CN110969192A - Intelligent identification method for power inspection image - Google Patents
Intelligent identification method for power inspection image Download PDFInfo
- Publication number
- CN110969192A CN110969192A CN201911105622.XA CN201911105622A CN110969192A CN 110969192 A CN110969192 A CN 110969192A CN 201911105622 A CN201911105622 A CN 201911105622A CN 110969192 A CN110969192 A CN 110969192A
- Authority
- CN
- China
- Prior art keywords
- layer
- model
- inspection image
- intelligent
- power inspection
- 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
- 238000007689 inspection Methods 0.000 title claims abstract description 35
- 238000000034 method Methods 0.000 title claims abstract description 17
- 239000010410 layer Substances 0.000 claims abstract description 33
- 238000013135 deep learning Methods 0.000 claims abstract description 18
- 238000012360 testing method Methods 0.000 claims abstract description 16
- 238000005516 engineering process Methods 0.000 claims abstract description 11
- 238000004458 analytical method Methods 0.000 claims abstract description 9
- 238000001514 detection method Methods 0.000 claims abstract description 9
- 238000003062 neural network model Methods 0.000 claims abstract description 6
- 230000001133 acceleration Effects 0.000 claims abstract description 5
- 238000005065 mining Methods 0.000 claims abstract description 4
- 239000002356 single layer Substances 0.000 claims abstract description 4
- 238000012549 training Methods 0.000 claims description 24
- 230000007547 defect Effects 0.000 claims description 12
- 238000013527 convolutional neural network Methods 0.000 claims description 10
- 238000012795 verification Methods 0.000 claims description 3
- 230000002787 reinforcement Effects 0.000 claims description 2
- 238000012423 maintenance Methods 0.000 abstract description 6
- 238000012545 processing Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 238000011176 pooling Methods 0.000 description 3
- 239000012212 insulator Substances 0.000 description 2
- 206010063385 Intellectualisation Diseases 0.000 description 1
- 238000009825 accumulation Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 235000005770 birds nest Nutrition 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 239000002131 composite material Substances 0.000 description 1
- 238000005260 corrosion Methods 0.000 description 1
- 230000007797 corrosion Effects 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000003014 reinforcing effect Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 235000005765 wild carrot Nutrition 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
- 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
Abstract
The invention discloses an intelligent identification system for power inspection images, which adopts an image identification algorithm based on deep learning to fuse a plurality of deep convolution neural network models, a main network covers objects with various sizes based on an FPN method, an RPN outputs a thin eigen layer acceleration model to reason, a head uses a single-layer RCNN sub-network to reduce weight and avoid over-fitting, an online difficult sample mining technology is adopted to strengthen the detection capability of a difficult sample, and data enhancement and detailed parameter analysis and comparison tests are realized according to the characteristics of a data set. It has the following advantages: the inspection big data is fully utilized, automatic image identification and classification are achieved, the workload of operation and maintenance personnel is reduced, and the working efficiency is improved.
Description
Technical Field
The invention relates to the field of intelligent power inspection, in particular to an intelligent identification method for power inspection images.
Background
The intelligent processing of the inspection image of the power system has the influence of a plurality of factors, and a plurality of problems need to be overcome, such as the insubstantial characteristics of the self-explosion umbrella skirt of the insulator, the larger difference of the dirty colors of the composite insulator, the similarity between the wire and the broken strand area and the background, the insubstantial corrosion of the pole tower, the smaller area, the too small number of pixels and the occupation ratio of the shockproof hammer area, the similarity between the hay and the bird nest characteristics and the like, so that the field does not have too many technical breakthroughs, and mature tools do not appear. The current electric power inspection image processing technology hardly meets the requirement of detection accuracy, so the characteristics of inspection images are combined, advanced technologies such as artificial intelligence and the like are introduced to carry out deep exploration and attack, the inspection images are subjected to state detection, potential safety hazard points and fault points are highly accurately positioned, and then maintenance teams are guided to quickly carry out line maintenance, the labor intensity of maintenance workers can be reduced, the inspection period is shortened, the operation and maintenance capacity of a power transmission line and a transformer substation is improved, powerful information technology support is provided for guaranteeing stable operation of a power grid, and the intellectualization and management refinement of power grid services are comprehensively improved.
In summary, how to provide an intelligent identification method for power inspection images, which can meet the requirement of inspection image detection accuracy and realize intelligent state monitoring, is one of the problems to be solved urgently by those skilled in the art.
Disclosure of Invention
The invention provides an intelligent identification method for power inspection images, which overcomes the defect that the current power inspection image processing technology in the background technology is difficult to meet the requirement of detection accuracy.
The technical scheme adopted by the invention for solving the technical problems is as follows:
an intelligent identification system for power inspection images adopts an image identification algorithm based on deep learning, integrates multiple deep convolutional neural network models, covers objects with various sizes by a trunk network based on a FPN method, adopts RPN to output thin sign layer acceleration model reasoning, reduces weight by using a single-layer RCNN (recursive neural network) subnetwork at the head, avoids overfitting, adopts an online difficult sample mining technology, strengthens the detection capability of a difficult sample, and realizes data enhancement and detailed parameter analysis and comparison tests according to the characteristics of a data set.
In one embodiment: the intelligent identification system architecture of the power inspection image is divided into an algorithm layer, a model layer, a learning and training layer and a data layer, wherein the algorithm layer adopts a deep learning algorithm, the model layer uses a neural network model base, the learning and training layer comprises a defect identification model reinforcement learning module and a deep learning and training platform, and the data layer comprises an inspection original database, a defect database, an intelligent analysis sample base and a test and verification base.
In one embodiment: after the new data automatically identify the marked hidden danger points through the standard test library, the marking results are verified manually, the marking is standardized, a newly added sample is formed and enters the standard training sample library, the generated data set is provided for the deep learning training platform to form a new model, the new model is marked by the new model through the standard test library to test the new model, the autonomous learning process is completed, and the accuracy of the model is continuously improved.
In one embodiment: and an Mxnet, Caffe, Tensorflow and DarkNet mainstream deep learning framework is fused, and a power inspection image deep learning training platform is constructed.
In one embodiment: the deep convolutional neural network model comprises Zfnet, VGG, Alexnet and ResNet.
Compared with the background technology, the technical scheme has the following advantages:
the system has the intelligent analysis capability of the defects of the inspection images, the automatic classification capability of the images, the screening and labeling capability of training samples, the learning and training capability of defect characteristics and the like. The method is characterized in that a deep network feature learning training system based on data driving is established, closed loops of data updating and model updating are completed, historical data can be fully utilized and data can be fully mined along with the accumulation of time, a closed loop system from defect finding to defect image warehousing is formed, then the closed loop system is updated through model self-learning training, large data in inspection can be fully utilized, automatic image identification and classification are achieved, the workload of operation and maintenance personnel is reduced, and the working efficiency is improved.
Drawings
Fig. 1 is an image recognition autonomous learning process based on deep learning.
Detailed Description
The utility model provides an image intelligent recognition system is patrolled and examined to electric power, adopt the image recognition algorithm based on the degree of depth study, fuse multiple degree of depth convolution neural network model, the image intelligent recognition system architecture is divided into algorithm layer, model layer, study and training layer, data layer, the algorithm layer adopts the degree of depth study algorithm, neural network model base is used in the model layer, study and training layer include defect identification model reinforcing learning module and degree of depth study training platform, the data layer is including patrolling and examining original database, the defect database, intelligent analysis sample storehouse, test and verification storehouse.
Referring to fig. 1, after the new data passes through the standard test library and automatically identifies the marked hidden danger points, the marking result is checked manually, the marking is normalized, a newly added sample is formed and enters the standard training sample library, the generated data set is provided for the deep learning training platform to form a new model, the new model is marked by the new model in the standard test library to test the new model, the autonomous learning process is completed, and the accuracy of the model is continuously improved.
The system integrates Mxnet, Caffe, Tensorflow and DarkNet mainstream deep learning frameworks, and a power inspection image deep learning training platform is constructed.
The system integrates multiple deep convolution neural network models including Zfnet, VGG, Alexnet and ResNet.
Compared with the common detection algorithms such as Faster R-CNN and the like, the image recognition algorithm based on deep learning in the scheme has the following differences: the trunk network covers objects with various sizes based on an FPN method, RPN outputs a thin eigen layer acceleration model to reason, a single-layer RCNN sub-network is used for the head, weight is reduced, overfitting is avoided, an online difficult sample mining technology is adopted, the detection capability of a difficult sample is enhanced, and data enhancement and detailed parameter analysis and comparison tests are realized according to the characteristics of a data set.
The specific implementation process is as follows: (1) the global context information module is introduced, the global context information is widely applied to the classification task, and a full connection layer is used in the classification network finally, so that global features can be simultaneously used for reasoning. The separable large convolution is used for efficiently bringing a larger receptive field for detecting the network, meanwhile, a lightweight global context information module is designed, and the separable large convolution complexity can be controlled more flexibly by introducing a smaller number of intermediate channels. (2) The characteristic layer used for the RoI Pooling is thinned, namely, the characteristic diagram channel used by the RoI Pooling is compressed to be small, on one hand, the channel number of the characteristic diagram is reduced, so that the RoI Pooling efficiency can be improved, on the other hand, the channel number of the input characteristic diagram of the R-CNN part can be reduced, and therefore the purpose of accelerating the R-CNN is achieved. (3) The design using lightweight R-CNN, unlike Faster R-CNN, which uses two strong fully-connected layers to regress and classify each candidate box, in this algorithm, R-CNN uses only one fully-connected layer. By combining the "RoI-pooled thin feature layers", the second stage of the two-stage object detector really achieves low time consumption, thereby achieving overall network acceleration. (4) Aiming at the condition that hidden danger exists, a random cutout technology is used for data enhancement, so that a model can learn hidden danger characteristics under the shielding condition; aiming at the problem of less hidden danger in each picture, the decomposition threshold of the foreground background in the RPN network is reduced to ensure the class balance.
The method combines a powerful inspection image training platform and a power inspection image depth characteristic resource library, can provide comprehensive and efficient inspection image defect automatic analysis and discrimination functions, and can classify a large number of inspection original images according to inspection large targets; and the archiving management of the data set and the training model is enhanced, and the class increase and the data increase are convenient for users. The specific identification types include the following:
the system inputs unmanned aerial vehicles or manual inspection photos, outputs classification results of targets to be detected, has the processing speed of 2 pieces/second on average, and has the identification accuracy rate of more than 90%. The specific identified target species results are as follows:
the above description is only a preferred embodiment of the present invention, and therefore should not be taken as limiting the scope of the invention, which is defined by the appended claims and their equivalents.
Claims (5)
1. The utility model provides an image intelligent recognition system is patrolled and examined to electric power which characterized in that: the method is characterized in that an image recognition algorithm based on deep learning is adopted, a plurality of deep convolutional neural network models are fused, a backbone network is based on an FPN method to cover objects with various sizes, RPN outputs a thin eigen layer acceleration model for reasoning, a single-layer RCNN sub-network is used for a head, weight is reduced, overfitting is avoided, an online difficult sample mining technology is adopted, the detection capability of difficult samples is enhanced, and data enhancement and detailed parameter analysis and comparison tests are realized according to the characteristics of a data set.
2. The intelligent power inspection image identification system according to claim 1, wherein: the intelligent identification system architecture of the power inspection image is divided into an algorithm layer, a model layer, a learning and training layer and a data layer, wherein the algorithm layer adopts a deep learning algorithm, the model layer uses a neural network model base, the learning and training layer comprises a defect identification model reinforcement learning module and a deep learning and training platform, and the data layer comprises an inspection original database, a defect database, an intelligent analysis sample base and a test and verification base.
3. The intelligent power inspection image identification system according to claim 2, wherein: after the new data automatically identify the marked hidden danger points through the standard test library, the marking results are verified manually, the marking is standardized, a newly added sample is formed and enters the standard training sample library, the generated data set is provided for the deep learning training platform to form a new model, the new model is marked by the new model through the standard test library to test the new model, the autonomous learning process is completed, and the accuracy of the model is continuously improved.
4. The intelligent power inspection image identification system according to claim 3, wherein: and an Mxnet, Caffe, Tensorflow and DarkNet mainstream deep learning framework is fused, and a power inspection image deep learning training platform is constructed.
5. The intelligent power inspection image identification system according to claim 3 or 4, wherein: the deep convolutional neural network model comprises Zfnet, VGG, Alexnet and ResNet.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911105622.XA CN110969192A (en) | 2019-11-13 | 2019-11-13 | Intelligent identification method for power inspection image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911105622.XA CN110969192A (en) | 2019-11-13 | 2019-11-13 | Intelligent identification method for power inspection image |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110969192A true CN110969192A (en) | 2020-04-07 |
Family
ID=70030521
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911105622.XA Pending CN110969192A (en) | 2019-11-13 | 2019-11-13 | Intelligent identification method for power inspection image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110969192A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112311092A (en) * | 2020-10-26 | 2021-02-02 | 杭州市电力设计院有限公司余杭分公司 | Method and system for identifying monitoring information of power system |
| CN113052296A (en) * | 2021-03-02 | 2021-06-29 | 贵州电网有限责任公司 | Power grid power transmission defect intelligent management system based on deep learning convolutional neural network technology |
| CN113159166A (en) * | 2021-04-19 | 2021-07-23 | 国网山东省电力公司威海供电公司 | Embedded image identification detection method, system, medium and equipment based on edge calculation |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190147245A1 (en) * | 2017-11-14 | 2019-05-16 | Nuro, Inc. | Three-dimensional object detection for autonomous robotic systems using image proposals |
| CN110008901A (en) * | 2019-04-04 | 2019-07-12 | 天津工业大学 | A kind of insulator breakdown identification and localization method based on Mask R-CNN |
-
2019
- 2019-11-13 CN CN201911105622.XA patent/CN110969192A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190147245A1 (en) * | 2017-11-14 | 2019-05-16 | Nuro, Inc. | Three-dimensional object detection for autonomous robotic systems using image proposals |
| CN110008901A (en) * | 2019-04-04 | 2019-07-12 | 天津工业大学 | A kind of insulator breakdown identification and localization method based on Mask R-CNN |
Non-Patent Citations (3)
| Title |
|---|
| 刘思言 等: "基于R-FCN的航拍巡检图像目标检测方法", 《电力系统自动化》 * |
| 寇大磊 等: "基于深度学习的目标检测框架进展研究", 《计算机工程与应用》 * |
| 汤踊 等: "深度学习在输电线路中部件识别与缺陷检测的研究", 《电子测量技术》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112311092A (en) * | 2020-10-26 | 2021-02-02 | 杭州市电力设计院有限公司余杭分公司 | Method and system for identifying monitoring information of power system |
| CN113052296A (en) * | 2021-03-02 | 2021-06-29 | 贵州电网有限责任公司 | Power grid power transmission defect intelligent management system based on deep learning convolutional neural network technology |
| CN113159166A (en) * | 2021-04-19 | 2021-07-23 | 国网山东省电力公司威海供电公司 | Embedded image identification detection method, system, medium and equipment based on edge calculation |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108037770B (en) | Unmanned aerial vehicle power transmission line inspection system and method based on artificial intelligence | |
| Li et al. | Automatic safety helmet wearing detection | |
| US9251425B2 (en) | Object retrieval in video data using complementary detectors | |
| CN110969192A (en) | Intelligent identification method for power inspection image | |
| Gyawali et al. | Comparative analysis of multiple deep CNN models for waste classification | |
| Adou et al. | Insulator faults detection based on deep learning | |
| Lin et al. | Detection of strawberry flowers in outdoor field by deep neural network | |
| Cepni et al. | Vehicle detection using different deep learning algorithms from image sequence | |
| CN110992307A (en) | Insulator positioning and identifying method and device based on YOLO | |
| CN109815798A (en) | Unmanned plane image processing method and system | |
| CN113128476A (en) | Low-power consumption real-time helmet detection method based on computer vision target detection | |
| CN116416613A (en) | Citrus fruit identification method and system based on improved YOLO v7 | |
| CN116052082A (en) | Power distribution station room anomaly detection method and device based on deep learning algorithm | |
| CN115393347A (en) | Intelligent power grid inspection method and system based on urban brain | |
| Rahman et al. | Identification of mature grape bunches using image processing and computational intelligence methods | |
| CN116739963A (en) | Power grid equipment defect detection method based on multi-level multi-scale feature fusion | |
| Chourasia et al. | Safety helmet detection: a comparative analysis using YOLOv4, YOLOv5, and YOLOv7 | |
| Wang et al. | Automatic Target Recognition using multiple-aspect sonar images | |
| CN114724002A (en) | Method for rapidly detecting contraband | |
| CN113191273A (en) | Oil field well site video target detection and identification method and system based on neural network | |
| CN109003254B (en) | Method, device, equipment, system and medium for detecting icing based on logistic regression | |
| Zhan | Electric Equipment Inspection on High Voltage Transmission Line Via Mobile Net-SSD | |
| Na et al. | Multi-attention YOLOv5 for Recognition of Abnormal Helmets Wearing in Power Stations | |
| Zhang et al. | A Novel Intelligent Recognition of Video Objects in Drone Routing Inspection based on Deep Learning | |
| CN114898291B (en) | Visual monitoring method based on camera inspection path |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20200407 |
|
| WD01 | Invention patent application deemed withdrawn after publication |