculating the probability that the candidate frame corresponding to the candidate features contains the diagnostic objects belonging to different categories through a classification layer to determine the categories of the diagnostic objects, and on the other hand, calculating the adjustment parameters of the candidate frame again through a regression layer to perform secondary adjustment on the candidate frame.
The beneficial effects of the further scheme are as follows: on one hand, image feature information of infrared image information of a cable terminal is extracted through a convolutional neural network CNN and forms a feature map, a candidate frame is automatically generated on the feature map through a region suggestion network RPN, the position and the size of the candidate frame are adjusted, the position of a diagnostic object can be preliminarily determined, on the other hand, block pooling processing is carried out through a region-of-interest pooling layer RoI, the probability that the candidate frame corresponding to the candidate feature contains different types of diagnostic objects is calculated through a target detection layer, the type corresponding to diagnosis can be accurately determined, and meanwhile, the regression layer calculates the adjustment parameters of the candidate frame again, so that the position of the diagnostic object is accurately determined, and training of a fast RCNN is completed.
Further: in step S22, the adjusting the position and size of the candidate frame according to the deviation between the labeling frame and the candidate frame specifically includes the following steps:
s221: generating adjustment parameter information for adjusting the candidate frame according to the deviation between the labeling frame and the candidate frame;
s222: adjusting the position and the size of the candidate frame according to the adjustment parameter information;
wherein the adjustment parameter information includes a translation parameter and a scaling parameter.
The beneficial effects of the further scheme are as follows: and generating adjustment parameter information through the deviation between the marking frame and the candidate frame, so that the position and the size of the candidate frame can be accurately adjusted, and the positioning accuracy of the whole fast RCNN network is improved conveniently.
Further: in step S221, the generating of adjustment parameter information for adjusting the candidate frame according to the deviation between the labeling frame and the candidate frame specifically includes the following steps:
s2211: the classification layer of the regional suggestion network calculates the probability that the candidate frame contains the diagnostic object by using a Softmax classifier, and preliminarily screens out the candidate frame containing the diagnostic object according to the probability;
s2212: constructing a loss function of the regional proposed network, and calculating the adjustment parameter information of the candidate frame by taking the minimum loss function as a target;
Figure BDA0002829016150000041
wherein:
Figure BDA0002829016150000042
Figure BDA0002829016150000043
Figure BDA0002829016150000044
in the formula: i represents the ith candidate box; p is a radical ofiRepresenting the probability that the candidate frame calculated by the classification layer contains the diagnosis object; t is tiRepresenting the prediction offset obtained by the regression layer, wherein the prediction offset comprises a translation parameter and a scaling parameter;
Figure BDA0002829016150000051
the value is determined by the intersection ratio IoU (intersection over Union) between the candidate frame and the labeling frame predicted by the area suggestion network, and is specifically shown as a formula (9);
Figure BDA0002829016150000052
the actual offset is shown as a formula (11); n is a radical ofcls、NregAnd λ are balance parameters;
Figure BDA0002829016150000053
wherein:
Figure BDA0002829016150000054
in the formula: anchor represents a candidate box generated by the regional suggestion network; the GrountTruth represents a marking frame;
Figure BDA0002829016150000055
in the formula: (A)x,Ay,Aw,Ah) Representing the coordinates of the candidate frame before adjustment; (G)x,Gy,Gw,Gh) Representing the coordinates of the label box.
The beneficial effects of the further scheme are as follows: by constructing the loss function of the regional suggestion network and taking the minimum loss function as a target, the adjustment parameters for adjusting the candidate frame are determined, so that the deviation between the adjusted candidate frame and the marking frame is minimum, the adjusted candidate frame can be ensured to accurately cover the diagnostic object, the training effect is ensured, and the positioning accuracy of the trained Faster RCNN network is improved.
Further: the specific calculation formula for adjusting the position and size of the candidate frame according to the adjustment parameter information is as follows:
Figure BDA0002829016150000056
Figure BDA0002829016150000061
wherein (A)x,Ay,Aw,Ah) Representing the coordinates of the candidate frame before adjustment; (G)x’,Gy’,G’w,Gh') indicates the adjusted candidate frame coordinates, dx(A) And dy(A) Representing a translation parameter; dw(A) And dh(A) Representing a scaling parameter.
The beneficial effects of the further scheme are as follows: the position and the size of the candidate frame can be accurately adjusted through the translation parameter and the scaling parameter determined in the previous steps, so that the adjusted candidate frame can accurately represent the position of the diagnostic object, and the candidate feature can be accurately determined subsequently.
Further: the training of the fast RCNN network by adopting the training samples for carrying out secondary training on the fast RCNN network specifically comprises the following steps:
s25: initializing the region suggestion network RPN by using the trained fast RCNN, and repeating the step S22 based on the initialized region suggestion network RPN to perform secondary training on the region suggestion network RPN;
s26: and repeating the step S23 and the step S24 based on the secondarily trained region suggestion network RPN to secondarily train the target detection layer.
The beneficial effects of the further scheme are as follows: through carrying out secondary training on the Faster RCNN, the identification and positioning accuracy of the fast RCNN can be greatly improved, and the intelligent diagnosis accuracy of the thermal state of the cable terminal can be improved.
Further: after the target infrared image information of the cable terminal to be detected is input into the trained fast RCNN network for recognition and positioning, the method further comprises the following steps:
and respectively setting the RGB components of the pixel points outside the target candidate frame containing the target diagnosis object to zero based on the identification and positioning results of the Faster RCNN.
The beneficial effects of the further scheme are as follows: the RGB components of the pixel points outside the target candidate frame containing the target diagnosis object are respectively set to be zero, so that the target diagnosis object can be highlighted, the influence of other interference information in the image on subsequent processing is removed, and the diagnosis accuracy is improved.
Further: in step S4, the step of performing thermal state diagnosis on the identified and positioned target diagnostic object according to the preset diagnostic criteria specifically includes the following steps:
s41: carrying out graying processing on the region where the target candidate frame containing the target diagnosis object is located after identification and positioning processing to obtain a grayed image;
s42: respectively extracting ABC three-phase highest gray value I of the target diagnosis object based on the gray imageA、IBAnd ICAnd calculating the ABC three-phase maximum temperature T of the target diagnosis objectA、TBAnd TCThe calculation formula is as follows:
T*＝(Tmax-Tmin)*I*/255+Tmin (13)
wherein the subscript denotes an a phase, a B phase or a C phase; t isminIndicates the initialA minimum temperature of the infrared image; t ismaxRepresents the maximum temperature of the initial infrared image;
s43: ABC three-phase maximum temperature T of object according to target diagnosisA、TBAnd TCCalculating the temperature parameter of the cable terminal to be detected, wherein the calculation formula is as follows:
