CN110136098A - A kind of order of cables detection method based on deep learning - Google Patents
A kind of order of cables detection method based on deep learning Download PDFInfo
- Publication number
- CN110136098A CN110136098A CN201910299722.4A CN201910299722A CN110136098A CN 110136098 A CN110136098 A CN 110136098A CN 201910299722 A CN201910299722 A CN 201910299722A CN 110136098 A CN110136098 A CN 110136098A
- Authority
- CN
- China
- Prior art keywords
- cable
- order
- detection
- cables
- method based
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10024—Color image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20021—Dividing image into blocks, subimages or windows
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The order of cables detection method based on deep learning that the invention discloses a kind of, comprising the following steps: step 1: carrying out gray processing to cable image first, does basis for subsequent order detection；Step 2: calculus of differences being carried out to the neighbor pixel gray value of image a line, a threshold value is chosen and difference result compares, export positioning result figure.Step 3: the characteristics of being directed to three colo(u)r streak cable images proposes the anchor frame of simplified feature extraction network and optimization, ELU activation primitive improves Faster R-CNN；Step 4: order of cables detection being carried out using improved Faster R-CNN algorithm of target detection, and counts accuracy rate and detection time.The method of the present invention can more fully extract the feature of three colo(u)r streak cables, have many advantages, such as easy to operate, high-efficient, saves many manpower consumptions, used innovatory algorithm can reduce the detection time of cable simultaneously, reduce erroneous detection, false retrieval and the accuracy rate for greatly improving cable detection.
Description
Technical field
The present invention relates to field of image processings, specifically design a kind of order of cables detection method based on deep learning.
Background technique
Existing order of cables detection is all to need many artificial detections.Take this mode that can consume a large amount of manpower
Resource manually carries out checking the efficiency that will affect detection, with high costs, and be easy to cause mistake using high-intensitive manual inspection
Inspection, false retrieval are not able to satisfy the demand of nowadays industrial automation production；Existing cable detection algorithm lacks versatility, detection knot
Fruit is unsatisfactory, practicability be not it is very high, affect to the product quality and production efficiency of enterprise.
Summary of the invention
For the deficiency in above-mentioned background technique, the present invention is to solve existing technical problem and provide a kind of quick, essence
The quasi- order of cables detection method based on deep learning.
Technical scheme is as follows: a kind of order of cables detection method based on deep learning includes the following steps:
Step 1: gray processing being carried out to three colo(u)r streak cable images first, makees basis for subsequent order detection；Step 2: to image
The neighbor pixel gray value of certain a line carries out calculus of differences, chooses a threshold value and difference result compares, and exports positioning result
Figure；Step 3: the characteristics of being directed to three colo(u)r streak cable images proposes anchor frame, the ELU activation of simplified feature extraction network and optimization
Function improves Faster R-CNN；Step 4: improved Faster R-CNN algorithm being subjected to order of cables detection, and is united
Count accuracy rate and detection time.
Further, in the step 2, in the grayscale image in three colo(u)r streak cable regions, the gray value of certain a line pixel is taken.
The grey scale pixel value adjacent to this line carries out calculus of differences.
Further, in the step 2, setting threshold value is 5, counts the coordinate of the pixel greater than threshold value 5, last defeated
Out greater than the coordinate of threshold value 5 and the positioning result figure in three colo(u)r streak cable regions of output.
Further, the feature extraction network that the step 3 simplifies, it is specific to be mentioned using simplified shared convolutional layer network
Feature is taken, deletes a convolutional layer in shared convolutional layer network in convolutional layer 3 and convolutional layer 4 respectively.
Further, some optimizations have been done to anchor frame in the step 3, the anchor frame of design includes cable region as far as possible
Inside, the anchor frame of optimization extracts target area feature using 1: 1.5,1: 1,1: 2 anchor frame.
Further, the step 3 replaces ReLU function using ELU function in Fast R-CNN network.
