CN109255347A - Method and apparatus based on wide-angle automatic identification - Google Patents
Method and apparatus based on wide-angle automatic identification Download PDFInfo
- Publication number
- CN109255347A CN109255347A CN201811105825.4A CN201811105825A CN109255347A CN 109255347 A CN109255347 A CN 109255347A CN 201811105825 A CN201811105825 A CN 201811105825A CN 109255347 A CN109255347 A CN 109255347A
- Authority
- CN
- China
- Prior art keywords
- infrared
- wide
- automatic identification
- under test
- infrared chart
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 18
- 238000012549 training Methods 0.000 claims description 35
- 238000012360 testing method Methods 0.000 claims description 24
- 238000005457 optimization Methods 0.000 claims description 15
- 238000012545 processing Methods 0.000 claims description 11
- 238000013528 artificial neural network Methods 0.000 claims description 9
- 230000002708 enhancing effect Effects 0.000 claims description 9
- 230000009466 transformation Effects 0.000 claims description 9
- NJPPVKZQTLUDBO-UHFFFAOYSA-N novaluron Chemical compound C1=C(Cl)C(OC(F)(F)C(OC(F)(F)F)F)=CC=C1NC(=O)NC(=O)C1=C(F)C=CC=C1F NJPPVKZQTLUDBO-UHFFFAOYSA-N 0.000 claims description 6
- 230000000007 visual effect Effects 0.000 claims description 2
- 238000010606 normalization Methods 0.000 claims 1
- 230000007547 defect Effects 0.000 abstract description 4
- 238000004458 analytical method Methods 0.000 abstract description 3
- 230000002159 abnormal effect Effects 0.000 abstract description 2
- 238000000605 extraction Methods 0.000 description 6
- 230000005540 biological transmission Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 239000012212 insulator Substances 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 230000003028 elevating effect Effects 0.000 description 2
- 238000003475 lamination Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 230000002093 peripheral effect Effects 0.000 description 2
- 238000011176 pooling Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 239000003989 dielectric material Substances 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/10—Image acquisition
- G06V10/12—Details of acquisition arrangements; Constructional details thereof
- G06V10/14—Optical characteristics of the device performing the acquisition or on the illumination arrangements
- G06V10/147—Details of sensors, e.g. sensor lenses
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F16—ENGINEERING ELEMENTS AND UNITS; GENERAL MEASURES FOR PRODUCING AND MAINTAINING EFFECTIVE FUNCTIONING OF MACHINES OR INSTALLATIONS; THERMAL INSULATION IN GENERAL
- F16M—FRAMES, CASINGS OR BEDS OF ENGINES, MACHINES OR APPARATUS, NOT SPECIFIC TO ENGINES, MACHINES OR APPARATUS PROVIDED FOR ELSEWHERE; STANDS; SUPPORTS
- F16M11/00—Stands or trestles as supports for apparatus or articles placed thereon ; Stands for scientific apparatus such as gravitational force meters
- F16M11/02—Heads
- F16M11/04—Means for attachment of apparatus; Means allowing adjustment of the apparatus relatively to the stand
- F16M11/06—Means for attachment of apparatus; Means allowing adjustment of the apparatus relatively to the stand allowing pivoting
- F16M11/10—Means for attachment of apparatus; Means allowing adjustment of the apparatus relatively to the stand allowing pivoting around a horizontal axis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/22—Image preprocessing by selection of a specific region containing or referencing a pattern; Locating or processing of specific regions to guide the detection or recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Vascular Medicine (AREA)
- Mechanical Engineering (AREA)
- Radiation Pyrometers (AREA)
Abstract
The invention discloses a kind of method and apparatus based on wide-angle automatic identification, device includes rack, the torsion shaft hingedly cooperated with rack, thermal infrared imager, the mounting rack for installing thermal infrared imager and being fixed in shaft shot to image, and the cylinder rotated for drive shaft around hinged place, the mounting rack include the rotary shaft that rotation is arranged on torsion shaft and are set to rotary shaft one end and are used to install the mounting base of thermal infrared imager.The present invention can carry out multi-angled shooting to object to be identified, improve recognition accuracy, and method of the invention can be used for picture name and the defect diagonsis of thermal infrared imager floor, with the automated analysis of later period status of electric power and abnormal conditions, the present invention identifies that the accuracy rate of infrared chart power equipment is high.
