CN113963277A - Aerial photography-based railway power through line hidden danger target detection method - Google Patents
Aerial photography-based railway power through line hidden danger target detection method Download PDFInfo
- Publication number
- CN113963277A CN113963277A CN202111245053.6A CN202111245053A CN113963277A CN 113963277 A CN113963277 A CN 113963277A CN 202111245053 A CN202111245053 A CN 202111245053A CN 113963277 A CN113963277 A CN 113963277A
- Authority
- CN
- China
- Prior art keywords
- target
- line
- railway power
- algorithm
- aerial photography
- 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
- 238000001514 detection method Methods 0.000 title claims description 39
- 238000007689 inspection Methods 0.000 claims abstract description 29
- 238000000034 method Methods 0.000 claims abstract description 28
- 238000010276 construction Methods 0.000 claims abstract description 24
- 238000013527 convolutional neural network Methods 0.000 claims description 24
- 238000012360 testing method Methods 0.000 claims description 17
- 238000012549 training Methods 0.000 claims description 17
- 238000013135 deep learning Methods 0.000 claims description 13
- 238000000605 extraction Methods 0.000 claims description 13
- 238000002372 labelling Methods 0.000 claims description 11
- 230000008569 process Effects 0.000 claims description 9
- 230000003213 activating effect Effects 0.000 claims description 6
- 238000005516 engineering process Methods 0.000 claims description 6
- 230000006870 function Effects 0.000 claims description 6
- 238000011161 development Methods 0.000 claims description 5
- 238000013528 artificial neural network Methods 0.000 claims description 4
- 238000013136 deep learning model Methods 0.000 claims description 4
- 238000013507 mapping Methods 0.000 claims description 4
- 230000006872 improvement Effects 0.000 claims description 3
- 238000004891 communication Methods 0.000 description 7
- 238000012423 maintenance Methods 0.000 description 6
- 238000013473 artificial intelligence Methods 0.000 description 4
- 239000013307 optical fiber Substances 0.000 description 3
- 230000003287 optical effect Effects 0.000 description 2
- 230000002093 peripheral effect Effects 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000009313 farming Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000006855 networking Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 230000009466 transformation Effects 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
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0635—Risk analysis of enterprise or organisation activities
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/40—Business processes related to the transportation industry
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Strategic Management (AREA)
- General Health & Medical Sciences (AREA)
- Economics (AREA)
- Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- Mathematical Physics (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Business, Economics & Management (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Entrepreneurship & Innovation (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Game Theory and Decision Science (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Educational Administration (AREA)
- Development Economics (AREA)
- Probability & Statistics with Applications (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Primary Health Care (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a method for detecting a hidden danger target of a railway power through line based on aerial photography, which comprises the following steps: s1, identifying the target; s2, marking the identified picture target; s3: detecting the marked target; s4: the invention relates to a railway power through line inspection method, which comprises the steps of identifying a construction site in an aerial photo of an unmanned aerial vehicle to quickly mark potential safety hazard information around a power line, introducing the unmanned aerial vehicle into a railway power through line for inspection, combining an artificial intelligent identification scheme aiming at the safety of the railway power through line, and realizing an intelligent inspection technical method of the railway power through line by quickly marking the potential safety hazard information around the railway power through line by identifying the construction site in the aerial photo of the unmanned aerial vehicle so as to replace most of manual complicated work and improve inspection efficiency.
Description
Technical Field
The invention belongs to the technical field of target detection of potential damage hazards of railway trunk lines, and particularly relates to a target detection method of potential hazards of a railway electric power through line based on aerial photography.
Background
With the development of information technology, the railway industry in China basically and comprehensively realizes optical fiber communication on the laying of communication cables, the communication quality is rapidly improved, but with the development of the railway industry, the laying length and the laying range of the optical fiber cables are also continuously improved, and the workload of maintenance personnel is increased. Because a lot of optical cables are newly added, constructors cannot mark in time, errors can often occur in communication of some construction units, the situation that cables are damaged by digging appears, and the situation that the optical cables laid in the early stage are aged under the action of time can cause fire hazards and other hidden dangers.
The railway industry in China still has a lot of problems in laying optical fiber cables, mainly in the selection of cable positions. Generally, cables are laid near railways, and railway maintenance personnel are very easy to dig and damage the cables during construction operation, for example, when the cables are laid near highways, although maintenance is facilitated, the safety of the cables is affected due to highway construction, house construction and the like, and difficulty is brought to maintenance work of communication lines.
The traditional manual tour has lower efficiency, poorer quality and lack of timeliness. In contrast, a method combining manual patrol and vehicle patrol is used for patrol of the communication line, and the patrol is performed manually in areas with complex road conditions and difficult vehicle passing; and the method of vehicle inspection can be adopted in the areas with good road condition and wide vision, and inspection supervisors need to be added no matter which method, so that the inspection working condition of maintenance personnel is supervised and recorded, and the inspection strength and quality are promoted to be increased for the maintenance personnel. The above methods all need to invest more manpower and material resources, and the potential safety hazard is still difficult to find in time in areas with poor road conditions.
Therefore, an artificial intelligence recognition scheme based on deep learning is provided, the aim is to quickly mark potential safety hazard information around a power line by recognizing a construction site in an aerial photo of an unmanned aerial vehicle, and an intelligent inspection technical method of the power through line is realized, so that most of manual complicated work is replaced, and the aerial-photo-based railway power through line potential target detection method for improving inspection efficiency is used for solving the problems.
Disclosure of Invention
The invention aims to provide a railway power through line hidden danger target detection method based on aerial photography, and aims to solve the problems in the background technology.
In order to achieve the purpose, the invention provides the following technical scheme: a railway power through line hidden danger target detection method based on aerial photography comprises the following steps:
s1, identifying the target: detecting a construction site which can damage a cable trunk line by adopting an aerial photography technology in an aerial photography graph, wherein the characteristic information of the construction site comprises a temporary building, vegetation damage and construction vehicles;
s2, marking the identified picture target: selecting labelImg software as labeling software, wherein the labeling information of the target needs to be accurate, the minimum circumscribed rectangle frame of the target is selected, and the category information of the target needs to be correct; whether the labeled information reaches the standard is checked in a manual detection mode, finally, labeled data are trained, and a training data set is used as a deep learning model;
s3: detecting the marked target: detecting the marked target by adopting a target detection algorithm based on a deep neural network, wherein the object detection development is concentrated on a two-stage algorithm and a one-stage algorithm, the two-stage algorithm divides the whole into two parts to generate a candidate frame and an object in an identification frame, the two-stage algorithm comprises R-CNN, SPP Net, Fast R-CNN and Fast RCNN, the one-stage algorithm unifies the whole flow and directly gives a detection result, and the one-stage algorithm comprises SSD and YOLO;
adopting fast R-CNN to detect the marked target in the two-stage algorithm, wherein the detection process comprises the following steps:
(a) inputting a test image;
(b) inputting the whole picture into CNN for feature extraction;
(c) generating suggestion windows by using an RPN (resilient packet network), and generating 300 suggestion windows for each picture;
(d) mapping the suggestion window to the last layer convolution feature map of the CNN;
(e) enabling each RoI to generate a feature map with a fixed size through a RoI posing layer;
(f) performing regression joint training on the classification probability and the frame by utilizing Softmax Loss and Smooth L1 Loss;
the labeled target is detected by using YOLO V4 in a one-stage algorithm, and YOLO V4 is improved compared with YOLO V3 by the following points:
(a) the trunk feature extraction network is changed from Darknet53 to CPSDarknet 53;
(b) the improvement of a characteristic extraction network is enhanced, and SPP and PANET structures are used;
(c) the data enhancement uses Mosaic;
(d) the activating function uses a hash activating function;
s4: and (4) analyzing results: and analyzing the performance of the fast RCNN algorithm model and the performance of the YOLO V4 algorithm model and the training and testing time according to the precision ratio, the recall ratio and the workload test criterion, and taking the YOLO V4 as an optimal detection algorithm.
The recall ratio represents how many targets are identified, the precision ratio represents how many targets are identified by the AI, the workload represents the comparison of workload of applying deep learning and only manually identifying the inspection photos without applying deep learning.
The algorithm formula of precision ratio, recall ratio and workload is as follows:
precision = AI identification correct number/AI identification hidden danger number;
recall = AI identification correct number/number of target photos;
workload = AI identification number/total number of photographs.
The label img software is open source free software, is used for deep learning data labeling work, and can generate an xml format or txt format labeling file.
And in the process of manually checking and marking whether the marking information reaches the standard, checking by adopting a random extraction mode.
The deep learning training data set is a database established with a data size of 4500, 4000 training sets and 500 testing sets, and each photo is 6000 x 4000.
The invention has the technical effects and advantages that: this railway electric power through line hidden danger target detection method based on take photo by plane, the peripheral potential safety hazard information of electric power line is mark out fast through the construction place in the discernment unmanned aerial vehicle aerial photo, introduce railway electric power through line and patrol and examine unmanned aerial vehicle, combine to the safe artificial intelligence identification scheme of railway electric power through line and be can be real effectual promotion operating efficiency, optimize railway electric power through line and patrol and examine the working method, the peripheral potential safety hazard information of electric power through line of mark out fast through the construction place in the discernment unmanned aerial vehicle aerial photo, the intelligent of railway electric power through line that realizes patrols and examines the technical method, in order to replace most artifical numerous and diverse work, improve and patrol and examine efficiency.
Drawings
FIG. 1 is a comparison spectrum of the detection data result of the invention based on two algorithms, namely, the master RCNN and the YOLO V4.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In recent years, the unmanned aerial vehicle technology is developed and matured continuously, and industrial-grade large-range unmanned aerial vehicles are widely applied to the fields of emergency, security and the like. Unmanned aerial vehicle has the speed fast, advantages such as reconnaissance scope is wide, also there is the operation simultaneously more complicated, control the distance limited not enough, to current unmanned aerial vehicle's weak point, through the networking transformation to unmanned aerial vehicle, reduce and use the degree of difficulty, enlarge the operation scope, utilize industrial unmanned aerial vehicle as flight platform, combine ripe 4G/5G communication network and artificial intelligence technique to carry out the work of patrolling and examining of railway electric power through line, the target is reduction artifical intensity of labour, greatly improve work efficiency.
At present, a certain theoretical basis exists in the aspect of intelligent routing inspection of a power line, but similar application does not exist in the railway industry, particularly in the aspect of routing inspection of a power through line, and the unmanned aerial vehicle is utilized for routing inspection, so that the method is blank in domestic production application.
The unmanned aerial vehicle inspection system mainly aims at the current inspection working situation, is based on introducing an unmanned aerial vehicle into an electric power line for inspection through technical innovation, and is combined with artificial intelligent identification aiming at the safety of the electric power line to research a set of novel intelligent inspection working means so as to improve the working efficiency and lay a foundation for optimizing the working method. The invention relates to an artificial intelligence recognition scheme based on deep learning, aiming at quickly marking potential safety hazard information around a power line by recognizing a construction site in an aerial photo of an unmanned aerial vehicle, and realizing an intelligent inspection technical method of a power through line so as to replace most of complicated manual work and improve inspection efficiency.
A railway power through line hidden danger target detection method based on aerial photography comprises the following steps:
s1, identifying the target: the method comprises the steps that an aerial photography technology is adopted, a construction site which can damage a cable trunk line in an aerial photography graph is detected, and characteristic information of the construction site comprises a temporary building house, vegetation damage and construction vehicles, wherein the characteristics of the temporary building house do not necessarily exist in small-sized engineering and are similar to factory buildings such as rural suburbs, the vegetation damage characteristic is easy to be confused with normal farming, so that the two characteristics are not available, the construction vehicles are an indispensable part of each construction site, and most of the construction sites which cause the damage of the cable trunk line are sites using large-scale construction machinery such as excavators and bulldozers, so that the excavators and the bulldozers around the cable trunk line are used as marking targets for identifying hidden danger information;
s2, marking the identified picture target:
s2.1, selecting a marking tool: the label software is LabelImg software which is open source free software and is used for deep learning data labeling work, and label files in an xml format or a txt format can be generated;
s2.2, determining the labeling requirement:
the marking information of the target needs to be accurate in frame line, the minimum circumscribed rectangle frame of the target is selected, and the category information of the target needs to be correct;
s2.3, checking the labeling information:
whether the labeling information reaches the standard is checked in a manual detection mode, and in the labeling process of manual detection, a random extraction mode is adopted for checking;
s2.4, determining the number of labels:
training the labeled data to establish a database with the data size of 4500, wherein 4000 are training sets and 500 are testing sets, and each photo is 6000 x 4000 in size;
s3: detecting the marked target: detecting the marked target by adopting a target detection algorithm based on a deep neural network, wherein the object detection development is concentrated on a two-stage algorithm and a one-stage algorithm, the two-stage algorithm divides the whole into two parts to generate a candidate frame and an object in an identification frame, the two-stage algorithm comprises R-CNN, SPP Net, Fast R-CNN and Fast RCNN, the one-stage algorithm unifies the whole flow and directly gives a detection result, and the one-stage algorithm comprises SSD and YOLO;
adopting fast R-CNN to detect the marked target in the two-stage algorithm, wherein the detection process comprises the following steps: (1) inputting a test image, (2) inputting the whole image into a CNN (CNN) for feature extraction; (3) generating suggestion windows (popsals) by using RPN, and generating 300 suggestion windows for each picture; (4) mapping the suggestion window to the last layer convolution feature map of the CNN; (5) enabling each RoI to generate a feature map with a fixed size through a RoI posing layer; (6) jointly training the classification probability and Bounding box regression (Bounding box regression) by utilizing Softmax Loss and Smooth L1 Loss;
the Faster R-CNN discards an external candidate region recommendation algorithm, provides an RPN network structure, and integrates a candidate region recommendation process into a neural network, so that all processes in target detection are unified into one network, complete end-to-end training can be performed, and as can be seen from Table 1, the speed and the precision of detection are improved from the R-CNN to the Faster RCNN;
TABLE 1
Adopting fast R-CNN to detect the marked target in the two-stage algorithm, wherein the detection process comprises the following steps:
(a) inputting a test image;
(b) inputting the whole picture into CNN for feature extraction;
(c) generating suggestion windows by using an RPN (resilient packet network), and generating 300 suggestion windows for each picture;
(d) mapping the suggestion window to the last layer convolution feature map of the CNN;
(e) enabling each RoI to generate a feature map with a fixed size through a RoI posing layer;
(f) performing regression joint training on the classification probability and the frame by utilizing Softmax Loss and Smooth L1 Loss;
the difference of the fast R-CNN lies in that RPN (region pro positive network) is used for replacing the original Selective Search method to generate a suggestion window, and the CNN for generating the suggestion window is shared with the CNN for target detection;
the labeled target is detected by using YOLO V4 in a one-stage algorithm, and YOLO V4 is improved compared with YOLO V3 by the following points:
(a) the trunk feature extraction network is changed from Darknet53 to CPSDarknet 53;
(b) the improvement of a characteristic extraction network is enhanced, and SPP and PANET structures are used;
(c) the data enhancement uses Mosaic;
(d) the activating function uses a hash activating function;
s4: and (4) analyzing results: analyzing the performance of the two algorithm models of the master RCNN and the YOLO V4 and the speed of training and testing time through the precision ratio, the recall ratio and the workload test criterion, and taking the YOLO V4 as an optimal detection algorithm;
the recall ratio represents how many targets are identified, the precision ratio represents how many targets are identified in AI, the workload represents the comparison of workload of applying deep learning and only manually identifying the inspection photos without applying deep learning, and the algorithm formulas of the precision ratio, the recall ratio and the workload are as follows:
precision = AI identification correct number/AI identification hidden danger number;
recall = AI identification correct number/number of target photos;
workload = AI identification number/total number of photographs.
Two algorithm models, namely, the faster RCNN and the YOLO V4, are trained for two types of targets, namely, an excavator target and a bulldozer target, and as shown in Table 2, the test results of a test set are tested by taking precision ratio, recall ratio and workload as test criteria:
TABLE 2
Please refer to fig. 1, which shows the analysis by plotting the data in table 2, and it can be seen from table 2 and fig. 1 that the workload of the railway trunk inspection can be greatly reduced and the working efficiency can be improved after applying the deep learning model, and the performance of YOLO V4 is better than that of the faster RCNN from the inspection rate; from the perspective of precision, the performance of the master RCNN is better than that of the YOLO V4; from the time of training and testing, the YOLO V4 with the recognition step one-step in place was faster than the fast RCNN with the two-step walk.
In practical application, the line inspection should take on the principle of discovering potential safety hazards as much as possible and solving the potential safety hazards in time, so the practical application value of the recall ratio is higher than the precision ratio. In conclusion, we choose to use the YOLO V4 model on the final deep learning model.
Therefore, the unmanned aerial vehicle is introduced into the railway power through line for inspection, the working efficiency can be really and effectively improved and the railway power through line inspection working method can be optimized by combining the artificial intelligent identification scheme aiming at the safety of the railway power through line, the potential safety hazard information around the railway power through line is quickly marked by identifying the construction site in the aerial photo of the unmanned aerial vehicle, and the realized intelligent inspection technical method of the railway power through line is used for replacing most of manual complicated work and improving the inspection efficiency.
Finally, it should be noted that: although the present invention has been described in detail with reference to the foregoing embodiments, it will be apparent to those skilled in the art that modifications may be made to the embodiments or portions thereof without departing from the spirit and scope of the invention.
Claims (6)
1. A railway power through line hidden danger target detection method based on aerial photography is characterized by comprising the following steps: the method comprises the following steps:
s1, identifying the target: detecting a construction site which can damage a cable trunk line by adopting an aerial photography technology in an aerial photography graph, wherein the characteristic information of the construction site comprises a temporary building, vegetation damage and construction vehicles;
s2, marking the identified picture target: selecting labelImg software as labeling software, wherein the labeling information of the target needs to be accurate, the minimum circumscribed rectangle frame of the target is selected, and the category information of the target needs to be correct; whether the labeled information reaches the standard is checked in a manual detection mode, finally, labeled data are trained, and a training data set is used as a deep learning model;
s3: detecting the marked target: detecting the marked target by adopting a target detection algorithm based on a deep neural network, wherein the object detection development is concentrated on a two-stage algorithm and a one-stage algorithm, the two-stage algorithm divides the whole into two parts to generate a candidate frame and an object in an identification frame, the two-stage algorithm comprises R-CNN, SPP Net, Fast R-CNN and Fast RCNN, the one-stage algorithm unifies the whole flow and directly gives a detection result, and the one-stage algorithm comprises SSD and YOLO;
adopting fast R-CNN to detect the marked target in the two-stage algorithm, wherein the detection process comprises the following steps: inputting a test image; then, inputting the whole picture into CNN for feature extraction; generating suggestion windows by using the RPN, and generating 300 suggestion windows by each picture; then mapping the suggestion window to the last convolution feature map of the CNN, and enabling each RoI to generate a fixed-size feature map through a RoI posing layer; finally, performing regression joint training on the classification probability and the frame by utilizing Softmax Loss and Smooth L1 Loss;
the labeled target is detected by using YOLO V4 in a one-stage algorithm, and YOLO V4 is improved compared with YOLO V3 by the following points: the trunk feature extraction network is changed from Darknet53 to CPSDarknet 53; the improvement of a characteristic extraction network is enhanced, and SPP and PANET structures are used; the data enhancement uses Mosaic; the activating function uses a hash activating function;
s4: and (4) analyzing results: and analyzing the performance of the fast RCNN algorithm model and the performance of the YOLO V4 algorithm model and the training and testing time according to the precision ratio, the recall ratio and the workload test criterion, and taking the YOLO V4 as an optimal detection algorithm.
2. The aerial photography-based target detection method for hidden dangers of the railway power through line, according to claim 1, is characterized in that: in step S4, the recall ratio represents how many objects are recognized, the precision ratio represents how many objects are recognized by the AI, and the workload represents comparison between application of deep learning and then application of deep learning, and only manual recognition of the inspection photo.
3. The aerial photography-based target detection method for hidden dangers of the railway power through line, according to claim 1, is characterized in that: the algorithm formula of precision ratio, recall ratio and workload in step S4 is as follows:
precision = AI identification correct number/AI identification hidden danger number;
recall = AI identification correct number/number of target photos;
workload = AI identification number/total number of photographs.
4. The aerial photography-based target detection method for hidden dangers of the railway power through line, according to claim 1, is characterized in that: the labelImg software in the step S2 is open source free software, is used for deep learning data labeling work, and can generate an xml-format or txt-format labeling file.
5. The aerial photography-based target detection method for hidden dangers of the railway power through line, according to claim 1, is characterized in that: in the step S2, in the process of manually checking and marking whether the marking information reaches the standard, a random extraction method is adopted for checking.
6. The aerial photography-based target detection method for hidden dangers of the railway power through line, according to claim 1, is characterized in that: the deep learning training data set in step S2 is a database with 4500 data sizes, wherein 4000 training sets, 500 testing sets and 6000 x 4000 photos are created.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111245053.6A CN113963277A (en) | 2021-10-26 | 2021-10-26 | Aerial photography-based railway power through line hidden danger target detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111245053.6A CN113963277A (en) | 2021-10-26 | 2021-10-26 | Aerial photography-based railway power through line hidden danger target detection method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113963277A true CN113963277A (en) | 2022-01-21 |
Family
ID=79466932
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111245053.6A Pending CN113963277A (en) | 2021-10-26 | 2021-10-26 | Aerial photography-based railway power through line hidden danger target detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113963277A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115373416A (en) * | 2022-08-08 | 2022-11-22 | 北京中润惠通科技发展有限公司 | Intelligent inspection method for railway power through line |
| CN115512098A (en) * | 2022-09-26 | 2022-12-23 | 重庆大学 | Electronic bridge inspection system and inspection method |
| CN117876800A (en) * | 2024-03-11 | 2024-04-12 | 成都千嘉科技股份有限公司 | Method for identifying potential safety hazard of flue of gas water heater |
-
2021
- 2021-10-26 CN CN202111245053.6A patent/CN113963277A/en active Pending
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115373416A (en) * | 2022-08-08 | 2022-11-22 | 北京中润惠通科技发展有限公司 | Intelligent inspection method for railway power through line |
| CN115512098A (en) * | 2022-09-26 | 2022-12-23 | 重庆大学 | Electronic bridge inspection system and inspection method |
| CN115512098B (en) * | 2022-09-26 | 2023-09-01 | 重庆大学 | Bridge electronic inspection system and inspection method |
| CN117876800A (en) * | 2024-03-11 | 2024-04-12 | 成都千嘉科技股份有限公司 | Method for identifying potential safety hazard of flue of gas water heater |
| CN117876800B (en) * | 2024-03-11 | 2024-05-17 | 成都千嘉科技股份有限公司 | Method for identifying potential safety hazard of flue of gas water heater |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113963277A (en) | Aerial photography-based railway power through line hidden danger target detection method | |
| CN111696075A (en) | Intelligent fan blade defect detection method based on double-spectrum image | |
| US20220309772A1 (en) | Human activity recognition fusion method and system for ecological conservation redline | |
| CN110544293B (en) | Building scene recognition method through visual cooperation of multiple unmanned aerial vehicles | |
| CN109711099B (en) | BIM automatic modeling system based on image recognition machine learning | |
| CN108022235A (en) | High-voltage power transmission tower critical component defect identification method | |
| CN111626092B (en) | Unmanned aerial vehicle image ground crack identification and extraction method based on machine learning | |
| AU2020102181A4 (en) | An intelligent recognition system and method of tunnel structure health based on robot vision recognition | |
| CN112149522A (en) | Intelligent visual external-damage-prevention monitoring system and method for cable channel | |
| CN112367400B (en) | Intelligent inspection method and system for power internet of things with edge cloud coordination | |
| CN109344853A (en) | A kind of the intelligent cloud plateform system and operating method of customizable algorithm of target detection | |
| US20230368354A1 (en) | Fault detection method and system for tunnel dome lights based on improved localization loss function | |
| CN104463242A (en) | Multi-feature motion recognition method based on feature transformation and dictionary study | |
| CN114863118A (en) | Self-learning identification system and method based on external hidden danger of power transmission line | |
| CN116846059A (en) | Edge detection system for power grid inspection and monitoring | |
| CN112613453A (en) | Method and system for checking violation of regulations on construction site of electric power infrastructure | |
| CN115311241A (en) | Coal mine down-hole person detection method based on image fusion and feature enhancement | |
| CN115984263A (en) | Bolt looseness detection algorithm and detection system based on improved twin neural network | |
| CN110490261B (en) | Positioning method for power transmission line inspection image insulator | |
| CN115830533A (en) | Helmet wearing detection method based on K-means clustering improved YOLOv5 algorithm | |
| CN116700290B (en) | Intelligent trolley positioning control system and method based on UWB | |
| CN117496426A (en) | Precast beam procedure identification method and device based on mutual learning | |
| CN105354591A (en) | High-order category-related prior knowledge based three-dimensional outdoor scene semantic segmentation system | |
| CN117423157A (en) | Mine abnormal video action understanding method combining migration learning and regional invasion | |
| CN112465072B (en) | Excavator image recognition method based on YOLOv4 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 |