Further, totally 4000, cable picture, cable picture is divided into two pictures: 2800 are used as training set,
1200 are used as test set
1. carrying out the gray processing of cable image first, three colo(u)r streak cable images are converted to gray level image；
2. taking the gray value of image a line and carrying out calculus of differences to the gray value of neighbor pixel；
3. obtaining a suitable threshold value m, the result and threshold value comparison that step 2 is obtained by experiment, output is greater than threshold
The coordinate and output cord positioning result figure of value.
4. proposing improved Faster R-CNN algorithm of target detection, the anchor frame for more meeting cable feature of image is had chosen
And the feature extraction network simplified；
5. carrying out the detection of order of cables using Faster R-CNN algorithm of target detection.MAP index is finally taken to come pair
The effect of model is evaluated.
The invention has the following advantages: the method for the present invention using improved Faster R-CNN algorithm of target detection into
Line cable sequence detection, first positions cable area-of-interest, then with simplified convolution feature extraction network, optimization
Anchor frame, ELU activation primitive improves Faster R-CNN algorithm.The method of the present invention can more fully extract three colors
The feature of cable, has many advantages, such as easy to operate, high-efficient, saves many manpower consumptions.
The present invention deletes conv3_4 layer and conv4_4 layers in shared convolutional layer network in conv3 and conv4 respectively,
The anchor frame of optimization uses length-width ratio to extract target area feature, when by threshold value m=5, three for 1: 1.5,1: 1,1: 2 anchor frame simultaneously
Colo(u)r streak cable zone location effect is more satisfactory, and improved ingenious place is used to be, can reduce the detection time of cable simultaneously,
Greatly improve the speed of detection；Additionally it is possible to effectively reduce erroneous detection, false retrieval simultaneously greatly improves the accurate of cable detection
Rate.
Further effect, the convolutional layer of shared convolutional layer and pond layer are made certain improvements, and are deleted respectively altogether
The conv3_4 layer in convolution layer network in conv3 and conv4 and conv4_4 layers are enjoyed, the effect for improving model can also be brought in this way
The effect of rate；Original ReLU activation primitive is replaced using ELU activation primitive, there is soft saturability on the functional image left side, facilitates
Noise robustness is promoted, the characteristic of ELU function can make output mean value be close to zero, accelerate convergent speed.
Further effect uses the anchor frame of length-width ratio 2: 1,1: 1,1: 2 for original Faster R-CNN, optimization
Anchor frame extracts target area feature using the anchor frame of length-width ratio 1: 1.5,1: 1,1: 2, and the anchor frame designed in this way can also more fill
The feature for dividing ground to extract image.
Detailed description of the invention
Fig. 1 is Faster R-CNN algorithm network structure；
Fig. 2 is ReLU functional arrangement；
Fig. 3 is shared convolutional layer；
Fig. 4 is the cable image of correct sequence.
Fig. 5 is the cable image of wrong sequence.
Fig. 6 is the image after gray processing.
Fig. 7 is the image carried out after cable zone location.
Fig. 8 is the optimization anchor frame schematic diagram proposed.
Fig. 9 is ELU functional arrangement；
Figure 10 is using the result figure for improving the detection of Faster R-CNN algorithm.
Specific embodiment
It following is a brief introduction of the relevant Faster R-CNN algorithm network structure principle with the present invention.
As shown in Figure 1, being Faster R-CNN algorithm network structure.Faster R-CNN algorithm is mainly by two modules
Composition: original image respectively enters two network modules, first is in Fast R-CNN network module by shared convolutional layer
Sequentially connected characteristic layer, ROI layers, full articulamentum, finally treated, and picture by classification or returns；Second is region
It is recommended that sequentially connected characteristic pattern in network RPN module, low-rank vector, finally treated, and picture by classification or is returned；
Fast R-CNN network module, main purpose are that the candidate region generated to Area generation network is detected and identified in candidate region
Target category；Suggest that network RPN module, main purpose are to generate candidate region in region.Suggest that network connects shared in region
On convolutional layer later layer convolutional layer, a sliding window network, each sliding window are used in the last layer convolutional layer of RPN network
The position of mouth is corresponding with original image, and corresponding position is mesh target area Suggestion box in original image, these region Suggestion box are claimed
For anchor frame.Anchor frame includes three kinds of different areas { 128*128,256*256,512*512 }, and different length-width ratios is { 1: 1,1
: 2,2: 1 }.