Tr＝T1-T0 (14)
Td＝T1-T2 (15)
δ＝(T1-T2)/(T1-T0) (16)
in the formula, T1ABC three-phase maximum temperature T for target diagnosis objectA、TBAnd TCMaximum value of (1), T2ABC three-phase maximum temperature T for target diagnosis objectA、TBAnd TCMinimum value of (1), T0Is ambient temperature, TrFor a temperature rise, TdIs the temperature difference, δ is the relative temperature difference;
s44: ABC three-phase maximum temperature T of object according to target diagnosisA、TBAnd TCAnd a temperature parameter Tr、TdAnd delta, determining the thermal state of the target diagnosis object of the cable terminal by combining the preset diagnosis standard.
The beneficial effects of the further scheme are as follows: the temperature parameter of the target diagnosis object of the cable terminal to be detected is calculated according to the ABC three-phase highest gray value of the diagnosis object by carrying out gray processing on the area where the target candidate frame is located, so that automatic accurate diagnosis of the thermal state is realized by combining a preset diagnosis standard.
The invention also provides an intelligent diagnosis system for the thermal state of the cable terminal, which comprises a collecting and labeling module, a training module, an identifying and positioning module and a diagnosis module;
the acquisition marking module is used for marking infrared image information of the cable terminal shot in the routing inspection process and constructing a training sample;
the training module is used for constructing a Faster RCNN and training the Faster RCNN by adopting the training samples;
the identification and positioning module is used for inputting target infrared image information of a cable terminal to be detected into the trained fast RCNN for identification and positioning to obtain the category and the position of a target diagnosis object;
and the diagnosis module is used for carrying out thermal state diagnosis on the target diagnosis object after the identification and positioning processing according to a preset diagnosis standard to obtain a thermal state diagnosis result of the cable terminal to be detected.
According to the intelligent diagnosis system for the thermal state of the cable terminal, the fast RCNN is constructed, the training sample constructed by the acquired infrared image information of the cable terminal is used for training, the trained fast RCNN is used for automatically identifying and positioning the target infrared image information of the cable terminal to be detected, thermal diagnosis is completed, the part of manual operation in the current infrared diagnosis can be effectively replaced, the dependence on manual processing and analysis is reduced, the diagnosis efficiency is greatly improved, the diagnosis accuracy is improved, and the deviation caused by the manual diagnosis is effectively avoided.
Drawings
Fig. 1 is a schematic flow chart of a cable termination thermal state intelligent diagnosis method according to an embodiment of the present invention;
FIG. 2 is a diagram illustrating an example of the positioning and identifying result of the trained Faster RCNN network according to an embodiment of the present invention;
FIG. 3 is a diagram illustrating the effect of retaining only the images inside the candidate frames for positioning by the Faster RCNN network according to an embodiment of the present invention;
fig. 4 is a block diagram of a cable termination thermal state intelligent diagnosis system according to an embodiment of the present invention.
Detailed Description
The principles and features of this invention are described below in conjunction with the following drawings, which are set forth by way of illustration only and are not intended to limit the scope of the invention.
As shown in fig. 1, a cable termination thermal state intelligent diagnosis method includes the following steps:
s1: marking infrared image information of the cable terminal shot in the routing inspection process, and constructing a training sample;
s2: constructing a fast RCNN network, and training the fast RCNN network by adopting the training sample;
s3: inputting target infrared image information of a cable terminal to be detected into the trained fast RCNN for identification and positioning to obtain the category and the position of a target diagnosis object;
s4: and carrying out thermal state diagnosis on the target diagnosis object subjected to the identification and positioning treatment according to a preset diagnosis standard to obtain a thermal state diagnosis result of the cable terminal to be detected.
According to the intelligent diagnosis method for the thermal state of the cable terminal, the fast RCNN is constructed, the training sample constructed by the acquired infrared image information of the cable terminal is used for training, the trained fast RCNN is used for automatically identifying and positioning the infrared image information of the cable terminal to be detected, thermal diagnosis is completed, the part of manual operation in the current infrared diagnosis can be effectively replaced, the dependence on manual processing and analysis is reduced, the diagnosis efficiency is greatly improved, the diagnosis accuracy is improved, and the deviation caused by the manual diagnosis is effectively avoided.
In one or more embodiments of the present invention, in step S1, the infrared image information of the cable terminal captured in the label inspection process specifically includes:
the infrared image information of the cable terminal is collected, the infrared image information for the experiment is shot by the FLIR thermal infrared imager, and the image size is 480 × 640.
After infrared image information of a cable terminal is collected, marking a marking frame at the position of a diagnostic object in the infrared image information, and inputting the name of the marking frame;
wherein the diagnostic object is contained within the label box.
In the invention, the software labelImg is used for labeling in the experimental process, the diagnostic object is shown in a rectangular frame, and the name corresponding to the labeling frame is input. Specifically, the position of the cable terminal where the overheat defect often occurs is taken as a diagnosis object, and the specific position and the corresponding label during manual labeling are as follows: an outdoor terminal connecting hardware fitting-label 'b' and an outdoor terminal sleeve-label 'c'; based on the labeled content, each piece of infrared image information forms an xml file, and the file comprises the following information: the name of the image, the coordinate information (including the coordinates of the upper left corner and the lower right corner) of the labeling box and the name corresponding to the labeling box.
By marking the acquired infrared image information of the cable terminal, the subsequent Faster RCNN network can be conveniently trained according to the marking frame as a reference, so that the trained Faster RCNN network can automatically identify and position the diagnostic object in the infrared image of the cable terminal.
In one or more embodiments of the present invention, in step S2, the training the fast RCNN network with the training samples specifically includes the following steps:
s21: initializing the Faster RCNN, inputting the training sample into the convolutional neural network CNN, extracting one or more image characteristics of the infrared image information of the cable terminal by using the convolutional neural network CNN, and forming a characteristic diagram;
specifically, an ImageNet classification task training model is downloaded, parameters of the model are used for initializing a fast RCNN model trained by the method, and then a training sample is input into a convolutional neural network CNN, wherein the convolutional neural network is of a VGG16 type and comprises 13 convolutional layers, 13 excitation layers and 4 pooling layers, and the output of the convolutional layers is as shown in a formula (1); the output of the excitation layer is shown in equation (2):
Figure BDA0002829016150000101
wherein
Figure BDA0002829016150000102
A jth output vector representing the mth convolutional layer; mjRepresenting a set of input feature vectors for the layer;
Figure BDA0002829016150000103
a convolution kernel representing the layer;
Figure BDA0002829016150000104