Description
Technical field
The invention belongs to electrical equipment technical fields, and in particular to a kind of method and apparatus based on wide-angle automatic identification.
Background technique
There are many power equipment in power transmission and transformation line, in order to avoid the generation of failure, because of long-term work, and it is most
Power equipment is subjected to exposing to the sun and rain, and the probability of failure is larger, it is therefore desirable to regular visit.
The conventional method that power equipment in power transmission and transformation line is identified are as follows: power equipment in acquisition power transmission and transformation line
Picture or video, be transmitted to monitoring backstage, monitor the staff on backstage by eye-observation picture or video, it is artificial to judge
Power equipment in power transmission and transformation line whether there is defect.This kind is by the staff's eye-observation for monitoring backstage and judgement is
The mode of no existing defects brings very big workload to staff, can not achieve intelligent defect recognition, and shoot
Image can not carry out multi-angled shooting, and object to be identified cannot be carried out to the complete shooting of multi-angle.
Currently, also there are a variety of intelligent polling methods, the infrared of power equipment to be detected is obtained using thermal infrared imager
Thermal map realizes identifying and positioning for power equipment by technologies such as computer vision, infrared chart processing, and then can carry out shape
State monitoring and fault diagnosis.It is existing for the infrared of power equipment although the inspection for power equipment is more intelligent
Thermal map identification technology still remains some shortcomings, such as weaker for the noise anti-interference ability of background and shooting equipment.
Summary of the invention
It to solve the above-mentioned problems, can be to be identified the present invention provides a kind of device based on wide-angle automatic identification
Object carries out the shooting of multi-angle, to improve recognition effect.
The technical solution of the present invention is as follows: a kind of device based on wide-angle automatic identification, including rack, hingedly cooperate with rack
Torsion shaft, thermal infrared imager, the mounting rack for installing thermal infrared imager and being fixed in shaft that image is shot,
And the cylinder rotated for drive shaft around hinged place, the mounting rack include the rotary shaft that is arranged on torsion shaft of rotation with
And it is set to rotary shaft one end and the mounting base for installing thermal infrared imager.
The device of the invention can make thermal infrared imager realize the shooting for carrying out multi-angle to object to be identified, pass through gas
Cylinder drives torsion shaft to overturn and then drives the overturning of infrared image instrument, and rotary shaft can also be driven to rotate, so that infrared thermal imagery
Instrument rotates, and the driving device of driving thermal infrared imager rotation can be the common structures such as motor in the present invention.
Preferably, the rack includes that pedestal, the vertical rack being set on pedestal and vertical rack are slidably matched
Lifting shaft is fixedly connected with lifter plate and the driving mechanism for driving lifting shaft to slide relative to vertical rack with lifting shaft,
The torsion shaft hingedly cooperates with lifter plate.
Preferably, the driving mechanism includes:
The rack gear being set on lifting shaft;
The gear matched with rack gear;
Motor for the rotation of sliding tooth wheel.
When motor-driven gear rotation, since wheel and rack matches, and rack gear is set on lifting shaft, therefore
Lifting shaft realizes elevating movement when gear rotates, and then realizes the lifting to thermal infrared imager.
The work of motor and cylinder can be controlled in the present invention by controller.
