CN109472229A - Shaft tower Bird's Nest detection method based on deep learning - Google Patents
Shaft tower Bird's Nest detection method based on deep learning Download PDFInfo
- Publication number
- CN109472229A CN109472229A CN201811275856.4A CN201811275856A CN109472229A CN 109472229 A CN109472229 A CN 109472229A CN 201811275856 A CN201811275856 A CN 201811275856A CN 109472229 A CN109472229 A CN 109472229A
- Authority
- CN
- China
- Prior art keywords
- nest
- shaft tower
- model
- bird
- deep learning
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- 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
Abstract
The present invention proposes a kind of shaft tower Bird's Nest detection method based on deep learning, the shaft tower Bird's Nest image for acquisition of taking photo by plane is divided using deep neural network frame, electric power line pole tower information is extracted first with deep learning network, deep learning network is recycled to extract the depth characteristic information of Bird's Nest, depth prediction is carried out using extracted characteristic information, each positioning confidence level for dividing region in shaft tower image of taking photo by plane is predicted, region will be finally divided and is merged into shaft tower Bird's Nest target detection rectangle frame.Including using Darknet as frame, deep learning network model is built using yolo v2, the regression forecasting of target rectangle frame is carried out using convolutional neural networks and confidence level is predicted.On this basis, the less disadvantage of deep learning web database technology can also be made up, and allow new data to continue to train on old model using fine-tune in the way of deep learning image enhancement, enhances the feature skewed popularity and generalization ability of model.
Description
Technical field
The invention belongs to electric inspection process failure and unusual condition identification field more particularly to a kind of bars based on deep learning
Tower Bird's Nest detection method.
Background technique
As power transmission network is increasingly huge, the workload rapid growth of power grid operation management, in addition adjoint landform and traffic
Complexity, bring huge difficulty to manual inspection.Novel unmanned plane inspection technology becomes current hot technology,
Unmanned plane is flexible, at low cost in construction of line traction and line data-logging upper type, can not only find the small size portion on route
Part, such as insulator, Bird's Nest, but also it can be found that the manual inspections such as the split pin of fitting burn into and nut missing are difficult to find
The problem of.
The fault detection that unmanned plane is applied to power components judges generally by direct labor after live shooting or will
Data band is diagnosed back into row.But unmanned plane line walking process can generate a large amount of picture, whether artificial judgment equipment breaks down
Necessarily inefficiency and not accurate enough.It is normal during electric inspection process all the time for the Bird's Nest on electric force pole tower
See hidden danger, if timely cannot be checked and be handled, may cause serious fault.
Summary of the invention
The shaft tower Bird's Nest detection method based on deep learning that the purpose of the present invention is to provide a kind of, utilizes depth nerve net
Network frame divides the shaft tower Bird's Nest image for acquisition of taking photo by plane, and extracts electric power line pole tower letter first with deep learning network
Breath recycles deep learning network to extract the depth characteristic information of Bird's Nest, carries out depth prediction using extracted characteristic information,
Each positioning confidence level for dividing region in shaft tower image of taking photo by plane is predicted, region will be finally divided and is merged into shaft tower Bird's Nest target
Detect rectangle frame.Including using Darknet as frame, deep learning network model is built using yolo v2, utilizes convolution mind
Regression forecasting and the confidence level prediction of target rectangle frame are carried out through network.On this basis, deep learning image can also be utilized
The mode of enhancing makes up the less disadvantage of deep learning web database technology, and allows new data old using fine-tune
Continue to train on model, enhances the feature skewed popularity and generalization ability of model.
To achieve the above object, the present invention specifically uses following technical scheme:
A kind of shaft tower Bird's Nest detection method based on deep learning, which comprises the following steps:
Step S1: establishing shaft tower Bird's Nest training image library, makes corresponding label file to each image pattern；The label text
Part meets the xml label file standard of Pascal VOC format；
Step S2: establishing deep learning network model, using based on Darknet network frame and yolo v2 algorithm of target detection
Construct deep learning target detection network；
Step S3: by shaft tower Bird's Nest training image library all image patterns and corresponding label file be divided into training set and
Test set；
Step S4: being trained the training set using yolo v2, obtains initial shaft tower detection model and the detection of initial Bird's Nest
Model；
Step S5: model fine tuning is carried out using fine-tune；
Step S6: using test set test model performance and curing model, shaft tower final mask and Bird's Nest final mask are obtained；
Step S7: testing image successively passes through shaft tower final mask and the detection of Bird's Nest final mask, obtains testing result.