represents an additive bias; denotes a convolution operation;
f(x)＝max(0,x) (2)
wherein f (x) denotes the ReLU activation function; x represents a characteristic parameter of the convolutional layer output.
S22: automatically generating a plurality of candidate frames containing the diagnostic object on the feature map by using a regional suggestion network (RPN), adjusting the positions and the sizes of the candidate frames according to the deviation between the labeling frame and the candidate frames, mapping the adjusted candidate frames to the feature map of the Convolutional Neural Network (CNN) to obtain corresponding candidate features, and determining the position of the diagnostic object;
s23: dividing the candidate features into a plurality of regions by using a region-of-interest pooling layer RoI, and performing maximum pooling on each region;
s24: inputting the candidate features into a target detection layer, on one hand, calculating the probability that the candidate frame corresponding to the candidate features contains the diagnostic object belonging to different categories through a classification layer so as to determine the category of the diagnostic object; on the other hand, the adjustment parameters of the candidate frame are calculated again through the regression layer so as to perform secondary adjustment on the candidate frame.
On one hand, image feature information of infrared image information of a cable terminal is extracted through a convolutional neural network CNN and forms a feature map, a candidate frame is automatically generated on the feature map through a region suggestion network RPN, the position and the size of the candidate frame are adjusted, the position of a diagnostic object can be preliminarily determined, on the other hand, block pooling processing is carried out through a region-of-interest pooling layer RoI, the probability that the candidate frame corresponding to the candidate feature contains different types of diagnostic objects is calculated through a target detection layer, the type corresponding to diagnosis can be accurately determined, and meanwhile, the adjustment parameters of the candidate frame are calculated again through a regression layer, so that the position of the diagnostic object is accurately determined, and training of a fast RCNN is completed.
In one or more embodiments of the present invention, the adjusting the position and size of the candidate frame according to the deviation between the label frame and the candidate frame in step S22 specifically includes the following steps:
s221: generating adjustment parameter information for adjusting the candidate frame according to the deviation between the labeling frame and the candidate frame;
s222: adjusting the position and the size of the candidate frame according to the adjustment parameter information;
wherein the adjustment parameter information includes a translation parameter and a scaling parameter.
And generating adjustment parameter information through the deviation between the marking frame and the candidate frame, so that the position and the size of the candidate frame can be accurately adjusted, and the positioning accuracy of the whole fast RCNN network is improved conveniently.
In one or more embodiments of the present invention, in step S221, the generating, according to the deviation between the labeling frame and the candidate frame, adjustment parameter information for adjusting the candidate frame specifically includes the following steps:
s2211: the classification layer of the regional suggestion network calculates the probability that the candidate frame contains the diagnostic object by using a Softmax classifier, and preliminarily screens out the candidate frame containing the diagnostic object according to the probability;
s2212: constructing a loss function of the regional proposed network, and calculating the adjustment parameter information of the candidate frame by taking the minimum loss function as a target;
Figure BDA0002829016150000121
wherein:
Figure BDA0002829016150000122
Figure BDA0002829016150000123
Figure BDA0002829016150000124
in the formula: i represents the ith candidate box; p is a radical ofiRepresenting the probability that the candidate frame calculated by the classification layer contains the diagnosis object; t is tiRepresenting the prediction offset obtained by the regression layer, wherein the prediction offset comprises a translation parameter and a scaling parameter; p is a radical ofi *The value is determined by the intersection ratio IoU (intersection over Union) between the candidate frame and the labeling frame predicted by the area suggestion network, and is specifically shown as a formula (9); t is ti *The actual offset is shown as a formula (11); n is a radical ofcls、NregAnd λ are balance parameters;
Figure BDA0002829016150000125
wherein:
Figure BDA0002829016150000126
in the formula: anchor represents a candidate box generated by the regional suggestion network; the GrountTruth represents a marking frame;
Figure BDA0002829016150000131
in the formula: (A)x,Ay,Aw,Ah) Representing the coordinates of the candidate frame before adjustment; (G)x,Gy,Gw,Gh) Representing the coordinates of the label box.
By constructing the loss function of the regional suggestion network and taking the minimum loss function as a target, the adjustment parameters for adjusting the candidate frame are determined, so that the deviation between the adjusted candidate frame and the marking frame is minimum, the adjusted candidate frame can be ensured to accurately cover the diagnostic object, the training effect is ensured, and the positioning accuracy of the trained Faster RCNN network is improved.
In one or more embodiments of the present invention, the specific calculation formula for adjusting the position and size of the candidate frame according to the adjustment parameter information is as follows:
Figure BDA0002829016150000132
Figure BDA0002829016150000133
wherein (A)x,Ay,Aw,Ah) Indicates the coordinates of the candidate frame before adjustment (specifically, A)xAnd AyRespectively representing the abscissa and the ordinate of the center point of the candidate frame before adjustment; a. thewAnd AhRespectively representing the width and height of the candidate frame before adjustment); (G'x,G’y,G’w,G’h) Represents the adjusted candidate frame coordinates (specifically, G'xAnd G'yRespectively represent the abscissa and ordinate G 'of the adjusted candidate frame center point'wAnd G'hRespectively representing the width and height of the adjusted candidate frame), dx(A) And dy(A) Representing a translation parameter; dw(A) And dh(A) Representing a scaling parameter.
The position and the size of the candidate frame can be accurately adjusted through the translation parameter and the scaling parameter determined in the previous steps, so that the adjusted candidate frame can accurately represent the position of the diagnostic object, and the candidate feature can be accurately determined subsequently.
In the embodiment of the present invention, in the step S23, the input candidate features are equally divided into 7 × 7 block Regions by using a Regions of Interest (RoI) pooling layer, and then maximum pooling is performed on each block region, that is, the maximum value of each block region is retained, and finally, the sizes of all candidate features are fixed to 7 × 7.
In an embodiment of the present invention, in step S24, the candidate features are input into a target detection layer, which includes a classification layer and a regression layer, and on one hand, the probability that a candidate box corresponding to the candidate features contains the diagnostic object belonging to different categories is calculated by the classification layer, so as to implement target identification based on this; and on the other hand, the translation parameter and the scaling parameter of the candidate frame are obtained through the regression layer, so that the secondary adjustment of the candidate frame is realized.
Optionally, in one or more embodiments of the present invention, the training the fast RCNN network by using the training samples further includes performing secondary training on the fast RCNN network, which specifically includes the following steps:
s25: initializing the region suggestion network RPN by using the trained fast RCNN, and repeating the step S22 based on the initialized region suggestion network RPN to perform secondary training on the region suggestion network RPN;