The present invention also provides a kind of methods based on wide-angle automatic identification, include the following steps:
S1: it includes the infrared of object under test to be identified that the above-mentioned device based on wide-angle automatic identification, which is acquired multiple,
Thermal map, and the shooting at multiple visual angles is carried out, obtain training sample set;
S2: enhancing processing and normalizing are successively carried out respectively to each object under test infrared chart in training sample set
Change processing obtains pretreatment infrared chart;
S3: extracting the object under test infrared chart feature in the pretreatment infrared chart: building VGG16 depth network,
Wherein from second two-dimensional convolution third into third convolution module two in the first convolution module in VGG16 depth network
Two-dimensional convolution between dimension convolution is the modular convolution module Fire Module in Squeeze Net, i.e. acquisition base
In the neural network framework of VGG16 and Squeeze Net;
S4: bottleneck characteristic is saved: input training sample set and test sample set, it is defeated before extracting full articulamentum
Out, that is, bottleneck characteristic is obtained, and uses Nadam optimizer training bottleneck characteristic;
S5: the network optimization: the pre- weight of VGG16 depth network and bottleneck characteristic weight freeze the of neural network framework
One convolution module, the second convolution module and third convolution module, and connect Volume Four volume module, the 5th convolution module and entirely
It connects layer and carries out global optimization training, obtain object under test network model, obtain the probability threshold value that infrared chart is object under test;
S6: utilizing the object under test network model, inputs or read infrared chart to be identified, obtains infrared heat to be identified
It include the probability of the object under test in figure, if the probability in infrared chart to be identified including the object under test is general higher than described
When rate threshold value, then the infrared chart to be identified is the object under test.
Power equipment (object to be identified) infrared chart, general power equipment infrared chart size are acquired in the present invention first
For 512*512, and should be comprising power equipment for identification in power equipment infrared chart, and power equipment is in infrared warm
Relative position in figure keeps roughly the same, and wherein power equipment can be a variety of existing power equipments, such as power equipment
It can be electric insulator, or current transformer.
Power equipment information is obtained using thermal infrared imager, wherein the equipment comprising classifying, the present invention is to red
Outer thermal map has carried out enhancing processing, and due to the otherness of thermal infrared imager and acquisition environment, the present invention is red by power equipment
Outer thermal map is normalized, and can influence to avoid caused by the otherness of thermal infrared imager and acquisition environment.
Extracted in the present invention it is described pretreatment infrared chart in power equipment infrared chart feature, including characteristics of image and
It is VGG16 network technology the present invention is based on existing depth network when heat distribution feature, wherein preceding several layers of convolutional layers in the present invention
Network depth can be increased using only 3 × 3 convolution kernels, successively reduce by every layer of mind by max pooling (maximum pond)
Through first quantity, last three-layer coil lamination is two full connection infrared chart characteristic layers with 4096 neurons respectively, and
The full articulamentum of one 2 value output, is finally Softmax layers.VGG16 network belongs to the prior art, and the present invention not writes.
In order to accelerate overall training and recognition speed, the present invention is by VGG16 from second two in the first convolution module
Tieing up two-dimensional convolution of the convolution between the third two-dimensional convolution into third convolution module is the modularization in Squeeze Net
Convolution module Fire Module, in conjunction with modular convolution (Fire Module) structure in VGG16 and Squeeze Net
It builds to obtain a kind of new network frame, meanwhile, the convolutional layer that the present invention also utilizes disclosed pre-training weights initialisation all will
The infrared chart enhanced after pretreatment, inputs the convolutional layer of depth network, and the result of output is that the power equipment of extraction is red
Outer thermal map feature.It is taken using the power equipment infrared chart feature of said extracted using Nadam optimizer training bottleneck characteristic
Fully-connected network frame is built, and Nadam optimizer is set, using the bottleneck characteristic of extraction to obtain weight.
Using the power equipment infrared chart feature of extraction, freeze the first convolution module, the volume Two of neural network framework
Volume module and third convolution module only carry out Volume Four volume module, the 5th convolution module and the full articulamentum in network
Global optimization training, to update the bottleneck characteristic weight, wherein global optimization training can be carried out using SGD optimizer.
Preferably, the power equipment infrared chart of acquisition to be carried out to the transformation of different scale in the step S1, obtain
Training sample set.There are many mapping modes for carrying out different scale in the present invention to the power equipment infrared chart of acquisition, example
It such as include drawing high and rotating, transformation forms a training sample set abundant.