As shown in Fig. 2, for the ReLU activation primitive in Fast R-CNN network module, ReLU function expression are as follows:
F (x)=max (0, x)
X is the input of neural network node, and f (x) is the input and output of neural network node；
As shown in figure 3, this shared convolutional layer is the model for having 16 layers: it one shares five sections of convolution i.e. convolutional layer 1 and arrives
Convolutional layer 5, and then pond layer after every section of convolution.Convolutional layer 1 and convolutional layer 2 contain 2 convolutional layers respectively, and convolutional layer 3 arrives
Convolutional layer 5 contains 3 convolutional layers respectively, and there are also full articulamentum 1, full articulamentum 2, full articulamentums 3.Convolution interlayer has used pond
Method, in this way processing mainly retain the main feature extracted, while reducing by next layer of parameter and calculation amount, prevent from intending
It closes.Finally classified with classifier.
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description.
A kind of detection method of order of cables is provided in the present embodiment, comprising the following steps:
Cable picture is divided into two pictures by cable image set totally 4000 picture: 2800 are used as training set, and 1200
Zhang Zuowei test set.
It is the core wire image of wrong sequence in Fig. 5 1. as shown in Figure 4,5, being the cable image of correct sequence in Fig. 4.Such as
Shown in Fig. 6, gray level image is converted by colored cable image first, gray level image tails off than color image pixel information, reduces
Calculation amount；
2. taking the gray value of certain a line pixel in the grayscale image in three colo(u)r streak cable regions.The pixel adjacent to this line
Gray value carries out calculus of differences, thus obtains some difference values；
3. obtain threshold value m=5 by experiment, three colo(u)r streak cable zone location effects are more satisfactory, therefore it is 5 that threshold value, which is arranged,.
The difference value that this threshold value and second step obtain is compared, the coordinate of the pixel greater than threshold value m is counted, finally output is big
In the coordinate of threshold value m and the positioning result figure in three colo(u)r streak cable regions of output, result figure is as shown in Figure 7；
4. the preceding features network in pair Faster R-CNN algorithm of target detection extracts part and has made some improvements, according to
The three uncomplicated features of colo(u)r streak cable image, then the condition of the network number of plies is without very high for extraction feature, in order to improve mould
The efficiency of type, the convolutional layer of shared convolutional layer and pond layer are made certain improvements, and delete shared convolutional layer network respectively
The conv4_3 layer in conv3_3 layer and conv4 in middle conv3, greatly improves detection under the premise of not reducing accuracy rate
Speed；
5. having done some optimizations to anchor frame according to the characteristics of three colo(u)r streak cable images, the anchor frame of design is as far as possible cable region
It is included in the inside, to complete Detection task better.The characteristics of three colo(u)r streak cable images is that height is more more than length, analysis
Obtain original aspect ratio be 2: 1 anchor frame cannot be appropriate three colo(u)r streak cable image of correspondence.As shown in figure 8, by the anchor of length-width ratio 2: 1
Frame is changed to 1: 1.5, remains 1: 1,1: 2 anchor frame, and effect therein is on the one hand to preferably match the cable figure of this paper
On the other hand picture can greatly save the memory of some computers of occupancy by experimental verification.
6. replacing original ReLU activation primitive using ELU activation primitive, there is soft saturability on the functional image left side, facilitates
Noise robustness is promoted, the characteristic of ELU function can make output mean value be close to zero, accelerate convergent speed；Wherein ELU function
Some neurons can be activated, the characteristics of it combines Sigmoid and ReLU, there is soft saturability on the functional image left side, facilitates
Promote noise robustness.And the characteristic of ELU function can make output mean value be close to zero, and accelerate convergent speed.