Preferably, in step s 4, the initialization model that training uses is Darknet pre-training model, parameter update side
Formula is SGD, and initial learning rate is 0.003, batch_size 1, and train epochs are 100,000 steps.
Preferably, in step s3, the image pattern in training set is overturn and/or rotated and/or scale and/or
Trimming operation generates new image pattern, and makes corresponding label file, carries out dilatation to training set.
When preferably, in step s 6, using test set test model performance, also model is carried out using fine-tune
Fine tuning.
The present invention and its preferred embodiment realize high detection rate under the premise of efficient, being capable of quick, real-time, effectively identification
Bird's Nest on electric force pole tower can exclude the security risk of power grid to improve the efficiency of electric inspection process in time.
Detailed description of the invention
The present invention is described in more detail with reference to the accompanying drawings and detailed description:
Fig. 1 is present invention method overall flow schematic diagram；
Fig. 2 is yolo v2 algorithm of target detection schematic network structure in the embodiment of the present invention；
Fig. 3 is shaft tower Bird's Nest image data base sample schematic diagram in the embodiment of the present invention；
Fig. 4 is to carry out turning operation to shaft tower Bird's Nest image data base sample in the embodiment of the present invention to realize the signal of dilatation
Figure；
Fig. 5 is shaft tower testing result legend in the embodiment of the present invention；
Fig. 6 is Bird's Nest testing result legend in the embodiment of the present invention；
Fig. 7 is test result part sample in the embodiment of the present invention.
Specific embodiment
For the feature and advantage of this patent can be clearer and more comprehensible, special embodiment below is described in detail below:
As shown in Figure 1, the overall flow of the present embodiment method the following steps are included:
Step S1: establishing shaft tower Bird's Nest training image library, and each image pattern is similar to Figure 3, preferably includes various city back
Shaft tower Bird's Nest image under scape, loess background and mountain forest background, and the shaft tower Bird's Nest image including distant view, close shot and feature,
And keep the shaft tower Bird's Nest amount of images of each type relatively uniform.Secondly, establishing and unmanned plane shaft tower Bird's Nest picture number
According to the corresponding shaft tower Bird's Nest tag database of taking photo by plane in library, every tension rod tower Bird's Nest image has corresponding mark in tag database
Sign file.Label file meets the xml label file standard of Pascal VOC format, including image ID, image path, Image Name
Title, the pixels tall of image and width.Wherein the pixels tall of image and width are by four coordinates of a rectangle frame come table
Show, including xmin, ymin, xmax, ymax.Wherein (xmin, ymin) is the coordinate of the left upper apex of rectangle frame, (xmax,
Ymax) be rectangle frame bottom right vertex coordinate.
Step S2: establishing deep learning network model, using based on Darknet network frame and yolo v2 target detection
Algorithm constructs deep learning target detection network；Darknet is a more light-duty open source depth based entirely on C and CUDA
Learning framework, installation is easy and transplantability is good.Yolo v2 is full convolutional network structure, as shown in Fig. 2, yolo v2 can be applicable in
It is used as input in the picture for not having to size, BN layers therein can be such that trained convergence rate accelerates, and use Faster-Rcnn
In anchor mechanism improve the precision and recall rate of model, and yolo v2 has used the residual block connection structure of resnet,
Wisp can preferably be detected.
Step S3: by shaft tower Bird's Nest training image library all image patterns and corresponding label file be divided into training
Collection and test set.Darknet is trained using VOC data set format, including Annotations, ImageSets,
JPEGImages etc..Wherein JPEGImages stores shaft tower Bird's Nest image data to be trained, and stores in Annotations wait train
Shaft tower Bird's Nest label data, label is corresponding with Bird's Nest image, and ImageSets stores the title of all images.
Step S4: being trained training set using yolo v2, obtains initial shaft tower detection model and the detection of initial Bird's Nest
Model；The initialization model that training uses is Darknet pre-training model, and parameter update mode is SGD, and initial learning rate is
0.003, batch_size 1,100,000 step of train epochs.Initial Bird's Nest detection model is by training set in initial shaft tower detection model
On be trained acquisition.
Step S5: model fine tuning is carried out using fine-tune；Fine-tune is a part of transfer learning, in pre-training
Fine-tune is carried out on the basis of model, is not required to re -training model, to improve efficiency.On the basis of data volume is little,
The effect of fine-tune is preferable.Model after fine-tune is more biased towards the characteristics of image in new data, tests in new data
Test effect on collection is preferable.
Step S6: using test set test model performance and curing model, shaft tower final mask and the final mould of Bird's Nest are obtained
Type；The wherein concrete operations of curing model are: saving the model of training in training process every certain step number.Mould after preservation
Type tests its model performance in test data set, observing and nursing performance with step number variation.Preference pattern performance is relatively stable
Model, the parameters such as its weight, biasing are saved, i.e. curing model.The solidification of model can reduce model volume, facilitate model
Transplanting.
Step S7: testing image successively passes through shaft tower final mask and the detection of Bird's Nest final mask, obtains testing result: will
It is detected in the Tower Model that testing image inputs after solidifying, obtains the shaft tower image detection such as Fig. 5 as a result, by shaft tower image
Testing result figure input Bird's Nest detection model obtains the target rectangle frame of testing result shown in Fig. 6 after network propagated forward
Coordinate and confidence level.The confidence level comprising shaft tower obtains in profile and target detection frame comprising detection shaft tower in rectangle frame
Point, and test result preservation is as shown in Figure 7 at txt file.
It, can be in training set in step S3 or step S6 when training data is inadequate or when new supplementary data is less
Image pattern overturn and/or rotated and/or scaled and/or trimming operation, generate new image pattern, and make correspondence
Label file, to training set carry out dilatation, be illustrated in figure 4 overturning dilatation schematic diagram.Image dilatation can be to a certain degree
The upper accuracy rate for increasing model.
When in step s 6, using test set test model performance, also model is finely adjusted using fine-tune.
This patent is not limited to above-mentioned preferred forms, anyone can obtain other each under the enlightenment of this patent
The shaft tower Bird's Nest detection method based on deep learning of kind of form, all equivalent changes done according to scope of the present invention patent with
Modification, should all belong to the covering scope of this patent.
Claims (4)
1. a kind of shaft tower Bird's Nest detection method based on deep learning, which comprises the following steps:
Step S1: establishing shaft tower Bird's Nest training image library, makes corresponding label file to each image pattern；The label text
Part meets the xml label file standard of Pascal VOC format；
Step S2: establishing deep learning network model, using based on Darknet network frame and yolo v2 algorithm of target detection
Construct deep learning target detection network；
Step S3: by shaft tower Bird's Nest training image library all image patterns and corresponding label file be divided into training set and
Test set；
Step S4: being trained the training set using yolo v2, obtains initial shaft tower detection model and the detection of initial Bird's Nest
Model；
Step S5: model fine tuning is carried out using fine-tune；
Step S6: using test set test model performance and curing model, shaft tower final mask and Bird's Nest final mask are obtained；
Step S7: testing image successively passes through shaft tower final mask and the detection of Bird's Nest final mask, obtains testing result.
2. the shaft tower Bird's Nest detection method according to claim 1 based on deep learning, which is characterized in that in step S4
In, the initialization model that training uses is Darknet pre-training model, and parameter update mode is SGD, and initial learning rate is
0.003, batch_size 1, train epochs are 100,000 steps.
3. the shaft tower Bird's Nest detection method according to claim 1 based on deep learning, which is characterized in that in step S3
In, the image pattern in training set is overturn and/or is rotated and/or is scaled and/or trimming operation, new image sample is generated
This, and corresponding label file is made, dilatation is carried out to training set.
4. the shaft tower Bird's Nest detection method according to claim 1 based on deep learning, which is characterized in that in step S6
In, when using test set test model performance, also model is finely adjusted using fine-tune.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811275856.4A CN109472229A (en) | 2018-10-30 | 2018-10-30 | Shaft tower Bird's Nest detection method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811275856.4A CN109472229A (en) | 2018-10-30 | 2018-10-30 | Shaft tower Bird's Nest detection method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109472229A true CN109472229A (en) | 2019-03-15 |
Family
ID=65666583
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811275856.4A Pending CN109472229A (en) | 2018-10-30 | 2018-10-30 | Shaft tower Bird's Nest detection method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109472229A (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110503623A (en) * | 2019-05-24 | 2019-11-26 | 深圳供电局有限公司 | A method of Bird's Nest defect on the identification transmission line of electricity based on convolutional neural networks |
| CN110569762A (en) * | 2019-08-27 | 2019-12-13 | 许昌许继软件技术有限公司 | pin falling detection method and device based on multistage neural network |
| CN110705542A (en) * | 2019-04-15 | 2020-01-17 | 中国石油大学(华东) | Crane intrusion detection mechanism under power transmission scene based on HDNet |
| CN110706185A (en) * | 2019-09-30 | 2020-01-17 | 上海数禾信息科技有限公司 | Image processing method and device, equipment and storage medium |
| CN111104906A (en) * | 2019-12-19 | 2020-05-05 | 南京工程学院 | Transmission tower bird nest fault detection method based on YOLO |
| CN111191679A (en) * | 2019-12-11 | 2020-05-22 | 中国地质大学（武汉） | Underground non-explosive detection and identification method based on fast R-CNN |
| CN111489354A (en) * | 2020-05-18 | 2020-08-04 | 国网浙江省电力有限公司检修分公司 | Method and device for detecting bird nest on power tower, server and storage medium |
| CN111582072A (en) * | 2020-04-23 | 2020-08-25 | 浙江大学 | Transformer substation picture bird nest detection method combining ResNet50+ FPN + DCN |
| CN112307851A (en) * | 2019-08-02 | 2021-02-02 | 上海交通大学烟台信息技术研究院 | Method and system for identifying bird nest on electric power iron tower |
| CN117253179A (en) * | 2023-11-20 | 2023-12-19 | 合肥中科类脑智能技术有限公司 | Distribution line bird nest detection method, storage medium and electronic equipment |
| CN117392572A (en) * | 2023-12-11 | 2024-01-12 | 四川能投发展股份有限公司 | Transmission tower bird nest detection method based on unmanned aerial vehicle inspection |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106778472A (en) * | 2016-11-17 | 2017-05-31 | 成都通甲优博科技有限责任公司 | The common invader object detection and recognition method in transmission of electricity corridor based on deep learning |
| CN106980827A (en) * | 2017-03-16 | 2017-07-25 | 天津大学 | A kind of method of Bird's Nest in identification transmission line of electricity based on Aerial Images |
| CN108229390A (en) * | 2018-01-02 | 2018-06-29 | 济南中维世纪科技有限公司 | Rapid pedestrian detection method based on deep learning |
| CN108389197A (en) * | 2018-02-26 | 2018-08-10 | 上海赛特斯信息科技股份有限公司 | Transmission line of electricity defect inspection method based on deep learning |
| CN108597053A (en) * | 2018-04-25 | 2018-09-28 | 北京御航智能科技有限公司 | Shaft tower and channel targets identification based on image data and neural network and defect diagnostic method |
| CN108694724A (en) * | 2018-05-11 | 2018-10-23 | 西安天和防务技术股份有限公司 | A kind of long-time method for tracking target |
-
2018
- 2018-10-30 CN CN201811275856.4A patent/CN109472229A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106778472A (en) * | 2016-11-17 | 2017-05-31 | 成都通甲优博科技有限责任公司 | The common invader object detection and recognition method in transmission of electricity corridor based on deep learning |
| CN106980827A (en) * | 2017-03-16 | 2017-07-25 | 天津大学 | A kind of method of Bird's Nest in identification transmission line of electricity based on Aerial Images |
| CN108229390A (en) * | 2018-01-02 | 2018-06-29 | 济南中维世纪科技有限公司 | Rapid pedestrian detection method based on deep learning |
| CN108389197A (en) * | 2018-02-26 | 2018-08-10 | 上海赛特斯信息科技股份有限公司 | Transmission line of electricity defect inspection method based on deep learning |
| CN108597053A (en) * | 2018-04-25 | 2018-09-28 | 北京御航智能科技有限公司 | Shaft tower and channel targets identification based on image data and neural network and defect diagnostic method |
| CN108694724A (en) * | 2018-05-11 | 2018-10-23 | 西安天和防务技术股份有限公司 | A kind of long-time method for tracking target |
Non-Patent Citations (3)
| Title |
|---|
| JOSEPH REDMON.ET AL: ""YOL09000:Better, Faster, Stronger"", 《ARXIV:1612.08242V1》 * |
| JOSEPH REDMON.ET AL: ""You Only Look Once:Unified, Real-Time Object Detection"", 《2016 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 * |
| 杨罡等: ""基于YOLO算法的无人机航拍图片绝缘子缺失检测"", 《科技资讯》 * |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110705542A (en) * | 2019-04-15 | 2020-01-17 | 中国石油大学(华东) | Crane intrusion detection mechanism under power transmission scene based on HDNet |
| CN110503623A (en) * | 2019-05-24 | 2019-11-26 | 深圳供电局有限公司 | A method of Bird's Nest defect on the identification transmission line of electricity based on convolutional neural networks |
| CN112307851A (en) * | 2019-08-02 | 2021-02-02 | 上海交通大学烟台信息技术研究院 | Method and system for identifying bird nest on electric power iron tower |
| CN110569762A (en) * | 2019-08-27 | 2019-12-13 | 许昌许继软件技术有限公司 | pin falling detection method and device based on multistage neural network |
| CN110706185A (en) * | 2019-09-30 | 2020-01-17 | 上海数禾信息科技有限公司 | Image processing method and device, equipment and storage medium |
| CN111191679A (en) * | 2019-12-11 | 2020-05-22 | 中国地质大学（武汉） | Underground non-explosive detection and identification method based on fast R-CNN |
| CN111104906A (en) * | 2019-12-19 | 2020-05-05 | 南京工程学院 | Transmission tower bird nest fault detection method based on YOLO |
| CN111582072A (en) * | 2020-04-23 | 2020-08-25 | 浙江大学 | Transformer substation picture bird nest detection method combining ResNet50+ FPN + DCN |
| CN111489354A (en) * | 2020-05-18 | 2020-08-04 | 国网浙江省电力有限公司检修分公司 | Method and device for detecting bird nest on power tower, server and storage medium |
| CN111489354B (en) * | 2020-05-18 | 2023-07-14 | 国网浙江省电力有限公司检修分公司 | Method and device for detecting bird nest on electric power tower, server and storage medium |
| CN117253179A (en) * | 2023-11-20 | 2023-12-19 | 合肥中科类脑智能技术有限公司 | Distribution line bird nest detection method, storage medium and electronic equipment |
| CN117253179B (en) * | 2023-11-20 | 2024-02-02 | 合肥中科类脑智能技术有限公司 | Distribution line bird nest detection method, storage medium and electronic equipment |
| CN117392572A (en) * | 2023-12-11 | 2024-01-12 | 四川能投发展股份有限公司 | Transmission tower bird nest detection method based on unmanned aerial vehicle inspection |
| CN117392572B (en) * | 2023-12-11 | 2024-02-27 | 四川能投发展股份有限公司 | Transmission tower bird nest detection method based on unmanned aerial vehicle inspection |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109472229A (en) | Shaft tower Bird's Nest detection method based on deep learning | |
| CN104978580B (en) | A kind of insulator recognition methods for unmanned plane inspection transmission line of electricity | |
| CN110245644A (en) | A kind of unmanned plane image transmission tower lodging knowledge method for distinguishing based on deep learning | |
| CN109376606B (en) | Power inspection image tower foundation fault detection method | |
| CN109086826B (en) | Wheat drought identification method based on image deep learning | |
| CN109635875A (en) | A kind of end-to-end network interface detection method based on deep learning | |
| CN109344753A (en) | A kind of tiny fitting recognition methods of Aerial Images transmission line of electricity based on deep learning | |
| CN109643125A (en) | For training the 3D virtual world true to nature of automated driving system to create and simulation | |
| CN108257114A (en) | A kind of transmission facility defect inspection method based on deep learning | |
| CN108229589A (en) | A kind of ground cloud atlas sorting technique based on transfer learning | |
| CN109409263A (en) | A kind of remote sensing image city feature variation detection method based on Siamese convolutional network | |
| CN110200598A (en) | A kind of large-scale plant that raises sign exception birds detection system and detection method | |
| CN109376605A (en) | A kind of electric inspection process image bird-resistant fault detection method | |
| CN108898085A (en) | A kind of road disease intelligent detecting method based on mobile video | |
| CN107808375B (en) | Merge the rice disease image detecting method of a variety of context deep learning models | |
| CN108154102A (en) | A kind of traffic sign recognition method | |
| CN109543630A (en) | Remote sensing image forest land extracting method and system, storage medium, electronic equipment based on deep learning | |
| CN109886947A (en) | The high-tension bus-bar defect inspection method of convolutional neural networks based on region | |
| CN110472575A (en) | A kind of string tomato maturation detection method based on deep learning and computer vision | |
| CN108288269A (en) | Bridge pad disease automatic identifying method based on unmanned plane and convolutional neural networks | |
| CN109584293A (en) | A kind of chamber crop three-dimensional configuration panorama measuring system based on RGB-D vision technique | |
| CN111340060A (en) | Training method of image generator | |
| CN110517228A (en) | Trunk image rapid detection method based on convolutional neural networks and transfer learning | |
| CN110298330A (en) | A kind of detection of transmission line polling robot monocular and localization method | |
| CN115661032A (en) | Intelligent pavement disease detection method suitable for complex background |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190315 |