s26: and repeating the step S23 and the step S24 based on the secondarily trained region suggestion network RPN to secondarily train the target detection layer.
The fast RCNN network capable of automatically identifying and positioning the diagnostic object in the infrared image information of the cable terminal is obtained by performing secondary training on the fast RCNN network, so that the identification and positioning accuracy can be greatly improved, and the intelligent diagnosis accuracy of the thermal state of the cable terminal can be improved.
In the embodiment of the present invention, in step S3, the infrared image information of the cable terminal to be detected is input into the trained fast RCNN network for identification and positioning, so as to obtain the category and the position of the target diagnostic object, which specifically includes the following steps:
s31, inputting the infrared image information of the cable terminal to be detected into the trained Faster RCNN network;
s32: and extracting one or more target image characteristics of the infrared image information of the cable terminal by using a Convolutional Neural Network (CNN) and forming a target characteristic diagram.
S33: automatically generating a plurality of target candidate frames containing target diagnosis objects on the target feature map by using a regional suggestion network (RPN), adjusting the positions and sizes of the target candidate frames according to the deviation between the labeling frame and the target candidate frames, mapping the adjusted target candidate frames to the target feature map of the Convolutional Neural Network (CNN) to obtain corresponding target candidate features, and determining the positions of the target diagnosis objects;
s34: dividing the target candidate features into a plurality of regions by using a region-of-interest pooling layer RoI, performing maximum pooling treatment on each region, fixing the target candidate features to be 7 x 7 in size by using the pooling layer, and outputting the target candidate features to a target detection layer;
s35: according to the target candidate characteristics, the target detection layer determines a specific category (such as an outdoor terminal link fitting (b) or an outdoor terminal sleeve (c)) of a diagnostic object in a target candidate frame through the classification layer; on the other hand, the regression layer is used for carrying out secondary adjustment on the target candidate frame to realize accurate positioning of the target diagnosis object, so that identification and positioning of infrared image information of the cable terminal to be detected are completed, as shown in fig. 2, the image can only be presented in an achromatic color due to relevant provisions of patent implementation rules, and actually is a color image of an infrared background.
Optionally, in one or more embodiments of the present invention, after the infrared image information of the cable terminal to be detected is input into the trained fast RCNN network for identification and positioning, the method further includes the following steps:
s36: and respectively setting the RGB components of the pixel points outside the target candidate frame containing the target diagnosis object to zero based on the identification and positioning results of the Faster RCNN.
The RGB components of the pixel points outside the target candidate frame containing the target diagnosis object are respectively set to be zero, so that the target diagnosis object can be highlighted, the influence of other interference information in the image on subsequent processing is removed, and the diagnosis accuracy is improved. As shown in fig. 3a and 3b, the effect diagrams of only retaining the target candidate frames corresponding to the link fittings and the sleeves respectively are obtained by respectively setting the RGB components of the pixels in the region outside the target candidate frames to zero, that is, the outside of the target candidate frames are all black.
In one or more embodiments of the present invention, in step S4, the performing thermal status diagnosis on the identified and located target diagnostic object according to the preset diagnostic criteria specifically includes the following steps:
s41: performing graying processing on the area where the target candidate frame containing the target diagnosis object is located after the identification and positioning processing to obtain a grayed image, wherein the specific calculation formula is as follows:
I＝0.299R+0.587G+0.114B (12)
wherein, I represents the gray level of the pixel point; r, G and B are the red, green and blue components of the pixel, respectively. The effect graph after graying is shown in fig. 3a and 3 b.
S42: respectively extracting ABC three-phase highest gray value I of the target diagnosis object based on the gray imageA、IBAnd ICAnd calculating the ABC three-phase maximum temperature T of the target diagnosis objectA、TBAnd TCThe calculation formula is as follows:
T*＝(Tmax-Tmin)*I*/255+Tmin (13)
wherein the subscript denotes an a phase, a B phase or a C phase; t isminRepresents the minimum temperature of the initial infrared image; t ismaxRepresents the maximum temperature of the initial infrared image;
s43: ABC three-phase maximum temperature T of object according to target diagnosisA、TBAnd TCCalculating the temperature parameter of the cable terminal to be detected, wherein the calculation formula is as follows:
Tr＝T1-T0 (14)
Td＝T1-T2 (15)
δ＝(T1-T2)/(T1-T0) (16)
in the formula, T1ABC three-phase maximum temperature T for target diagnosis objectA、TBAnd TCMaximum value of (1), T2ABC three-phase maximum temperature T for target diagnosis objectA、TBAnd TCMinimum value of (1), T0Is ambient temperature, TrFor a temperature rise, TdIs the temperature difference, δ is the relative temperature difference;
s44: root of herbaceous plantABC three-phase maximum temperature T of object according to target diagnosisA、TBAnd TCAnd a temperature parameter Tr、TdAnd delta, determining the thermal state of the target diagnosis object of the cable terminal by combining the preset diagnosis standard.
Which diagnostic criterion is compared depends on the recognition of the positioning rectangle by the Faster RCNN network: if the label is identified as the label 'b', contrasting an outdoor terminal connection hardware tool diagnosis standard; if the label is identified as "c", the outdoor terminal bushing diagnostic criteria is compared.
The temperature parameter of the target diagnosis object of the cable terminal to be detected is calculated according to the ABC three-phase highest gray value of the diagnosis object by carrying out gray processing on the area where the target candidate frame is located, so that automatic accurate diagnosis of the thermal state is realized by combining a preset diagnosis standard.
In the embodiment of the invention, the outdoor terminal connecting hardware tool thermal state diagnosis:
1) the diagnosis standard of the thermal state of the outdoor terminal link fitting is shown in table 1:
TABLE 1 outdoor terminal link fitting thermal state diagnostic criteria
Figure BDA0002829016150000171
2) According to the diagnosis standard, the state diagnosis of the outdoor terminal link fitting is completed according to the following steps:
a) maximum temperature T of ABC three phasesA、TBAnd TCJudging whether the temperature exceeds 130 ℃, if so, judging that the outdoor terminal connecting hardware has an emergency defect, otherwise, entering the step b);
b) maximum temperature T of ABC three phasesA、TBAnd TCWhether the temperature exceeds 90 ℃ or not is judged, if yes, the outdoor terminal connection hardware fitting is judged to have major defects, and if not, the step c) is carried out;
c) temperature parameter Td(temperature difference) whether the temperature exceeds 40 ℃, if so, judging that the outdoor terminal connecting hardware has an emergency defect, otherwise, entering the step d);
d) temperature parameter Td(temperature difference) is judged whether to exceed 15 ℃, if so, the outdoor terminal connection hardware is judged to have major defects, otherwise, the step e is carried out);
e) temperature parameter Td(temperature difference) is judged whether to exceed 5 ℃, if so, the outdoor terminal connection hardware fitting is judged to have common defects, otherwise, the step f is carried out);
f) temperature parameter Tr(temperature rise) whether the temperature exceeds 15 ℃, if not, judging that the outdoor terminal connecting hardware is in a normal state, otherwise, entering the step g);
g) whether the temperature parameter delta (relative temperature difference) exceeds 95% or not, if yes, judging that the outdoor terminal connection hardware has an emergency defect, and otherwise, entering the step h);
h) whether the temperature parameter delta (relative temperature difference) exceeds 80%, if so, judging that the outdoor terminal connection hardware has a major defect, otherwise, entering the step i);
i) and judging whether the temperature parameter delta (relative temperature difference) exceeds 35%, if so, judging that the outdoor terminal connection hardware has a common defect, otherwise, judging that the outdoor terminal connection hardware is in a normal state.
In the embodiment of the invention, the outdoor terminal sleeve thermal state diagnosis:
1) the outdoor termination bushing thermal state diagnostic criteria are shown in table 2:
TABLE 2 outdoor termination bushing thermal state diagnostic criteria
Figure BDA0002829016150000181
2) According to the diagnosis standard, the state diagnosis of the outdoor terminal sleeve is completed according to the following steps:
a) temperature parameter Td(temperature difference) is judged whether to exceed 4 ℃, if so, the outdoor terminal sleeve is judged to have major defects, otherwise, the step b is carried out;
b) temperature parameter TdAnd (temperature difference) is judged whether to exceed 2 ℃, if so, the outdoor terminal sleeve is judged to have common defects, otherwise, the outdoor terminal sleeve is judged to be in a normal state.
As shown in fig. 4, the present invention further provides an intelligent diagnosis system for the thermal state of the cable terminal, which includes a collecting and labeling module, a training module, an identifying and positioning module, and a diagnosis module;
the acquisition marking module is used for marking infrared image information of the cable terminal shot in the routing inspection process and constructing a training sample;
the training module is used for constructing a Faster RCNN and training the Faster RCNN by adopting the training samples;
the identification and positioning module is used for inputting infrared image information of a cable terminal to be detected into the trained fast RCNN for identification and positioning to obtain the category and the position of a target diagnosis object;
and the diagnosis module is used for carrying out thermal state diagnosis on the target diagnosis object after the identification and positioning processing according to a preset diagnosis standard to obtain a thermal state diagnosis result of the cable terminal to be detected.
According to the intelligent diagnosis system for the thermal state of the cable terminal, the fast RCNN is constructed, the training sample constructed by the acquired infrared image information of the cable terminal is used for training, the trained fast RCNN is used for automatically identifying and positioning the infrared image information of the cable terminal to be detected, thermal diagnosis is completed, the part of manual operation in the current infrared diagnosis can be effectively replaced, the dependence on manual processing and analysis is reduced, the diagnosis efficiency is greatly improved, the diagnosis accuracy is improved, and the deviation caused by the manual diagnosis is effectively avoided.
In one or more embodiments of the present invention, the collection labeling module is specifically configured to:
acquiring infrared image information of a cable terminal, marking a marking frame at the position of a diagnostic object in the infrared image information, and inputting the name of the marking frame;
wherein the diagnostic object is contained within the label box.
By marking the acquired infrared image information of the cable terminal, the subsequent Faster RCNN network can be conveniently trained according to the marking frame as a reference, so that the trained Faster RCNN network can automatically identify and position the diagnostic object in the infrared image of the cable terminal.
In one or more embodiments of the present invention, the specific implementation of the training module for training the fast RCNN network by using the training samples is as follows:
initializing the Faster RCNN, inputting the training sample into the convolutional neural network CNN, extracting one or more image characteristics of the infrared image information of the cable terminal by using the convolutional neural network CNN, and forming a characteristic diagram;
automatically generating a plurality of candidate frames containing the diagnostic object on the feature map by using a regional suggestion network (RPN), adjusting the positions and the sizes of the candidate frames according to the deviation between the labeling frame and the candidate frames, mapping the adjusted candidate frames to the feature map of the Convolutional Neural Network (CNN) to obtain corresponding candidate features, and determining the position of the diagnostic object;
dividing the candidate features into a plurality of regions by using a region-of-interest pooling layer RoI, and performing maximum pooling on each region;
inputting the candidate features into a target detection layer, on one hand, calculating the probability that the candidate frame corresponding to the candidate features contains the diagnostic object belonging to different categories through a classification layer so as to determine the category of the diagnostic object; on the other hand, the adjustment parameters of the candidate frame are calculated again through the regression layer so as to perform secondary adjustment on the candidate frame.
On one hand, image feature information of infrared image information of a cable terminal is extracted through a convolutional neural network CNN and forms a feature map, a candidate frame is automatically generated on the feature map through a region suggestion network RPN, the position and the size of the candidate frame are adjusted, the position of a diagnostic object can be preliminarily determined, on the other hand, block pooling processing is carried out through a region-of-interest pooling layer RoI, the probability that the candidate frame corresponding to the candidate feature contains different types of diagnostic objects is calculated through a target detection layer, the type corresponding to diagnosis can be accurately determined, and meanwhile, the adjustment parameters of the candidate frame are calculated again through a regression layer, so that the position of the diagnostic object is accurately determined, and training of a fast RCNN is completed.
In one or more embodiments of the present invention, the implementation of the training module adjusting the position and size of the candidate frame according to the deviation between the labeled frame and the candidate frame is as follows:
generating adjustment parameter information for adjusting the candidate frame according to the deviation between the labeling frame and the candidate frame;
adjusting the position and the size of the candidate frame according to the adjustment parameter information;
wherein the adjustment parameter information includes a translation parameter and a scaling parameter.
And generating adjustment parameter information through the deviation between the marking frame and the candidate frame, so that the position and the size of the candidate frame can be accurately adjusted, and the positioning accuracy of the whole fast RCNN network is improved conveniently.
In one or more embodiments of the present invention, the specific implementation of the training module generating the adjustment parameter information for adjusting the candidate box according to the deviation between the labeled box and the candidate box is as follows:
the classification layer of the regional suggestion network calculates the probability that the candidate frame contains the diagnostic object by using a Softmax classifier, and preliminarily screens out the candidate frame containing the diagnostic object according to the probability;
constructing a loss function of the regional proposed network, and calculating the adjustment parameter information of the candidate frame by taking the minimum loss function as a target;
Figure BDA0002829016150000211
wherein:
Figure BDA0002829016150000212
Figure BDA0002829016150000213
Figure BDA0002829016150000214
in the formula: i represents the ith candidate box; p is a radical ofiRepresenting the probability that the candidate frame calculated by the classification layer contains the diagnosis object; t is tiRepresenting the prediction offset obtained by the regression layer, wherein the prediction offset comprises a translation parameter and a scaling parameter;
Figure BDA0002829016150000215
the value is determined by the intersection ratio IoU (intersection over Union) between the candidate frame and the labeling frame predicted by the area suggestion network, and is specifically shown as a formula (9);
Figure BDA0002829016150000216
the actual offset is shown as a formula (11); n is a radical ofcls、NregAnd λ are balance parameters;
Figure BDA0002829016150000217
wherein:
Figure BDA0002829016150000218
in the formula: anchor represents a candidate box generated by the regional suggestion network; the GrountTruth represents a marking frame;
Figure BDA0002829016150000221
in the formula: (A)x,Ay,Aw,Ah) Representing the coordinates of the candidate frame before adjustment; (G)x,Gy,Gw,Gh) Representing the coordinates of the label box.
In one or more embodiments of the present invention, the specific calculation formula for the training module to adjust the position and size of the candidate frame according to the adjustment parameter information is as follows:
Figure BDA0002829016150000222
Figure BDA0002829016150000223
wherein (A)x,Ay,Aw,Ah) Representing the coordinates of the candidate frame before adjustment; (G'x,G’y,G’w,G’h) Representing adjusted candidate frame coordinates, dx(A) And dy(A) Representing a translation parameter; dw(A) And dh(A) Representing a scaling parameter.
The position and the size of the candidate frame can be accurately adjusted through the determined translation parameter and the determined scaling parameter, so that the adjusted candidate frame can accurately represent the position of the diagnostic object, and the candidate feature can be accurately determined subsequently.
Optionally, in one or more embodiments of the present invention, the training module is further configured to perform secondary training on the fast RCNN network, and specifically implemented as:
s25: initializing the region suggestion network RPN by using the trained fast RCNN, and repeating the step S22 based on the initialized region suggestion network RPN to perform secondary training on the region suggestion network RPN;
s26: and repeating the step S23 and the step S24 based on the secondarily trained region suggestion network RPN to secondarily train the target detection layer.
Through carrying out secondary training on the Faster RCNN, the identification and positioning accuracy of the fast RCNN can be greatly improved, and the intelligent diagnosis accuracy of the thermal state of the cable terminal can be improved.
Optionally, in one or more embodiments of the present invention, after the recognition and positioning module inputs the infrared image information of the cable terminal to be detected into the trained fast RCNN network for recognition and positioning, the recognition and positioning module is further configured to:
and respectively setting the RGB components of the pixel points outside the candidate frame containing the diagnostic object to zero based on the identification and positioning results of the Faster RCNN.
The RGB components of the pixel points outside the candidate frame containing the diagnostic object are respectively set to be zero, so that the diagnostic object can be highlighted, the influence of other interference information in the image on subsequent processing is removed, and the diagnostic accuracy is improved.
In one or more embodiments of the invention, the diagnostic module is specifically configured to:
carrying out graying processing on the area where the candidate frame containing the target diagnosis object is located after identification and positioning processing to obtain a grayed image;
respectively extracting ABC three-phase highest gray value I of the target diagnosis object based on the gray imageA、IBAnd ICAnd calculating the ABC three-phase maximum temperature T of the target diagnosis objectA、TBAnd TCThe calculation formula is as follows:
T*＝(Tmax-Tmin)*I*/255+Tmin (13)
wherein the subscript denotes an a phase, a B phase or a C phase; t isminRepresents the minimum temperature of the initial infrared image; t ismaxRepresents the maximum temperature of the initial infrared image;
ABC three-phase maximum temperature T of object according to target diagnosisA、TBAnd TCCalculating the temperature parameter of the cable terminal to be detected, wherein the calculation formula is as follows:
Tr＝T1-T0 (14)
Td＝T1-T2 (15)
δ＝(T1-T2)/(T1-T0) (16)
in the formula, T1ABC three-phase maximum temperature T for target diagnosis objectA、TBAnd TCMaximum value of (1), T2ABC triphase for target diagnostic objectsMaximum temperature TA、TBAnd TCMinimum value of (1), T0Is ambient temperature, TrFor a temperature rise, TdIs the temperature difference, δ is the relative temperature difference;
ABC three-phase maximum temperature T of object according to target diagnosisA、TBAnd TCAnd a temperature parameter Tr、TdAnd delta, determining the thermal state of the target diagnosis object of the cable terminal by combining the preset diagnosis standard.
The temperature parameter of the target diagnostic object of the cable terminal to be detected is calculated according to the ABC three-phase highest gray value of the diagnostic object by carrying out gray processing on the area where the candidate frame is located, so that automatic accurate diagnosis of the thermal state is realized by combining a preset diagnostic standard.
It will be apparent to those skilled in the art that, for convenience and simplicity of description, the foregoing functional units and modules are merely illustrated in terms of division, and in practical applications, the foregoing functional allocation may be performed by different functional units and modules as needed, that is, the internal structure of the terminal is divided into different functional units or modules to perform all or part of the above described functions. Each functional unit and module in the embodiments may be integrated in one processing unit, or each unit may exist alone physically, or two or more units are integrated in one unit, and the integrated unit may be implemented in a form of hardware, or in a form of software functional unit. In addition, specific names of the functional units and modules are only for convenience of distinguishing from each other, and are not used for limiting the protection scope of the present application. The specific working processes of the units and modules in the above-mentioned apparatus refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.
In the above embodiments, the descriptions of the respective embodiments have respective emphasis, and reference may be made to the related descriptions of other embodiments for parts that are not described or illustrated in a certain embodiment.
Those of ordinary skill in the art will appreciate that the various illustrative elements and method steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the implementation. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.
In the embodiments provided by the present invention, it should be understood that the disclosed system and method can be implemented in other ways. For example, the above-described embodiments are merely illustrative, and for example, a module or a unit may be divided into only one logic function, and may be implemented in other ways, for example, a plurality of units or components may be combined or integrated into another device, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, terminals or units, and may be in an electrical, mechanical or other form.
Units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
In addition, functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit can be realized in a form of hardware, and can also be realized in a form of a software functional unit.
The invention is not limited solely to that described in the specification and embodiments, and additional advantages and modifications will readily occur to those skilled in the art, so that the invention is not limited to the specific details, representative apparatus, and illustrative examples shown and described herein, without departing from the spirit and scope of the general concept as defined by the appended claims and their equivalents.

Claims (10)

1. The intelligent diagnosis method for the thermal state of the cable terminal is characterized by comprising the following steps:
s1: marking infrared image information of the cable terminal shot in the routing inspection process, and constructing a training sample;
s2: constructing a fast RCNN network, and training the fast RCNN network by adopting the training sample;
s3: inputting target infrared image information of a cable terminal to be detected into the trained fast RCNN for identification and positioning to obtain the category and the position of a target diagnosis object;
s4: and carrying out thermal state diagnosis on the target diagnosis object subjected to the identification and positioning treatment according to a preset diagnosis standard to obtain a thermal state diagnosis result of the cable terminal to be detected.
2. The intelligent diagnosis method for the thermal state of the cable terminal according to claim 1, wherein in step S1, the step of labeling the infrared image information of the cable terminal captured in the inspection process specifically includes:
acquiring infrared image information of a cable terminal, marking a marking frame at the position of a diagnostic object in the infrared image information, and inputting the name of the marking frame;
wherein the diagnostic object is contained within the label box.
3. The intelligent diagnosis method for the thermal state of the cable terminal according to claim 2, wherein in step S2, the training of the fast RCNN network with the training samples specifically includes the following steps:
s21: initializing the Faster RCNN, inputting the training sample into the convolutional neural network CNN, extracting one or more image characteristics of the infrared image information of the cable terminal by using the convolutional neural network CNN, and forming a characteristic diagram;
s22: automatically generating a plurality of candidate frames containing the diagnostic object on the feature map by using a regional suggestion network (RPN), adjusting the positions and the sizes of the candidate frames according to the deviation between the labeling frame and the candidate frames, mapping the adjusted candidate frames to the feature map of the Convolutional Neural Network (CNN) to obtain corresponding candidate features, and determining the position of the diagnostic object;
s23: dividing the candidate features into a plurality of regions by using a region-of-interest pooling layer RoI, and performing maximum pooling on each region;
s24: inputting the candidate features into a target detection layer, on one hand, calculating the probability that the candidate frame corresponding to the candidate features contains the diagnostic object belonging to different categories through a classification layer so as to determine the category of the diagnostic object; on the other hand, the adjustment parameters of the candidate frame are calculated again through the regression layer so as to perform secondary adjustment on the candidate frame.
4. The intelligent diagnosis method for the thermal state of the cable termination according to claim 3, wherein in step S22, the step of adjusting the position and size of the candidate frame according to the deviation between the label frame and the candidate frame specifically comprises the following steps:
s221: generating adjustment parameter information for adjusting the candidate frame according to the deviation between the labeling frame and the candidate frame;
s222: adjusting the position and the size of the candidate frame according to the adjustment parameter information;
wherein the adjustment parameter information includes a translation parameter and a scaling parameter.
5. The intelligent diagnosis method for the thermal state of the cable termination according to claim 4, wherein in step S221, the generating of the adjustment parameter information for adjusting the candidate frame according to the deviation between the labeling frame and the candidate frame specifically includes the following steps:
s2211: the classification layer of the regional suggestion network calculates the probability that the candidate frame contains the diagnostic object by using a Softmax classifier, and preliminarily screens out the candidate frame containing the diagnostic object according to the probability;
s2212: constructing a loss function of the regional proposed network, and calculating the adjustment parameter information of the candidate frame by taking the minimum loss function as a target;
Figure FDA0002829016140000021
wherein:
Figure FDA0002829016140000022
Figure FDA0002829016140000023
Figure FDA0002829016140000031
in the formula: i represents the ith candidate box; p is a radical ofiRepresenting the probability that the candidate frame calculated by the classification layer contains the diagnosis object; t is tiRepresenting the prediction offset obtained by the regression layer, wherein the prediction offset comprises a translation parameter and a scaling parameter;
Figure FDA0002829016140000032
the value is determined by the intersection ratio IoU (intersection over Union) between the candidate frame and the labeling frame predicted by the area suggestion network, and is specifically shown as a formula (9);
Figure FDA0002829016140000033
the actual offset is shown as a formula (11); n is a radical ofcls、NregAnd λ are balance parameters;
Figure FDA0002829016140000034
wherein:
Figure FDA0002829016140000035
in the formula: anchor represents a candidate box generated by the regional suggestion network; the GrountTruth represents a marking frame;
Figure FDA0002829016140000036
in the formula: (A)x,Ay,Aw,Ah) Representing the coordinates of the candidate frame before adjustment; (G)x,Gy,Gw,Gh) Representing the coordinates of the label box.
6. The intelligent diagnosis method for the thermal state of the cable termination according to claim 4, wherein the specific calculation formula for adjusting the position and size of the candidate frame according to the adjustment parameter information is as follows:
Figure FDA0002829016140000037
Figure FDA0002829016140000038
wherein (A)x,Ay,Aw,Ah) Representing the coordinates of the candidate frame before adjustment; (G'x,G’y,G’w,G’h) Representing adjusted candidate frame coordinates, dx(A) And dy(A) Representing a translation parameter; dw(A) And dh(A) Representing a scaling parameter.
7. The intelligent diagnosis method for the thermal state of the cable terminal as claimed in claim 5, wherein the training of the Faster RCNN network using the training samples further comprises a secondary training of the fast RCNN network, specifically comprising the steps of:
s25: initializing the region suggestion network RPN by using the trained fast RCNN, and repeating the step S22 based on the initialized region suggestion network RPN to perform secondary training on the region suggestion network RPN;
s26: and repeating the step S23 and the step S24 based on the secondarily trained region suggestion network RPN to secondarily train the target detection layer.
8. The intelligent diagnosis method for the thermal state of the cable terminal according to claim 1, wherein the method further comprises the following steps after inputting the target infrared image information of the cable terminal to be detected into the trained fast RCNN network for identification and positioning:
and respectively setting the RGB components of the pixel points outside the target candidate frame containing the target diagnosis object to zero based on the identification and positioning results of the Faster RCNN.
9. The intelligent diagnosis method for the thermal state of the cable terminal according to any one of claims 1 to 8, wherein in step S4, the step of performing the thermal state diagnosis on the identified and located target diagnosis object according to the preset diagnosis standard specifically includes the following steps:
s41: carrying out graying processing on the region where the target candidate frame containing the target diagnosis object is located after identification and positioning processing to obtain a grayed image;
s42: respectively extracting ABC three-phase highest gray value I of the target diagnosis object based on the gray imageA、IBAnd ICAnd calculating the ABC three-phase maximum temperature T of the target diagnosis objectA、TBAnd TCThe calculation formula is as follows:
T*＝(Tmax-Tmin)*I*/255+Tmin (13)
wherein the subscript denotes an a phase, a B phase or a C phase; t isminRepresents the minimum temperature of the initial infrared image; t ismaxRepresents the maximum temperature of the initial infrared image;
s43: ABC three-phase maximum temperature T of object according to target diagnosisA、TBAnd TCCalculating the temperature parameter of the cable terminal to be detected, wherein the calculation formula is as follows:
Tr＝T1-T0 (14)
Td＝T1-T2 (15)
δ＝(T1-T2)/(T1-T0) (16)
in the formula, T1ABC three-phase maximum temperature T for target diagnosis objectA、TBAnd TCMaximum value of (1), T2ABC three-phase maximum temperature T for target diagnosis objectA、TBAnd TCMinimum value of (1), T0Is ambient temperature, TrFor a temperature rise, TdIs the temperature difference, δ is the relative temperature difference;
s44: ABC three-phase maximum temperature T of object according to target diagnosisA、TBAnd TCAnd a temperature parameter Tr、TdAnd delta, determining the thermal state of the target diagnosis object of the cable terminal by combining the preset diagnosis standard.
10. An intelligent diagnosis system for the thermal state of a cable terminal is characterized by comprising a collecting and labeling module, a training module, an identifying and positioning module and a diagnosis module;
the acquisition marking module is used for marking infrared image information of the cable terminal shot in the routing inspection process and constructing a training sample;
the training module is used for constructing a Faster RCNN and training the Faster RCNN by adopting the training samples;
the identification and positioning module is used for inputting target infrared image information of a cable terminal to be detected into the trained fast RCNN for identification and positioning to obtain the category and the position of a target diagnosis object;
and the diagnosis module is used for carrying out thermal state diagnosis on the target diagnosis object after the identification and positioning processing according to a preset diagnosis standard to obtain a thermal state diagnosis result of the cable terminal to be detected.
CN202011437678.8A 2020-12-10 2020-12-10 Intelligent diagnosis method and system for thermal state of cable terminal Pending CN112465797A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011437678.8A CN112465797A (en) 2020-12-10 2020-12-10 Intelligent diagnosis method and system for thermal state of cable terminal

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011437678.8A CN112465797A (en) 2020-12-10 2020-12-10 Intelligent diagnosis method and system for thermal state of cable terminal

Publications (1)

Publication Number Publication Date
CN112465797A true CN112465797A (en) 2021-03-09

Family

ID=74801555

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011437678.8A Pending CN112465797A (en) 2020-12-10 2020-12-10 Intelligent diagnosis method and system for thermal state of cable terminal

Country Status (1)

Country Link
CN (1) CN112465797A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113505837A (en) * 2021-07-14 2021-10-15 同济大学 Measuring mark automatic extraction method based on fast-RCNN

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105718687A (en) * 2016-01-26 2016-06-29 国家电网公司 Power-cable electric-heat degradation simulation method based on seasonal loads and temperature cycles
CN107607207A (en) * 2017-08-22 2018-01-19 中国科学院深圳先进技术研究院 A kind of power equipment thermal fault detection method, system and electronic equipment

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105718687A (en) * 2016-01-26 2016-06-29 国家电网公司 Power-cable electric-heat degradation simulation method based on seasonal loads and temperature cycles
CN107607207A (en) * 2017-08-22 2018-01-19 中国科学院深圳先进技术研究院 A kind of power equipment thermal fault detection method, system and electronic equipment

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
MERINGUE_ZZ: "目标定位和检测系列（7）：Faster R-CNN", 《HTTPS://BLOG.CSDN.NET/SINAT_34474705/ARTICLE/DETAILS/80663128/》 *
徐小冰: "基于红外图像的高压电缆附件热状态智能诊断方法", 《中国优秀硕士学位论文全文数据库工程科技II辑》 *
徐小冰;袁婧;廖雁群;韦亦龙;周承科;周文俊;: "基于Faster RCNN与Mean-Shift的电缆附件缺陷红外图像自动诊断方法", 《高电压技术》 *
徐小冰等: "基于 Faster RCNN 与 Mean-Shift 的电缆附件缺陷红外图像自动诊断方法", 《高电压技术》 *
靖小平;彭小圣;姜伟;周文俊;周承科;唐泽洋;: "基于K-Means聚类算法的自动图谱识别在电缆局部放电在线监测系统中的应用", 高电压技术 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113505837A (en) * 2021-07-14 2021-10-15 同济大学 Measuring mark automatic extraction method based on fast-RCNN

Similar Documents

Publication Publication Date Title
Wang et al. Automatic fault diagnosis of infrared insulator images based on image instance segmentation and temperature analysis
CN112734692B (en) Defect identification method and device for power transformation equipment
CN111624229B (en) Intelligent charged equipment fault diagnosis method based on infrared imaging
US10621717B2 (en) System and method for image-based target object inspection
CN110942458B (en) Temperature anomaly defect detection and positioning method and system
CN111751002B (en) Intelligent charged equipment fault diagnosis method based on infrared thermal imaging
EP3432268A1 (en) Diagnosis assisting device, image processing method in diagnosis assisting device, and computer program
CN114494185B (en) Electrical equipment fault detection method based on RGB-T multi-scale feature fusion
CN111798407A (en) Electrified equipment fault diagnosis method based on neural network model
CN112508019B (en) GIS isolation/grounding switch state detection method and system based on image recognition
CN113920097A (en) Power equipment state detection method and system based on multi-source image
CN113688817A (en) Instrument identification method and system for automatic inspection
CN113077416A (en) Welding spot welding defect detection method and system based on image processing
Chavan et al. Quality control of PCB using image processing
CN112465797A (en) Intelligent diagnosis method and system for thermal state of cable terminal
CN113269234B (en) Connecting piece assembly detection method and system based on target detection
CN113221805B (en) Method and device for acquiring image position of power equipment
CN107977959B (en) Respirator state identification method suitable for electric power robot
JP2010008159A (en) Visual inspection processing method
CN117079286A (en) Identification method and identification system for pointer instrument
CN115147591A (en) Transformer equipment infrared image voltage heating type defect diagnosis method and system
CN113205178B (en) Artificial intelligent infrared image sensing system and method
CN114494186A (en) Fault detection method for high-voltage power transmission and transformation line electrical equipment
CN111950558B (en) High-pressure oil immersed transformer oil liquid level image identification method
CN114155246A (en) Deformable convolution-based power transmission tower pin defect detection method

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
RJ01 Rejection of invention patent application after publication

Application publication date: 20210309