CN113610061A - Method and system for identifying unstressed conducting wire based on target detection and residual error network - Google Patents
Method and system for identifying unstressed conducting wire based on target detection and residual error network Download PDFInfo
- Publication number
- CN113610061A CN113610061A CN202111158595.XA CN202111158595A CN113610061A CN 113610061 A CN113610061 A CN 113610061A CN 202111158595 A CN202111158595 A CN 202111158595A CN 113610061 A CN113610061 A CN 113610061A
- Authority
- CN
- China
- Prior art keywords
- network
- feature
- target detection
- conductive
- sample
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
Abstract
The invention discloses a method and a system for identifying a conductor without stress based on target detection and a residual error network, and belongs to the technical field of contact network conductor identification. The existing conductive wire stress identification scheme has poor identification precision and larger error. According to the method for identifying the stress-free conducting wire based on the target detection and the residual error network, the stress condition of the conducting wire is accurately identified by using the target detection algorithm and the residual error network, compared with the existing priori knowledge and the inclined-pull wire angle difference scheme, the method is higher in identification precision, smaller in error, and more objective, accurate and efficient in data processing. Furthermore, the method can efficiently identify the state of the conductor wire, has good stability, can effectively avoid the damage of a contact network system, improves the current collection quality of the pantograph, has a simple and practical scheme, is practical, and has high economic value.
Description
Technical Field
The invention relates to a method and a system for identifying an electric lead without stress based on target detection and a residual error network, belonging to the technical field of contact net electric lead identification.
Background
In the electrified railway industry, a conductive wire plays a crucial role in stable operation of a contact network, namely in chain type suspension, a contact wire is suspended on a catenary wire through the conductive wire, and the length of the conductive wire is adjusted to ensure the structural height and the height of the contact wire of the contact suspension, so that the elasticity of the contact suspension is improved, and the current collection quality of a pantograph is improved. However, there are complex mechanical and electrical interactions between the pantograph and the catenary, which inevitably results in a high defect rate of the pantograph-catenary system and seriously affects the operational safety.
In particular, the loosening and deformation of the conductive wires due to vibrations and excitations during long-term operation are not stressed, which affects the structural height of the contact suspension and the height of the contact lines, resulting in a reduction of the current-carrying quality of the pantograph, which in the past causes unmortal damage to the entire system. Therefore, whether the conductive wire is stressed or not needs to be identified timely and accurately.
Chinese patent (publication No. CN 108734687B) discloses a method and a device for identifying unstressed defects of diagonal cables. Acquiring component linear information of the panoramic image through image processing processes such as edge detection, exclusion of no-sense interest areas, linear detection algorithm and the like; then, reasoning and judging the straight line corresponding to the key suspension through priori knowledge to obtain straight lines of the strut, the flat cantilever, the inclined cantilever, the positioning tube and the positioner component; then judging whether the inclined pull line is stressed or not; further, when the inclined pull wire is not stressed, straight line fitting is carried out on the area where the inclined pull wire is located, parameters forming fitting are extracted to serve as unstressed characteristics, bending is judged, and therefore unstressed defect identification is achieved.
According to the scheme, whether the stay wire is stressed or not is judged according to the priori knowledge and the angle difference of the stay wire, but the priori knowledge needs to presuppose some rules and set some necessary conditions, certain subjectivity and randomness exist, and data cannot be objectively and accurately processed; meanwhile, due to the influence of wind power and the tightness of the stay wires, whether the stay wires are stressed or not is judged by utilizing the angle difference of the inclined stay wires, and a large error exists, so that the scheme has poor identification precision and a large error.
Disclosure of Invention
Aiming at the defects of the prior art, the invention aims to provide the method and the system for identifying the conductor wire without stress based on the target detection and the residual error network, which have the advantages of high identification precision, small error, more objective, more accurate and more efficient data processing, can further effectively avoid the damage of a contact network system, and improve the current collection quality of the pantograph.
In order to achieve the purpose, the technical scheme of the invention is as follows:
the conductive lines based on object detection and residual error networks are not subjected to a force identification method,
the method comprises the following steps:
firstly, acquiring an image of a contact net sample of a conductive wire;
the contact net sample image comprises a conductor wire stressed sample and a conductor wire unstressed sample;
secondly, marking the conductive wires in the conductive wire stressed samples and the conductive wire unstressed samples in the first step respectively to obtain corresponding label files, wherein the label files, the conductive wire stressed samples and the conductive wire unstressed samples form an image data set together;
thirdly, scaling the pictures in the image data set in the second step to form scaled image data;
fourthly, constructing a detection model1 according to the zoomed image data in the third step and through the target detection network, and outputting the position coordinates of the conductive wire in the contact network sample image;
fifthly, utilizing the position coordinates in the fourth step to extract a conducting wire image from the contact net sample image;
and sixthly, comparing and scaling the conductive line images in the fifth step, constructing a recognition model2 by combining a residual error network ResNet, and outputting a recognition result of whether the conductive line is stressed or not.
According to the method, through continuous exploration and test, the stress condition of the conducting wire is accurately identified by using a target detection algorithm and a residual error network, and compared with the existing priori knowledge and an inclined-pull wire angle difference scheme, the method is higher in identification precision, smaller in error and more objective, accurate and efficient in data processing.
Furthermore, the method can efficiently identify the state of the conductor wire, has good stability, can effectively avoid the damage of a contact network system, improves the current collection quality of the pantograph, has a simple and practical scheme, is practical, and has high economic value.
Meanwhile, the method can also be used for identifying the defects of other parts of the contact network, has certain universality and is convenient to popularize and use.
Furthermore, when the target detection network performs deep learning, there is a degradation problem, so that the target detection network is not easy to train, and therefore a residual error network ResNet needs to be set to solve the degradation problem.
As a preferable technical measure:
the acquisition mode of the contact net sample image in the first step is as follows:
the method comprises the following steps of shooting a conducting wire in a contact net through a camera to obtain the conducting wire;
the camera is assembled on the patrol car;
the patrol car moves to the vicinity of the contact net according to the set route, the camera faces the contact net, and the contact net is taken as a target object for shooting;
the camera is a monitoring camera or/and a mobile phone camera.
As a preferable technical measure:
in the second step, the conductive line marking method is as follows:
marking through surrounding frames, wherein one surrounding frame corresponds to one conductive wire area, and each conductive wire stressed sample or conductive wire unstressed sample marks 1-3 surrounding frames;
the surrounding frame is a rectangular frame which is a 1 × 4 row vector;
the row vector is used for recording horizontal and vertical coordinates corresponding to the upper left corner and the lower right corner of the rectangular frame;
the information of the horizontal and vertical coordinates forms a corresponding label file;
the image data set is divided into a training set and a verification set according to the proportion of N: 1;
2≤N≤8。
as a preferable technical measure:
in the third step, the scaling method is as follows:
and adjusting the sizes of the stressed sample and the unstressed sample of the conductive wire in the contact network sample image to be the same, wherein the specific size is 1024 multiplied by 800, and the corresponding label files are also scaled in the same proportion.
As a preferable technical measure:
in the fourth step, the method for constructing the detection model1 specifically includes the following steps:
firstly, training a target detection network by using a training set;
step two, verifying the target detection network in the step one by using a verification set;
and step three, predicting and outputting the position coordinates of the conductive wire in the contact net sample image by using the target detection network in the step two.
As a preferable technical measure:
the target detection network is a single-stage target detection FCOS model (FCOS network), is used for inputting images and outputting a feature map of corresponding input images, and comprises a backbone network and a feature layer.
The single-phase detection network has the greatest advantage of high speed, and the double-phase detection algorithm has the greatest advantage of high precision. However, with the development of the target detection algorithm, the precision of the single-stage target detection algorithm is greatly improved and basically can be comparable to the double-stage detection algorithm, particularly the FCOS algorithm, so that the single-stage target detection FCOS model can meet the identification precision requirement of the invention.
The backbone network is a basic feature extractor of the target detection network and is represented by a feature map C3, a feature map C4 and a feature map C5;
the feature map C3, which has a size of 128 × 512, shows that the feature map has a size of 128 × 128, and 512 extracted catenary sample images are provided;
the size of the characteristic map C4 is 64 × 1024, the size of the characteristic map is 64 × 64, and 1024 extracted contact net sample images are arranged;
the size of the feature map C5 is 32 × 2048, which indicates that the size of the feature map is 32 × 32, and 2048 extracted catenary sample images are provided.
The feature layer comprises feature P3, feature P4, feature P5, feature P6, feature P7 for final prediction;
the characteristics P3, P4 and P5 utilize characteristic diagram C3, characteristic diagram C4 and characteristic diagram C5 and are connected from top to bottom
Generating a convolution layer;
feature P6 and feature P7 are implemented by applying one convolutional layer with step size of 2 to feature P5 and feature P6, respectively;
the height of the characteristic diagram is H, the width of the characteristic diagram is W, and the sampling rate of the characteristic diagram relative to the input contact net sample image is s;
the above-mentioned
；
Preferably, the step sizes of the feature layers P3, P4, P5, P6 and P7 are 8,16, 32, 64 and 128, respectively.
As a preferable technical measure:
in the fifth step, the method for scratching out the conductive line image is as follows:
the corresponding area is cut out from the contact net sample image according to the position coordinates, and then the corresponding area is subjected to image scaling, so that the size of the corresponding area is 224 x 224.
As a preferable technical measure:
the residual network ResNet in the sixth step is a residual network ResNet50, which is provided with a focus loss model.
The focus loss model is obtained by improving a cross entropy loss model, and a modulation factor is added on the basis of the cross entropy loss model
The method is used for reducing the loss of simple samples and expanding the range of low loss of sample receiving;
the calculation formula of the focus loss model is as follows:
the above-mentionedp t The calculation formula is as follows:
wherein the content of the first and second substances,
is the estimated probability of the focus loss model for the class with label y = 1,
is a focus parameter, can be adjusted
。
The recognition model2 uses class activation mapping to locate the defect locations in the conductive line image.
As a preferable technical measure:
the cross entropy loss model is calculated as follows:
wherein the content of the first and second substances,
determining a category of ground truth;
input to cross entropy loss modelp t To obtain
One significant characteristic of the cross-entropy loss model is: even with samples (p) that are easy to classifyt >>0.5) also results in large losses; these loss values may overwhelm the rare class if a large number of simple samples are added;
therefore, the modulation factor is set, the loss model of the simple sample can be effectively reduced, and the range of the sample receiving low loss model is expanded.
As a preferable technical measure:
a conductive line stress-free identification system based on target detection and residual error network,
it includes:
one or more processors;
storage means for storing one or more programs;
the one or more programs, when executed by the one or more processors, cause the one or more processors to implement a target detection and residual network based conductive line unstressed identification method as described above.
Firstly, detecting and positioning a conductive wire in an image through an FCOS (FCOS model), and then deducting the positioned conductive wire from an original image and zooming the conductive wire; and classifying and identifying the zoomed conductive line image through a ResNet50 model with focus loss added, and judging whether the conductive line has the problem of no stress.
The method can solve the problem of identifying whether the high-speed railway contact net conductor is stressed or not, and the identification result based on the scene of no stress of the Xinjiang high-speed railway contact net conductor shows that the accuracy rate of the identification method reaches 93.6%, and the method can be applied to identifying whether the high-speed railway contact net conductor is stressed or not. Meanwhile, the method has universality for defect identification of other contact net parts.
Compared with the prior art, the invention has the following beneficial effects:
according to the method, through continuous exploration and test, the stress condition of the conducting wire is accurately identified by using a target detection algorithm and a residual error network, and compared with the existing priori knowledge and an inclined-pull wire angle difference scheme, the method is higher in identification precision, smaller in error and more objective, accurate and efficient in data processing.
Furthermore, the method can efficiently identify the state of the conductor wire, has good stability, can effectively avoid the damage of a contact network system, improves the current collection quality of the pantograph, has a simple and practical scheme, is practical, and has high economic value.
Drawings
FIG. 1 is a sample image view of a catenary of the present invention;
FIG. 2 is a diagram showing a comparison between the stress on the conductive wire according to the present invention;
FIG. 3 is a flow chart of the present invention for constructing a detection model;
FIG. 4 is a flow chart of a training process of the FCOS model of the present invention;
FIG. 5 is a flow chart of the present invention for constructing a recognition model;
FIG. 6 is an identification flow chart employing the present invention;
fig. 7 is a diagram showing the result of the force recognition of a conductive wire to which the present invention is applied.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
On the contrary, the invention is intended to cover alternatives, modifications, equivalents and alternatives which may be included within the spirit and scope of the invention as defined by the appended claims. Furthermore, in the following detailed description of the present invention, certain specific details are set forth in order to provide a better understanding of the present invention. It will be apparent to one skilled in the art that the present invention may be practiced without these specific details.
The conductive wire unstressed identification method based on target detection and a residual error network comprises the following steps:
firstly, acquiring an image of a contact net sample of a conductive wire;
the contact net sample image comprises a conductor wire stressed sample and a conductor wire unstressed sample;
secondly, marking the conductive wires in the conductive wire stressed samples and the conductive wire unstressed samples in the first step respectively to obtain corresponding label files, wherein the label files, the conductive wire stressed samples and the conductive wire unstressed samples form an image data set together;
thirdly, scaling the pictures in the image data set in the second step to form scaled image data;
fourthly, constructing a detection model1 according to the zoomed image data in the third step and through the target detection network, and outputting the position coordinates of the conductive wire in the contact network sample image;
fifthly, utilizing the position coordinates in the fourth step to extract a conducting wire image from the contact net sample image;
and sixthly, comparing and scaling the conductive line images in the fifth step, constructing a recognition model2 by combining a residual error network ResNet, and outputting a recognition result of whether the conductive line is stressed or not.
According to the method, through continuous exploration and test, the stress condition of the conducting wire is accurately identified by using a target detection algorithm and a residual error network, and compared with the existing priori knowledge and an inclined-pull wire angle difference scheme, the method is higher in identification precision, smaller in error and more objective, accurate and efficient in data processing.
Furthermore, the method can efficiently identify the state of the conductor wire, has good stability, can effectively avoid the damage of a contact network system, improves the current collection quality of the pantograph, has a simple and practical scheme, is practical, and has high economic value.
Meanwhile, the method can also be used for identifying the defects of other parts of the contact network, has certain universality and is convenient to popularize and use.
Furthermore, when the target detection network performs deep learning, there is a degradation problem, so that the target detection network is not easy to train, and therefore a residual error network ResNet needs to be set to solve the degradation problem.
The best embodiment of the invention is as follows:
1) the method comprises the following steps of shooting and collecting a contact network sample image (shown in figure 1) comprising a conductive wire through a camera, wherein the image comprises two types of stressed and unstressed conductive wires (shown in figure 2).
In the examples of the present invention, there were 800 experimental pictures and 892 examples. 640 pictures are used for training, the rest 160 pictures are used as test pictures, 83 examples of the test pictures are not stressed, and 97 examples are stressed. The original picture has a size of 2585 × 1940 pixels.
2) Traversing all the contact network sample images, marking each contact network sample image by using a bounding box for the conductive wire to obtain a corresponding label file, forming an image data set by the label file and the original contact network sample image together, and distinguishing a training set and a verification set according to the image data set by a proportion of about 4: 1.
3) The method comprises the steps of carrying out image scaling on a contact network sample image in an image data set, adjusting the size of the contact network sample image to be the same, specifically 1024 × 800, carrying out the same scaling on a label file corresponding to the contact network sample image, then training an FCOS (fuzzy c-means operating system) model by using a training set, predicting and outputting position coordinates of a conducting wire in the contact network sample image by using the FCOS model, and verifying the trained FCOS model by using a verification set to obtain a detection model1 (the flow is shown in FIG. 3).
As shown in FIG. 4, the FCOS model is a training embodiment:
the FCOS model comprises a backbone network and a feature layer.
The backbone network is a basic feature extractor of the target detection network and is represented by a feature map C3, a feature map C4 and a feature map C5;
the feature map C3, which has a size of 128 × 512, shows that the feature map has a size of 128 × 128, and 512 extracted catenary sample images are provided;
the size of the characteristic map C4 is 64 × 1024, the size of the characteristic map is 64 × 64, and 1024 extracted contact net sample images are arranged;
the size of the feature map C5 is 32 × 2048, which indicates that the size of the feature map is 32 × 32, and 2048 extracted catenary sample images are provided.
The feature layer comprises feature P3, feature P4, feature P5, feature P6, feature P7 for final prediction;
the features P3, P4 and P5 utilize feature map C3, feature map C4 and feature map C5 and have 1 connected from top to bottom
1, generating a convolution layer;
feature P6 and feature P7 are implemented by applying one convolutional layer with step size of 2 to feature P5 and feature P6, respectively;
the height of the characteristic diagram is H, the width of the characteristic diagram is W, and the sampling rate of the characteristic diagram relative to the input contact net sample image is s;
the sum of said s =8,16,
,128；
the step sizes of feature layers P3, P4, P5, P6 and P7 are 8,16, 32, 64 and 128, respectively.
4) The method comprises the steps of inputting a contact network sample image and position coordinates of a conducting wire thereof into a ResNet50 model with focus loss for training, predicting and outputting classification recognition results of the conducting wire by the model, carrying out defect positioning on the conducting wire image by using class activation mapping to obtain a recognition model2, and verifying the trained ResNet50 model with focus loss by using a verification set to obtain a recognition model2 (the flow is shown in FIG. 5).
5) Zooming the contact network image to be detected acquired in real time according to the same image zooming method as that in the step 3), inputting the zoomed contact network image to the detection model1 to obtain the position coordinates of the conducting wire, pulling out the target object of the conducting wire from the original contact network image according to the position coordinates, zooming according to the same image zooming method as that in the step 3), inputting the recognition model2 to output the obtained non-stressed defect recognition result of the conducting wire, and displaying the defect position through class activation mapping. (this flow is shown in FIG. 6).
The detection model1 trained by the invention is adopted to detect the pictures of the catenary test set, and the recognition model2 trained by the invention is adopted to recognize the pictures of the conductive wires extracted from the pictures of the catenary test set, so that the obtained results are shown in table 1.
TABLE 1 results of conductive wire testing
As can be seen from table 1, the recognition accuracy of the stressed conductive wire is 90.7%, and the recognition accuracy of the unstressed conductive wire is 96.4%.
The result of identifying the conductor line without force is shown in fig. 7, wherein the left column in the figure represents an image of a catenary sample including the conductor line captured by a camera, the middle column represents an image of the conductor line cut from the image of the catenary sample according to the position coordinates of the conductor line and scaled to a fixed size of 224 × 224, and the right column represents an image of an important area marked by class activation mapping.
The right image contains characters and is the result automatically generated by the program, wherein label represents the class of the image, predict represents the class predicted by the model, and the following decimal represents the confidence coefficient of the model prediction result. As can be seen from the figure, the model has better prediction results, the confidence degrees are 0.81752 and 0.9834 respectively, and the haze areas in the right column of images represent marked important areas, so that the model can be known to really focus on the important difference between the two states of the conductor line without stress and the state with stress.
Therefore, the method can realize automatic identification of the unstressed state of the conducting wire, has the advantages of high accuracy, good stability, strong anti-interference capability, high universality and the like, and can be applied to a high-speed railway contact net defect identification system.
A system embodiment to which the method of the invention is applied:
a conductive line unstressed recognition system based on target detection and residual error network, comprising:
one or more processors;
storage means for storing one or more programs;
the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the above-described target detection and residual network-based conductive line unstressed identification method.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solutions of the present invention and not for limiting the same, and although the present invention is described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: modifications and equivalents may be made to the embodiments of the invention without departing from the spirit and scope of the invention, which is to be covered by the claims.
Claims (10)
1. A method for identifying a conductive line without stress based on target detection and a residual error network is characterized in that,
the method comprises the following steps:
firstly, acquiring an image of a contact net sample of a conductive wire;
the contact net sample image comprises a conductor wire stressed sample and a conductor wire unstressed sample;
secondly, marking the conductive wires in the conductive wire stressed samples and the conductive wire unstressed samples in the first step respectively to obtain corresponding label files, wherein the label files, the conductive wire stressed samples and the conductive wire unstressed samples form an image data set together;
thirdly, scaling the pictures in the image data set in the second step to form scaled image data;
fourthly, constructing a detection model1 according to the zoomed image data in the third step and through the target detection network, and outputting the position coordinates of the conductive wire in the contact network sample image;
fifthly, utilizing the position coordinates in the fourth step to extract a conducting wire image from the contact net sample image;
and sixthly, comparing and scaling the conductive line images in the fifth step, constructing a recognition model2 by combining a residual error network ResNet, and outputting a recognition result of whether the conductive line is stressed or not.
2. The method of claim 1, wherein the object detection and residual network-based force-free identification of conductive lines,
the acquisition mode of the contact net sample image in the first step is as follows:
the method comprises the following steps of shooting a conducting wire in a contact net through a camera to obtain the conducting wire;
the camera is assembled on the patrol car;
the patrol car moves to the vicinity of the contact net according to the set route, the camera faces the contact net, and the contact net is taken as a target object for shooting;
the camera is a monitoring camera or/and a mobile phone camera.
3. The method of claim 1, wherein the object detection and residual network-based force-free identification of conductive lines,
in the second step, the conductive line marking method is as follows:
marking through surrounding frames, wherein one surrounding frame corresponds to one conductive wire area, and each conductive wire stressed sample or conductive wire unstressed sample marks 1-3 surrounding frames;
the surrounding frame is a rectangular frame which is a 1 × 4 row vector;
the row vector is used for recording horizontal and vertical coordinates corresponding to the upper left corner and the lower right corner of the rectangular frame;
the information of the horizontal and vertical coordinates forms a corresponding label file;
the image data set is divided into a training set and a verification set according to the proportion of N: 1;
2≤N≤8。
4. the method of claim 1, wherein the object detection and residual network-based force-free identification of conductive lines,
in the third step, the scaling method is as follows:
and adjusting the sizes of the stressed sample and the unstressed sample of the conductive wire in the contact network sample image to be the same, wherein the specific size is 1024 multiplied by 800, and the corresponding label files are also scaled in the same proportion.
5. The method of claim 3, wherein the object detection and residual error network-based force-free identification of conductive lines,
in the fourth step, the method for constructing the detection model1 specifically includes the following steps:
firstly, training a target detection network by using a training set;
step two, verifying the target detection network in the step one by using a verification set;
and step three, predicting and outputting the position coordinates of the conductive wire in the contact net sample image by using the target detection network in the step two.
6. The method of claim 5, wherein the object detection and residual error network-based force-free identification of conductive lines,
the target detection network is a single-stage target detection FCOS model and comprises a backbone network and a feature layer;
the backbone network is a basic feature extractor of the target detection network and is represented by a feature map C3, a feature map C4 and a feature map C5;
the feature map C3, which has a size of 128 × 512, shows that the feature map has a size of 128 × 128, and 512 extracted catenary sample images are provided;
the size of the characteristic map C4 is 64 × 1024, the size of the characteristic map is 64 × 64, and 1024 extracted contact net sample images are arranged;
the size of the feature map C5 is 32 × 2048, the size of the feature map is 32 × 32, and 2048 extracted contact net sample images are provided;
the feature layer comprises feature P3, feature P4, feature P5, feature P6, feature P7 for final prediction;
the features P3, P4 and P5 utilize feature map C3, feature map C4 and feature map C5 and have 1 connected from top to bottom
1, generating a convolution layer;
feature P6 and feature P7 are implemented by applying one convolutional layer with step size of 2 to feature P5 and feature P6, respectively;
the height of the characteristic diagram is H, the width of the characteristic diagram is W, and the sampling rate of the characteristic diagram relative to the input contact net sample image is s;
said s =8,16, …, 128.
7. The method of claim 1, wherein the object detection and residual network-based force-free identification of conductive lines,
in the fifth step, the method for scratching out the conductive line image is as follows:
the corresponding area is cut out from the contact net sample image according to the position coordinates, and then the corresponding area is subjected to image scaling, so that the size of the corresponding area is 224 x 224.
8. The method of stress-free identification of conductive lines based on object detection and residual error networks according to any of claims 1-7,
the residual error network ResNet in the sixth step is a residual error network ResNet50 which is provided with a focus loss model;
the focus loss model is obtained by improving a cross entropy loss model, and a modulation factor is added on the basis of the cross entropy loss model
The method is used for reducing the loss of simple samples and expanding the range of low loss of sample receiving;
the calculation formula of the focus loss model is as follows:
the above-mentionedp t The calculation formula is as follows:
wherein the content of the first and second substances,
is the estimated probability of the focus loss model for the class with label y = 1,
is a focus parameter, can be adjusted
；
The recognition model2 uses class activation mapping to locate the defect locations in the conductive line image.