ELU functional arrangement as shown in Figure 9, ELU function expression:
X is the input of neural network node, and f (x) is the input and output of neural network node, and α is an adjustable ginseng
Number；
5. carrying out sequence detection using Faster R-CNN algorithm of target detection.As shown in Figure 10, finally, being referred to using mAP
Mark is to evaluate the effect of model.In order to embody the detection effect of method provided by the invention, offer of the present invention is compared
Method and original object detection algorithm accuracy rate and detection time.When table 1 illustrates the accuracy rate and detection of two methods
Between, as it can be seen from table 1 method provided by the invention improves certain detection accuracy, network consumption memory is less, have compared with
Good detection performance.
Table 1: cable testing result
| Method | MAP (%) | Detection time (S) |
| Original Faster R-CNN | 94.32 | 0.284 |
| The method of proposition | 96.55 | 0.236 |
In the description of this specification, reference term " one embodiment ", " some embodiments ", " illustrative examples ",
The description of " example ", " specific example " or " some examples " etc. means specific features described in conjunction with this embodiment or example, knot
Structure, material or feature are included at least one embodiment or example of the invention.In the present specification, to above-mentioned term
Schematic representation may not refer to the same embodiment or example.Moreover, specific features, structure, material or the spy of description
Point can be combined in any suitable manner in any one or more of the embodiments or examples.
Although an embodiment of the present invention has been shown and described, it will be understood by those skilled in the art that: not
A variety of change, modification, replacement and modification can be carried out to these embodiments in the case where being detached from the principle of the present invention and objective, this
The range of invention is defined by the claims and their equivalents.
Claims (7)
1. a kind of order of cables detection method based on deep learning, which comprises the following steps:
Step 1: gray processing being carried out to three colo(u)r streak cable images first, does basis for subsequent order detection；Step 2: a certain to image
Capable neighbor pixel gray value carries out calculus of differences, chooses a threshold value and difference result compares, and exports positioning result figure；Step
Rapid 3: the characteristics of being directed to three colo(u)r streak cable images proposes anchor frame, the ELU activation primitive pair of simplified feature extraction network and optimization
Faster R-CNN is improved；Step 4: improved Faster R-CNN algorithm being subjected to order of cables detection, and counts accurate
Rate and detection time.
2. a kind of order of cables detection method based on deep learning according to claim 1, which is characterized in that described
In step 2, in the grayscale image in three colo(u)r streak cable regions, the gray value of certain a line pixel is taken.To the pixel ash that this line is adjacent
Angle value carries out calculus of differences.
3. a kind of order of cables detection method based on deep learning according to claim 1, which is characterized in that described
In step 2, setting threshold value is 5, counts the coordinate of the pixel greater than threshold value 5, and finally output is greater than the coordinate of threshold value 5 and defeated
The positioning result figure in three colo(u)r streak cable regions out.
4. a kind of order of cables detection method based on deep learning according to claim 1, which is characterized in that described
The feature extraction network that step 3 simplifies, it is specific that feature is extracted using simplified shared convolutional layer network, it deletes respectively shared
A convolutional layer in convolution layer network in convolutional layer 3 and convolutional layer 4.
5. a kind of order of cables detection method based on deep learning according to claim 1, which is characterized in that described
Some optimizations are done to anchor frame in step 3, cable region is included in the inside as far as possible by the anchor frame of design, and the anchor frame of optimization uses
The anchor frame of 1:1.5,1:1,1:2 extract target area feature.
6. a kind of order of cables detection method based on deep learning according to claim 1, which is characterized in that described
Step 3, in FastR-CNN network, ReLU function is replaced using ELU function.