There are many modes that enhancing processing is carried out in the present invention, using existing a variety of enhancing processing modes, as excellent
It selects, enhancing processing is carried out by Laplace operator in the step S2.
There are many modes being normalized in the present invention, preferably, it is infrared to calculate power equipment in the step S2
Maximum value Rmax, minimum value Rmin and average value Ravg of the thermal map on the channel R, by following equation to training sample set
In all power equipment infrared chart be normalized:
Wherein RtAny pixel of power equipment infrared chart is represented in the numerical value in the channel R.
Preferably, each convolution module Fire Module includes X 1 × 1 convolutional layer and expansion, it is described
Expansion is made of Y 1 × 1 expansions and Z 3 × 3 expansion.
Preferably, carrying out global optimization training by stochastic gradient algorithm SGD optimization in the step S5.
Method of the invention can be applied to identify a variety of power equipments, preferably, the power equipment is insulator
And current transformer etc..
Preferably, in the infrared chart to be identified power equipment identification region be in the lateral rectangle at center or
In the rectangle of person longitudinal direction.Effective identification region is divided to infrared power equipment, since the power equipment that capture apparatus obtains is red
Outer thermal map background rich in and peripheral information, during practical business, power equipment identification that the present invention can require
Region is in the lateral rectangle at center, or in longitudinal rectangle.It simultaneously can be simultaneous with more in Classification and Identification picture
A insulator and other power equipments.
Compared with prior art, the beneficial effects of the present invention are embodied in:
(1) present invention can carry out multi-angled shooting to object to be identified, improve recognition accuracy;
(2) present invention uses analysis and study based on a large amount of power domain equipment infrared charts early period, to infrared electric power
The power equipments such as equipment have carried out automatic identification, effectively reduce the work named after the training cost of Field Force, and shooting
It measures, reduces on-site personnel manual identified power equipment, and be labeled required time and efforts cost.
(3) manual identified, mark and the name relative to a large amount of infrared power equipments, the present invention is based on set to infrared electric power
Standby high discrimination can be used for the automated analysis of the picture name and later period status of electric power and abnormal conditions at scene,
The present invention identifies that the accuracy rate of the power equipments such as infrared power equipment is high.
(4) texture recognition of the present invention and thermal imagery heat distribution identify the algorithm be combineding with each other, and are better than existing list
The accuracy rate of one recognition methods, identification is higher.
Detailed description of the invention
Fig. 1 is the neural network framework schematic diagram based on VGG16 and Squeeze Net in the present invention.
Fig. 2 is the schematic diagram of convolution module Fire Module in the present invention.
Fig. 3 is structural schematic diagram of the invention.
Specific embodiment
Embodiment 1
The present embodiment is a kind of device based on wide-angle automatic identification, as shown in Figure 1, the present embodiment includes rack and machine
Torsion shaft 1 that frame hingedly cooperates, the thermal infrared imager that image is shot, for installing thermal infrared imager and being fixed on shaft
On mounting rack, and the cylinder 2 rotated for drive shaft around hinged place, the mounting rack includes that rotation is arranged in torsion shaft
Rotary shaft 3 on 1 and it is set to 3 one end of rotary shaft and the mounting base 31 for installing thermal infrared imager.
The device of the invention can make thermal infrared imager realize the shooting for carrying out multi-angle to object to be identified, pass through gas
Cylinder 2 drives torsion shaft 1 to overturn and then drives the overturning of infrared image instrument, and rotary shaft 3 can also be driven to rotate, so that infrared heat
As instrument rotation, the driving device of driving thermal infrared imager rotation can be the common structures such as motor in the present invention.
As shown in Figure 1, wherein rack includes that pedestal 5, the vertical rack being set on pedestal 59 and vertical rack 9 slide
The lifting shaft 4 of cooperation is fixedly connected with lifter plate 6 with lifting shaft 4 and for driving lifting shaft 4 to slide relative to vertical rack
Driving mechanism, the torsion shaft 1 hingedly cooperate with lifter plate 6.Sliding slot can be set in the present invention on vertical rack, gone up and down
The sliding block that sliding rail matches is set on axis.
In the present embodiment driving mechanism can there are many, such as shown in Figure 1, driving mechanism includes:
The rack gear 7 being set on lifting shaft 4;
The gear 8 matched with rack gear 7;
Motor for driving gear 8 to rotate.
When motor-driven gear 8 rotates, since gear 8 is matched with rack gear 7, and rack gear 7 is set to lifting shaft 4
On, therefore lifting shaft 4 realizes elevating movement when gear 8 rotates, and then realizes the lifting to thermal infrared imager.The present invention
In can pass through the work that controller controls motor and cylinder 2.
Embodiment 2
A kind of power equipment infrared chart recognition methods, comprising the following steps:
S1: using embodiment 1 dress acquisition it is multiple include power equipment to be identified infrared chart, by the electricity of acquisition
Power equipment infrared chart carries out the transformation of different scale, obtains training sample set.To the power equipment of acquisition in the present embodiment
Infrared chart carries out there are many mapping modes of different scale, and for example including drawing high and rotating, transformation forms one and enriches
Training sample set.
S2: Laplace operator is passed sequentially through to each power equipment infrared chart in training sample set respectively and is carried out
Enhancing processing and normalized obtain pretreatment infrared chart;Wherein, power equipment infrared chart is calculated on the channel R
Maximum value Rmax, minimum value Rmin and average value Ravg, by following equation to electric power all in training sample set
Equipment infrared chart is normalized:
Wherein RtAny pixel of power equipment infrared chart is represented in the numerical value in the channel R.
S3: the power equipment infrared chart textural characteristics and heat distribution feature in pretreatment infrared chart: building are extracted
VGG16 depth network, as depicted in figs. 1 and 2, from second in the first convolution module in VGG16 depth network in the present embodiment
Two-dimensional convolution of a two-dimensional convolution between the third two-dimensional convolution into third convolution module is the mould in Squeeze Net
Convolution module the Fire Module, each convolution module Fire Module of block include X 1 × 1 convolutional layer and expansion
Part is opened up, expansion is made of Y 1 × 1 expansions and Z 3 × 3 expansion, the convolution number in Fire Module
Amount is shown in Table 1, i.e. neural network framework of the acquisition based on VGG16 and Squeeze Net;
Convolution quantity in the depth network architecture of 1 construction of table in Fire Module
Convolution quantity | X(1x1) | Y (1x1, extension) | Z (3x3, extension) |
Fire_1 | 8 | 32 | 32 |
Fire_2 | 16 | 64 | 64 |
Fire_3 | 16 | 64 | 64 |
Fire_4 | 32 | 128 | 128 |
Fire_5 | 32 | 128 | 128 |
Fore_6 | 32 | 128 | 128 |
S4: bottleneck characteristic is saved: input training sample set and test sample set, it is defeated before extracting full articulamentum
Out, that is, bottleneck characteristic is obtained, and uses Nadam optimizer training bottleneck characteristic;
S5: the network optimization: the pre- weight of VGG16 depth network and bottleneck characteristic weight freeze the of neural network framework
One convolution module, the second convolution module and third convolution module, and connect Volume Four volume module, the 5th convolution module and entirely
It connects layer and global optimization training is carried out by stochastic gradient algorithm SGD optimization, obtain power equipment network model, obtain infrared chart
For the probability threshold value of power equipment;
S6: power equipment network model is utilized, inputs or reads infrared chart to be identified, obtain in infrared chart to be identified
Probability including power equipment should be to if the probability in infrared chart to be identified including power equipment is higher than probability threshold value
Identification infrared chart is power equipment.
Power equipment infrared chart is acquired in the present invention first, general power equipment infrared chart size is 512*512, and
And should be comprising power equipment for identification in power equipment infrared chart, and opposite position of the power equipment in infrared chart
It sets and keeps roughly the same, wherein power equipment can be a variety of existing power equipments, such as insulator, current transformer etc..
Power equipment information is obtained using thermal infrared imager, wherein comprising the power equipment classified, the present invention
Enhancing processing is carried out to infrared chart, and due to the otherness of thermal infrared imager and acquisition environment, the present invention sets electric power
Standby infrared chart is normalized, and can influence to avoid caused by the otherness of thermal infrared imager and acquisition environment.
When extracting the power equipment infrared chart feature in pretreatment infrared chart in the present invention, the present invention is based on existing
Depth network is VGG16 network technology, wherein preceding several layers of convolutional layers can increase net using only 3 × 3 convolution kernels in the present invention
Network depth, successively reduces by every layer of neuronal quantity by max pooling (maximum pond), and last three-layer coil lamination is respectively
The full articulamentum of two full connection infrared chart characteristic layers and 2 values output with 4096 neurons, finally for
Softmax layers.Wherein VGG16 network belongs to the prior art, and the present invention not writes.
In order to accelerate overall training and recognition speed, the present invention is by VGG16 from second two in the first convolution module
Tieing up two-dimensional convolution of the convolution between the third two-dimensional convolution into third convolution module is the modularization in Squeeze Net
Convolution module Fire Module, in conjunction with modular convolution (Fire Module) structure in VGG16 and Squeeze Net
It builds to obtain a kind of new network frame, meanwhile, the convolutional layer that the present invention also utilizes disclosed pre-training weights initialisation all will
The infrared chart enhanced after pretreatment, inputs the convolutional layer of depth network, and the result of output is that the power equipment of extraction is red
Outer thermal map feature.It is taken using the power equipment infrared chart feature of said extracted using Nadam optimizer training bottleneck characteristic
Fully-connected network frame is built, and Nadam optimizer is set, using the bottleneck characteristic of extraction to obtain weight.
Using the power equipment infrared chart feature of extraction, freeze the first convolution module, the volume Two of neural network framework
Volume module and third convolution module only carry out Volume Four volume module, the 5th convolution module and the full articulamentum in network
Global optimization training, to update bottleneck characteristic weight, wherein global optimization training can be carried out using sgd optimizer.
Under normal circumstances, power equipment identification region is in the lateral rectangle at center in infrared chart to be identified in the present invention
In the middle or in longitudinal rectangle.Effective identification region is divided to infrared power equipment, the electric power obtained due to capture apparatus
Equipment infrared chart background rich in and peripheral information, during practical business, the electric power that the present invention can require is set
Standby identification region is in the lateral rectangle at center, or in longitudinal rectangle.It simultaneously can be simultaneously in Classification and Identification picture
With multiple power equipments.
Claims (10)
1. a kind of device based on wide-angle automatic identification, which is characterized in that the torsion shaft that hingedly cooperates including rack, with rack,
Thermal infrared imager, the mounting rack for installing thermal infrared imager and being fixed in shaft that image is shot, and be used for
The cylinder that drive shaft is rotated around hinged place, the mounting rack include the rotary shaft and be set to that rotation is arranged on torsion shaft
Rotary shaft one end and the mounting base for being used to install thermal infrared imager.
2. the device as described in claim 1 based on wide-angle automatic identification, which is characterized in that the rack includes pedestal, sets
The vertical rack that is placed on pedestal, is fixedly connected with lifter plate and use with lifting shaft at the lifting shaft being slidably matched with vertical rack
In the driving mechanism that driving lifting shaft is slided relative to vertical rack, the torsion shaft hingedly cooperates with lifter plate.
3. the device as claimed in claim 2 based on wide-angle automatic identification, which is characterized in that the driving mechanism includes:
The rack gear being set on lifting shaft;
The gear matched with rack gear;
Motor for the rotation of sliding tooth wheel.
4. a kind of method based on wide-angle automatic identification, which comprises the steps of:
S1: by any device acquisition based on wide-angle automatic identification of claims 1 to 3 it is multiple include it is to be identified to
The infrared chart of object is surveyed, and carries out the shooting at multiple visual angles, obtains training sample set;
S2: each object under test infrared chart in training sample set is successively carried out respectively at enhancing processing and normalization
Reason obtains pretreatment infrared chart;
S3: extracting the object under test infrared chart feature in the pretreatment infrared chart: building VGG16 depth network, wherein
Third two dimension volume in VGG16 depth network from second two-dimensional convolution in the first convolution module into third convolution module
Two-dimensional convolution between product is the modular convolution module Fire Module in Squeeze Net, that is, is based on
The neural network framework of VGG16 and Squeeze Net;
S4: save bottleneck characteristic: input training sample set and test sample set extract the output before full articulamentum, i.e.,
Bottleneck characteristic is obtained, and uses Nadam optimizer training bottleneck characteristic;
S5: the network optimization: the pre- weight of VGG16 depth network and bottleneck characteristic weight freeze the first volume of neural network framework
Volume module, the second convolution module and third convolution module, and by Volume Four volume module, the 5th convolution module and full articulamentum
Global optimization training is carried out, object under test network model is obtained, obtains the probability threshold value that infrared chart is object under test;
S6: the object under test network model is utilized, inputs or reads infrared chart to be identified, obtain in infrared chart to be identified
Probability including the object under test, if the probability in infrared chart to be identified including the object under test is higher than the probability threshold
When value, then the infrared chart to be identified is the object under test.
5. the method as claimed in claim 4 based on wide-angle automatic identification, which is characterized in that in the step S1, will acquire
Object under test infrared chart carry out different scale transformation, obtain training sample set.
6. the method as claimed in claim 4 based on wide-angle automatic identification, which is characterized in that general by drawing in the step S2
Laplacian operater carries out enhancing processing.
7. the method as claimed in claim 4 based on wide-angle automatic identification, which is characterized in that calculated in the step S2 to be measured
Maximum value Rmax, minimum value Rmin and average value Ravg of the object infrared chart on the channel R, by following equation to training
All object under test infrared charts are normalized in sample set:
Wherein RtAny pixel of object under test infrared chart is represented in the numerical value in the channel R.
8. the method as claimed in claim 4 based on wide-angle automatic identification, which is characterized in that each convolution module Fire
Module includes the convolutional layer and expansion of X 1 × 1, and the expansion is by Y 1 × 1 expansions and Z 3 × 3
Expansion composition.
9. the method as claimed in claim 4 based on wide-angle automatic identification, which is characterized in that by random in the step S5
Gradient algorithm SGD optimization carries out global optimization training.
10. the method as claimed in claim 4 based on wide-angle automatic identification, which is characterized in that the infrared heat of object under test
Figure feature includes the textural characteristics of image and the heat distribution feature of image.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811105825.4A CN109255347A (en) | 2018-09-21 | 2018-09-21 | Method and apparatus based on wide-angle automatic identification |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811105825.4A CN109255347A (en) | 2018-09-21 | 2018-09-21 | Method and apparatus based on wide-angle automatic identification |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109255347A true CN109255347A (en) | 2019-01-22 |
Family
ID=65048246
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811105825.4A Pending CN109255347A (en) | 2018-09-21 | 2018-09-21 | Method and apparatus based on wide-angle automatic identification |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109255347A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109934212A (en) * | 2018-11-29 | 2019-06-25 | 国家电网有限公司 | The method of imaging device and automatic identification image based on wide-angle |
CN110598588A (en) * | 2019-08-27 | 2019-12-20 | 杭州天铂云科光电科技有限公司 | Wide-angle automatic identification method for substation equipment |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN205781809U (en) * | 2016-07-06 | 2016-12-07 | 高戊凡 | A kind of stent-type cam device of band separate microphone |
CN107463054A (en) * | 2017-08-02 | 2017-12-12 | 张艳 | A kind of adjusting means of camera lens |
CN207094114U (en) * | 2017-06-23 | 2018-03-13 | 浙江机电职业技术学院 | Cradle head device for shoot control system |
CN207094131U (en) * | 2017-06-23 | 2018-03-13 | 浙江机电职业技术学院 | A kind of filming apparatus for being easy to multi-angled shooting |
CN207625679U (en) * | 2017-12-28 | 2018-07-17 | 广州华宇维视电子技术有限公司 | A kind of high-definition monitoring equipment being conveniently adjusted orientation |
US20180211117A1 (en) * | 2016-12-20 | 2018-07-26 | Jayant Ratti | On-demand artificial intelligence and roadway stewardship system |
-
2018
- 2018-09-21 CN CN201811105825.4A patent/CN109255347A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN205781809U (en) * | 2016-07-06 | 2016-12-07 | 高戊凡 | A kind of stent-type cam device of band separate microphone |
US20180211117A1 (en) * | 2016-12-20 | 2018-07-26 | Jayant Ratti | On-demand artificial intelligence and roadway stewardship system |
CN207094114U (en) * | 2017-06-23 | 2018-03-13 | 浙江机电职业技术学院 | Cradle head device for shoot control system |
CN207094131U (en) * | 2017-06-23 | 2018-03-13 | 浙江机电职业技术学院 | A kind of filming apparatus for being easy to multi-angled shooting |
CN107463054A (en) * | 2017-08-02 | 2017-12-12 | 张艳 | A kind of adjusting means of camera lens |
CN207625679U (en) * | 2017-12-28 | 2018-07-17 | 广州华宇维视电子技术有限公司 | A kind of high-definition monitoring equipment being conveniently adjusted orientation |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109934212A (en) * | 2018-11-29 | 2019-06-25 | 国家电网有限公司 | The method of imaging device and automatic identification image based on wide-angle |
CN110598588A (en) * | 2019-08-27 | 2019-12-20 | 杭州天铂云科光电科技有限公司 | Wide-angle automatic identification method for substation equipment |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108898077A (en) | The recognition methods of power equipment infrared chart and power equipment infrared recognition system | |
CN110598736B (en) | Power equipment infrared image fault positioning, identifying and predicting method | |
CN112200178B (en) | Transformer substation insulator infrared image detection method based on artificial intelligence | |
CN111784633B (en) | Insulator defect automatic detection algorithm for electric power inspection video | |
CN108686978A (en) | The method for sorting and system of fruit classification and color and luster based on ARM | |
CN109782603A (en) | The detection method and monitoring system of rotating machinery coupling fault | |
CN110378232B (en) | Improved test room examinee position rapid detection method of SSD dual-network | |
CN109255347A (en) | Method and apparatus based on wide-angle automatic identification | |
CN109801265A (en) | A kind of real-time transmission facility foreign matter detection system based on convolutional neural networks | |
CN110009601A (en) | Large-Scale Equipment irregular contour detection method of surface flaw based on HOG | |
CN113205039A (en) | Power equipment fault image identification and disaster investigation system and method based on multiple DCNNs | |
CN112200263A (en) | Self-organizing federal clustering method applied to power distribution internet of things | |
CN116681962A (en) | Power equipment thermal image detection method and system based on improved YOLOv5 | |
CN106846304A (en) | Electrical equipment detection method and device based on infrared detection | |
CN114387558B (en) | Transformer substation monitoring method and system based on multi-dimensional video | |
CN113361686A (en) | Multilayer heterogeneous multi-mode convolutional neural network integrated robot inspection method | |
CN112560896A (en) | Fruit quality screening and classifying system based on image processing | |
CN116388379A (en) | Remote infrared intelligent inspection method and system for transformer substation | |
CN114135991B (en) | Temperature preset control and equipment early warning method for subway station public area | |
CN109934212A (en) | The method of imaging device and automatic identification image based on wide-angle | |
CN108073873A (en) | Human face detection and tracing system based on high-definition intelligent video camera | |
CN206962973U (en) | A kind of centralized control system for substation inspection | |
CN106774190A (en) | Wet cooling gas turbine group condenser vacuum exception auto-check system and method | |
CN114943988B (en) | Planar target monitoring method based on instance segmentation and deep convolution neural network | |
CN113284103B (en) | Substation equipment defect online detection method based on space transformation fast R-CNN model |
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 | ||
AD01 | Patent right deemed abandoned | ||
AD01 | Patent right deemed abandoned |
Effective date of abandoning: 20220902 |