9. The method of claim 8, wherein the object detection and residual error network-based force-free identification of conductive lines,
the cross entropy loss model is calculated as follows:
wherein the content of the first and second substances,
determining a category of ground truth;
input to cross entropy loss modelp t To obtain
10. A system for identifying without stress a conductive wire based on target detection and residual error network is characterized in that,
it includes:
one or more processors;
storage means for storing one or more programs;
the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the target detection and residual network based conductive line unstressed identification method of any of claims 1-9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111158595.XA CN113610061A (en) | 2021-09-30 | 2021-09-30 | Method and system for identifying unstressed conducting wire based on target detection and residual error network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111158595.XA CN113610061A (en) | 2021-09-30 | 2021-09-30 | Method and system for identifying unstressed conducting wire based on target detection and residual error network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113610061A true CN113610061A (en) | 2021-11-05 |
Family
ID=78343307
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111158595.XA Pending CN113610061A (en) | 2021-09-30 | 2021-09-30 | Method and system for identifying unstressed conducting wire based on target detection and residual error network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113610061A (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108647665A (en) * | 2018-05-18 | 2018-10-12 | 西安电子科技大学 | Vehicle real-time detection method of taking photo by plane based on deep learning |
| CN110807376A (en) * | 2019-10-17 | 2020-02-18 | 北京化工大学 | Method and device for extracting urban road based on remote sensing image |
| CN111695493A (en) * | 2020-06-10 | 2020-09-22 | 国网山东省电力公司电力科学研究院 | Method and system for detecting hidden danger of power transmission line |
| CN112801166A (en) * | 2021-01-25 | 2021-05-14 | 上海海事大学 | Melanoma classification method based on feature data and dimension scaling |
| CN112819096A (en) * | 2021-02-26 | 2021-05-18 | 西北大学 | Method for constructing fossil image classification model based on composite convolutional neural network |
-
2021
- 2021-09-30 CN CN202111158595.XA patent/CN113610061A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108647665A (en) * | 2018-05-18 | 2018-10-12 | 西安电子科技大学 | Vehicle real-time detection method of taking photo by plane based on deep learning |
| CN110807376A (en) * | 2019-10-17 | 2020-02-18 | 北京化工大学 | Method and device for extracting urban road based on remote sensing image |
| CN111695493A (en) * | 2020-06-10 | 2020-09-22 | 国网山东省电力公司电力科学研究院 | Method and system for detecting hidden danger of power transmission line |
| CN112801166A (en) * | 2021-01-25 | 2021-05-14 | 上海海事大学 | Melanoma classification method based on feature data and dimension scaling |
| CN112819096A (en) * | 2021-02-26 | 2021-05-18 | 西北大学 | Method for constructing fossil image classification model based on composite convolutional neural network |
Non-Patent Citations (2)
| Title |
|---|
| ZHI TIAN等: "FCOS: Fully Convolutional One-Stage Object Detection", 《ARXIV》 * |
| 齐冬莲等: "一种基于RefineDet网络和霍夫变换的高速铁路接触网吊弦状态多尺度检测方法", 《电子与信息学报》 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109753929B (en) | High-speed rail insulator inspection image recognition method | |
| CN115170576B (en) | Aluminum pipe surface defect detection method based on machine vision | |
| CN110059694A (en) | The intelligent identification Method of lteral data under power industry complex scene | |
| CN111814686A (en) | Vision-based power transmission line identification and foreign matter invasion online detection method | |
| JP5500024B2 (en) | Image recognition method, apparatus, and program | |
| CN111260629A (en) | Pantograph structure abnormity detection algorithm based on image processing | |
| Aydin | A new approach based on firefly algorithm for vision-based railway overhead inspection system | |
| CN111784633A (en) | Insulator defect automatic detection algorithm for power inspection video | |
| CN109360188A (en) | A kind of detection method of high iron catenary equipotential line defective mode | |
| CN114333070A (en) | Examinee abnormal behavior detection method based on deep learning | |
| CN111539924A (en) | Defect detection method, device and equipment for suspension clamp and storage medium | |
| CN111640135A (en) | TOF camera pedestrian counting method based on hardware front end | |
| CN114694130A (en) | Method and device for detecting telegraph poles and pole numbers along railway based on deep learning | |
| CN114782410A (en) | Insulator defect detection method and system based on lightweight model | |
| CN104966095A (en) | Image target detection method and apparatus | |
| CN111767919B (en) | Multilayer bidirectional feature extraction and fusion target detection method | |
| CN113610061A (en) | Method and system for identifying unstressed conducting wire based on target detection and residual error network | |
| CN109934172B (en) | GPS-free full-operation line fault visual detection and positioning method for high-speed train pantograph | |
| CN111862023A (en) | Pantograph structure abnormity detection method based on computer vision | |
| CN114638989A (en) | Fault classification visualization method based on target detection and fine-grained identification | |
| CN112380970A (en) | Video target detection method based on local area search | |
| CN110807416A (en) | Digital instrument intelligent recognition device and method suitable for mobile detection device | |
| CN117173385B (en) | Detection method, device, medium and equipment of transformer substation | |
| CN116883391B (en) | Two-stage distribution line defect detection method based on multi-scale sliding window | |
| Zhang et al. | ECF-STPM: A robust crack detection method for railway catenary components |
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: 20211105 |