7. a kind of order of cables detection method based on deep learning according to claim 1, which is characterized in that cable figure
Piece totally 4000, cable picture is divided into two pictures: 2800 are used as training set, and 1200 are used as test set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910299722.4A CN110136098B (en) | 2019-04-15 | 2019-04-15 | Cable sequence detection method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910299722.4A CN110136098B (en) | 2019-04-15 | 2019-04-15 | Cable sequence detection method based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110136098A true CN110136098A (en) | 2019-08-16 |
| CN110136098B CN110136098B (en) | 2023-07-18 |
Family
ID=67569708
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910299722.4A Active CN110136098B (en) | 2019-04-15 | 2019-04-15 | Cable sequence detection method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110136098B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111738164A (en) * | 2020-06-24 | 2020-10-02 | 广西计算中心有限责任公司 | Pedestrian detection method based on deep learning |
| CN112270668A (en) * | 2020-11-06 | 2021-01-26 | 南京斌之志网络科技有限公司 | Suspended cable detection method and system and electronic equipment |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106846298A (en) * | 2016-12-22 | 2017-06-13 | 清华大学 | The recognition methods of optical fiber winding displacement and device |
| CN109409272A (en) * | 2018-10-17 | 2019-03-01 | 北京空间技术研制试验中心 | Cable Acceptance Test System and method based on machine vision |
| CN109596634A (en) * | 2018-12-30 | 2019-04-09 | 国网北京市电力公司 | The detection method and device of electric cable stoppage, storage medium, processor |
-
2019
- 2019-04-15 CN CN201910299722.4A patent/CN110136098B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106846298A (en) * | 2016-12-22 | 2017-06-13 | 清华大学 | The recognition methods of optical fiber winding displacement and device |
| CN109409272A (en) * | 2018-10-17 | 2019-03-01 | 北京空间技术研制试验中心 | Cable Acceptance Test System and method based on machine vision |
| CN109596634A (en) * | 2018-12-30 | 2019-04-09 | 国网北京市电力公司 | The detection method and device of electric cable stoppage, storage medium, processor |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111738164A (en) * | 2020-06-24 | 2020-10-02 | 广西计算中心有限责任公司 | Pedestrian detection method based on deep learning |
| CN112270668A (en) * | 2020-11-06 | 2021-01-26 | 南京斌之志网络科技有限公司 | Suspended cable detection method and system and electronic equipment |
| CN112270668B (en) * | 2020-11-06 | 2021-09-21 | 威海世一电子有限公司 | Suspended cable detection method and system and electronic equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110136098B (en) | 2023-07-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109961049B (en) | Cigarette brand identification method under complex scene | |
| CN115082683B (en) | Injection molding defect detection method based on image processing | |
| CN109272500B (en) | Fabric classification method based on adaptive convolutional neural network | |
| CN108562589B (en) | Method for detecting surface defects of magnetic circuit material | |
| CN111582294B (en) | Method for constructing convolutional neural network model for surface defect detection and application thereof | |
| CN109684922B (en) | Multi-model finished dish identification method based on convolutional neural network | |
| CN113160062B (en) | Infrared image target detection method, device, equipment and storage medium | |
| CN107016413B (en) | A kind of online stage division of tobacco leaf based on deep learning algorithm | |
| CN110210362A (en) | A kind of method for traffic sign detection based on convolutional neural networks | |
| CN107392925A (en) | Remote sensing image terrain classification method based on super-pixel coding and convolutional neural networks | |
| CN109191459A (en) | The automatic identification and ranking method of continuous casting billet macrostructure center segregation defect | |
| CN104992223A (en) | Intensive population estimation method based on deep learning | |
| CN107492094A (en) | A kind of unmanned plane visible detection method of high voltage line insulator | |
| CN111986125A (en) | Method for multi-target task instance segmentation | |
| CN109872313A (en) | A kind of method for detecting surface defects of products based on depth convolution self-encoding encoder | |
| CN109740572A (en) | A kind of human face in-vivo detection method based on partial color textural characteristics | |
| CN108154158B (en) | Building image segmentation method for augmented reality application | |
| CN109829507B (en) | Aerial high-voltage transmission line environment detection method | |
| CN109801295A (en) | Image partition method based on secondary restricted area growth method | |
| CN110135248A (en) | A kind of natural scene Method for text detection based on deep learning | |
| CN110136098A (en) | A kind of order of cables detection method based on deep learning | |
| CN114005081A (en) | Intelligent detection device and method for foreign matters in tobacco shreds | |
| CN109815957A (en) | A kind of character recognition method based on color image under complex background | |
| CN112767411A (en) | Synchronous segmentation method for multiple organs of lung | |
| CN117522864B (en) | European pine plate surface flaw detection method based on machine vision |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |