CN114387510A - Bird identification method and device for power transmission line and storage medium - Google Patents
Bird identification method and device for power transmission line and storage medium Download PDFInfo
- Publication number
- CN114387510A CN114387510A CN202111578096.6A CN202111578096A CN114387510A CN 114387510 A CN114387510 A CN 114387510A CN 202111578096 A CN202111578096 A CN 202111578096A CN 114387510 A CN114387510 A CN 114387510A
- Authority
- CN
- China
- Prior art keywords
- bird
- recognition
- training
- shielding
- model
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a bird identification method, a bird identification device and a storage medium for a power transmission line, wherein the bird identification method comprises the following steps: acquiring a bird image data set of the power transmission line, and performing data enhancement processing on the bird image data set to obtain a training image set; adding a shielding label to the training image set, and extracting a characteristic graph of the training image set by adopting a convolutional neural network; inputting the characteristic diagram into a mainstream SSD model, calculating to obtain a loss function, and combining the loss function with a shielding label to perform model training to obtain a bird recognition model; the method comprises the steps of collecting image data to be recognized of the power transmission line, inputting the image data to be recognized into a bird recognition model, outputting a plurality of recognition frames, and selecting an optimal recognition frame from the recognition frames to serve as a final recognition result. According to the embodiment of the invention, the loss compensation mechanism is formed by combining the shielding label with the loss function, and the model training is carried out by adopting the loss compensation mechanism to identify the birds, so that the bird identification precision under the condition that the birds are shielded is effectively improved.
Description
Technical Field
The invention relates to the technical field of bird identification, in particular to a method and a device for identifying birds of a power transmission line and a storage medium.
Background
With the improvement of ecological environment, returning to farming, wetland restoration and the construction of a green ecological environment-friendly society, the conditions for bird reproduction and rest are gradually improved, the number of various birds is continuously increased, and various accidents caused by the birds are increased. Bird damage in the transmission line occurs frequently, for example, insulator flashover caused by bird droppings, line tripping faults caused by bird nesting on the line are caused, birds have large body types, the wire discharges to air gaps of the cross arm along the surface of the insulator when standing on the cross arm, single-phase short circuit grounding and the like occur, and bird damage accidents occurring on the transmission line not only cause great pressure to stable operation of an electric power system, but also cause great manpower and material loss.
Aiming at the problem of bird damage prevention of the power transmission line, the bird repelling device is installed on the power transmission line to repel birds at home and abroad, but the bird repelling device is easy to increase power consumption after working for a long time, bird repelling efficiency is reduced, and the service life is prolonged, so that the bird repelling device is very important for monitoring bird activities on the power transmission line in real time and repelling birds in time by combining an image recognition technology. When the existing power transmission line bird identification method is used for identifying birds aiming at images shot by a monitoring camera, the problem that the birds cannot be accurately identified due to the fact that the birds are partially shielded by a steel tower or a bird repelling device exists.
Disclosure of Invention
The invention provides a bird identification method and device for a power transmission line and a storage medium, and solves the problem that birds cannot be accurately identified by the existing bird identification method for the power transmission line.
One embodiment of the invention provides a bird identification method for a power transmission line, which comprises the following steps:
acquiring a bird image data set of a power transmission line, and performing data enhancement processing on the bird image data set to obtain a training image set;
adding an occlusion label to the training image set, and extracting a characteristic diagram of the training image set by adopting a convolutional neural network;
inputting the characteristic diagram into a main stream SSD model to calculate a loss function, and combining the loss function with the shielding label to perform model training to obtain a bird recognition model;
acquiring image data to be recognized of the power transmission line, inputting the image data to be recognized into the bird recognition model, outputting a plurality of recognition frames, and selecting an optimal recognition frame from the recognition frames as a final recognition result.
Further, the adding an occlusion tag to the bird image dataset includes:
setting a corresponding shielding level according to the proportion of the shielded area of the birds in the bird image data set, and adding a shielding label to the bird image data set according to the shielding level, wherein the larger the proportion of the shielded area of the birds is, the higher the corresponding shielding level is.
Further, the training of the loss function in combination with the occlusion label includes:
detecting whether a shielding label exists in a labeling frame in the feature map, if so, acquiring a corresponding shielding grade according to the shielding label, and calculating to obtain a shielding coefficient corresponding to the shielding label according to the shielding grade;
and combining the shielding coefficient with a loss function to obtain a shielding compensation coefficient, and performing model training according to the shielding compensation coefficient.
Further, the higher the occlusion level is, the larger the occlusion compensation coefficient is.
Further, the combining the loss function and the shielding label for model training to obtain the bird recognition model includes:
generating a prior frame on the feature map through the SSD model, taking the prior frame with the intersection ratio with the real target being greater than a preset threshold value as a training positive sample, and defining the prior frame with the intersection ratio with the real target being less than or equal to the preset threshold value as a candidate negative sample;
sampling the candidate negative samples, sorting the sampled candidate negative samples in a descending order according to confidence errors, and selecting the candidate negative samples meeting the quantity requirement as training negative samples according to the quantity of the training positive samples and the quantity proportion of the training sample positive samples to the training negative samples and the maximum and descending order of errors;
and carrying out model training according to the training positive sample and the training negative sample to obtain a bird recognition model.
Further, the selecting an optimal recognition frame from the plurality of recognition frames as a final recognition result includes:
and filtering the recognition frames with the overlapping degree higher than a preset threshold value in the recognition frames by adopting a non-maximum value suppression algorithm, and selecting the optimal recognition frame from the filtered recognition frames as a final recognition result.
Further, the loss function is defined as:
wherein, N is the number of positive samples of the prior frame, c is the identification value of the category confidence coefficient,/is the identification value of the position of the boundary frame corresponding to the prior frame, g is the position parameter of the real frame, LconfAs confidence error, LlocAs the position error, α is the learning rate.
Further, the data enhancement processing includes at least one of horizontal flipping, random cropping, color warping, and randomly acquiring block fields.
One embodiment of the present invention provides a bird recognition device for a power transmission line, including:
the data preprocessing module is used for acquiring a bird image data set of the power transmission line and performing data enhancement processing on the bird image data set to obtain a training image set;
the characteristic image extraction module is used for adding an occlusion label to the training image set and extracting a characteristic image of the training image set by adopting a convolutional neural network;
the model training module is used for inputting the characteristic diagram into a main stream SSD model to calculate a loss function, and combining the loss function with the shielding label to perform model training to obtain a bird recognition model;
and the bird recognition module is used for acquiring the image data to be recognized of the power transmission line, inputting the image data to be recognized into the bird recognition model, outputting a plurality of recognition frames, and selecting the optimal recognition frame from the recognition frames as a final recognition result.
An embodiment of the present invention provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements the steps of the transmission line bird identification method as described above.
According to the embodiment of the invention, the shielding label is added to the training image set, the shielding label is combined with the loss function to form the loss compensation mechanism, model training is carried out in a loss compensation mechanism mode, a bird identification model capable of accurately identifying birds can be obtained through effective training, and the bird identification precision under the condition that the birds are shielded is improved, so that the accuracy of bird identification on the power transmission line can be effectively improved, the bird repelling device can be accurately controlled according to the bird identification result to repel the birds, the working efficiency of the bird repelling device is effectively improved, and the safety of the power transmission line is improved.
Drawings
Fig. 1 is a schematic flow chart of a bird identification method for a power transmission line according to an embodiment of the present invention;
fig. 2 is a schematic structural diagram of a bird identification device for a power transmission line provided in an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, 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 application.
In the description of the present application, it is to be understood that the terms "first", "second" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implying any number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of the present application, "a plurality" means two or more unless otherwise specified.
In the description of the present application, it is to be noted that, unless otherwise explicitly specified or limited, the terms "mounted," "connected," and "connected" are to be construed broadly, e.g., as meaning either a fixed connection, a removable connection, or an integral connection; can be mechanically or electrically connected; they may be connected directly or indirectly through intervening media, or they may be interconnected between two elements. The specific meaning of the above terms in the present application can be understood in a specific case by those of ordinary skill in the art.
Referring to fig. 1, an embodiment of the present invention provides a method for identifying birds in a power transmission line, including:
s1, acquiring a bird image data set of the power transmission line, and performing data enhancement processing on the bird image data set to obtain a training image set;
in the embodiment of the invention, a bird image data set can be acquired through an image acquisition device arranged on the power transmission line, wherein the image acquisition device comprises a camera arranged on a steel tower of the power transmission line; also can patrol and examine unmanned aerial vehicle and gather birds image data set. Wherein, birds image data set is including the birds image that shelters from and the birds image that does not shelter from, and the birds image that has the shelter from divides into different sheltering from grades according to sheltering from the degree to improve follow-up accuracy to birds discernment.
S2, adding an occlusion label to the training image set, and extracting a characteristic diagram of the training image set by adopting a convolutional neural network;
in the embodiment of the invention, the corresponding shielding labels are added to the training image set according to the shielding levels divided according to the shielding degrees of the birds in the bird images, and model training is carried out according to the different shielding degrees corresponding to the shielding labels, so as to obtain the model for accurately identifying the birds.
S3, inputting the characteristic diagram into a mainstream SSD model, calculating to obtain a loss function, and combining the loss function with a shielding label to perform model training to obtain a bird recognition model;
in the embodiment of the invention, the model training is carried out by combining the loss function and the shielding label, so that the sensitivity of the model to the shielding target can be effectively enhanced, and the accuracy of identifying whether birds exist on a power transmission line by the bird identification model can be effectively improved.
S4, collecting image data to be recognized of the power transmission line, inputting the image data to be recognized into the bird recognition model, outputting a plurality of recognition frames, and selecting the optimal recognition frame from the recognition frames as a final recognition result.
In a specific embodiment, when the bird in the image is marked with 1, namely the result of the identification frame is 1, the bird in the image is shown, and at the moment, the bird is repelled by starting a bird repelling device arranged on a steel tower; if the result of the identification frame is 0, the bird does not exist in the image, and the bird repelling device does not need to be started.
In one embodiment, adding occlusion tags to a bird image dataset comprises:
the corresponding shielding level is set according to the proportion of the shielded area of the birds in the bird image data set, and the shielding label is added to the bird image data set according to the shielding level, wherein the larger the proportion of the shielded area of the birds is, the higher the shielding level corresponding to the bird image data set is.
In a specific embodiment, the bird shielding levels are divided into four shielding levels according to the bird shielding degree, which are respectively: when the ratio of the blocked area of the target is more than 50%, determining that i is 3; when the ratio is 30-50%, i is 2; when the ratio is 10-30%, i is 1; when the proportion is 0-10%, i is 0; wherein i is the shielding level, and in order to further improve the accuracy of model training, the image quantity proportion of various shielding levels should be uniform.
In one embodiment, training a loss function in conjunction with an occlusion tag includes:
detecting whether a shielding label exists in a marking frame in the feature map, if so, acquiring a corresponding shielding grade according to the shielding label, and calculating to obtain a shielding coefficient corresponding to the shielding label according to the shielding grade;
in a specific implementation manner, the same feature map has a plurality of real frames, and if there is an occlusion tag in the feature map, there is an occlusion tag to obtain occlusion information, and further calculate an occlusion coefficient, specifically:
wherein, cover _ coefficient (i) is an occlusion coefficient, level (k) is an actual occlusion level of the kth real frame, bird _ cover (i) is an occlusion label, and cover (i) is an actual occlusion label.
For example, there is an occlusion label bird cover (2) in the image,the block factor corresponding to this image is
And combining the shielding coefficient with the loss function to obtain a shielding compensation coefficient, and performing model training according to the shielding compensation coefficient.
In an embodiment of the invention, the loss function is defined as:
wherein, N is the number of positive samples of the prior frame, c is the identification value of the category confidence coefficient,/is the identification value of the position of the boundary frame corresponding to the prior frame, g is the position parameter of the real frame, LconfAs confidence error, LlocAs the position error, α is the learning rate.
Position error LlocComprises the following steps:
wherein the content of the first and second substances,is an indication parameter whenThe time represents that the ith prior frame is matched with the jth real frame, and the category of the real frame is p;the position parameter g of the real frame is obtained by encoding.
Confidence error LconfComprises the following steps:
wherein the content of the first and second substances,representing the confidence recognition value of the ith prior box relative to the class p,a confidence identification value indicating that the ith prior box does not belong to any class.
wherein the content of the first and second substances,represents the occlusion compensation coefficient of the current target frame, i corresponds to i in the occlusion label bird _ cover (i).
In one embodiment, the higher the occlusion level, the larger the occlusion compensation factor.
In the embodiment of the invention, the coefficient is compensated through shieldingIncreasing confidence error LconfAnd identifying the error value of the category 0 in the identification box, thereby optimizing the loss function to improve the accuracy of the bird identification of the model.
In one embodiment, model training is performed by combining the loss function with the occlusion tag to obtain the bird recognition model, including:
generating a prior frame on the feature map through an SSD model, taking the prior frame with the intersection ratio with the real target being greater than a preset threshold value as a training positive sample, and defining the prior frame with the intersection ratio with the real target being less than or equal to the preset threshold value as a candidate negative sample;
sampling the candidate negative samples, sorting the sampled candidate negative samples in a descending order according to the confidence error, and selecting the candidate negative samples meeting the quantity requirement as training negative samples according to the maximum descending order of the error according to the quantity of the training positive samples and the quantity proportion of the training sample positive samples to the training negative samples;
in the embodiment of the invention, the SSD model confirms the final recognition result through the matching degree of the prior frame and the real target. First the SSD model generates a prior box on the feature map each pixel on the feature map may generate 4 or 6 prior boxes. In the prior frame, positive and negative samples are divided, wherein the prior frame with the intersection ratio (IOU) of the real target being more than a threshold value (0.5) is a training positive sample, and the rest are candidate negative samples. In general, negative samples are much larger than positive samples, and direct training may cause the network to weigh the negative samples too much, thereby causing the loss value of the loss function to be unstable. In order to ensure that positive and negative samples are balanced as much as possible, the embodiment of the invention firstly samples candidate negative samples through the SSD model, performs descending order arrangement according to confidence errors (the smaller the confidence of the recognition background is, the larger the error is) during sampling, and selects the candidate negative sample with the larger error as a training negative sample so as to ensure that the proportion of the training positive and negative samples is close to 1: 3.
And carrying out model training according to the training positive sample and the training negative sample to obtain the bird recognition model.
In the embodiment of the invention, the training positive sample and the training negative sample with the proportion close to 1:3 are adopted for model training, so that the balance of the positive sample and the negative sample in the training process is ensured, and the stability and the reliability of the training can be effectively improved.
In one embodiment, selecting the optimal recognition box from the plurality of recognition boxes as the final recognition result includes:
and filtering the recognition frames with the overlapping degree higher than a preset threshold value in the plurality of recognition frames by adopting a non-maximum value suppression algorithm, and selecting the optimal recognition frame from the plurality of filtered recognition frames as a final recognition result.
In the embodiment of the invention, the identification frames with the overlapping degree higher than the preset threshold are filtered by the filtering method, so that the response speed and accuracy of identification can be effectively improved, and the preset threshold can be set according to requirements, such as 60%, 50% and the like.
In one embodiment, the data enhancement processing includes at least one of horizontal flipping, random cropping, color warping, and randomly acquiring block fields.
The embodiment of the invention increases the diversity of training samples through data enhancement processing, can effectively provide the robustness of the model, avoids overfitting, and can reduce the dependence of the model on certain attributes, thereby improving the generalization capability of the model.
The embodiment of the invention has the following beneficial effects:
according to the embodiment of the invention, the shielding label is added to the training image set, the shielding label is combined with the loss function to form the loss compensation mechanism, model training is carried out in a loss compensation mechanism mode, a bird identification model capable of accurately identifying birds can be obtained through effective training, and the bird identification precision under the condition that the birds are shielded is improved, so that the accuracy of bird identification on the power transmission line can be effectively improved, the bird repelling device can be accurately controlled according to the bird identification result to repel the birds, the working efficiency of the bird repelling device is effectively improved, and the safety of the power transmission line is improved.
Referring to fig. 2, based on the same inventive concept as the above embodiment, an implementation of the present invention provides a bird recognition device for a power transmission line, including:
the data preprocessing module 10 is used for acquiring a bird image data set of the power transmission line, and performing data enhancement processing on the bird image data set to obtain a training image set;
the characteristic image extraction module 20 is configured to add an occlusion label to the training image set, and extract a characteristic image of the training image set by using a convolutional neural network;
the model training module 30 is used for inputting the feature map into the mainstream SSD model to calculate a loss function, and combining the loss function with the shielding label to perform model training to obtain a bird recognition model;
and the bird recognition module 40 is used for acquiring image data to be recognized of the power transmission line, inputting the image data to be recognized into the bird recognition model, outputting a plurality of recognition frames, and selecting the optimal recognition frame from the recognition frames as a final recognition result.
In one implementation, the feature map extraction module 20 includes functionality for:
the corresponding shielding level is set according to the proportion of the shielded area of the birds in the bird image data set, and the shielding label is added to the bird image data set according to the shielding level, wherein the larger the proportion of the shielded area of the birds is, the higher the shielding level corresponding to the bird image data set is.
In one embodiment, model training module 30 is to:
detecting whether a shielding label exists in a marking frame in the feature map, if so, acquiring a corresponding shielding grade according to the shielding label, and calculating to obtain a shielding coefficient corresponding to the shielding label according to the shielding grade;
and combining the shielding coefficient with the loss function to obtain a shielding compensation coefficient, and performing model training according to the shielding compensation coefficient.
In one embodiment, the higher the occlusion level, the larger the occlusion compensation factor.
In one embodiment, model training module 30 is specifically configured to:
generating a prior frame on the feature map through an SSD model, taking the prior frame with the intersection ratio with the real target being greater than a preset threshold value as a training positive sample, and defining the prior frame with the intersection ratio with the real target being less than or equal to the preset threshold value as a candidate negative sample;
sampling the candidate negative samples, sorting the sampled candidate negative samples in a descending order according to the confidence error, and selecting the candidate negative samples meeting the quantity requirement as training negative samples according to the maximum descending order of the error according to the quantity of the training positive samples and the quantity proportion of the training positive samples to the training negative samples;
and carrying out model training according to the training positive sample and the training negative sample to obtain the bird recognition model.
In one implementation, bird identification module 40 is specifically configured to:
and filtering the recognition frames with the overlapping degree higher than a preset threshold value in the plurality of recognition frames by adopting a non-maximum value suppression algorithm, and selecting the optimal recognition frame from the plurality of filtered recognition frames as a final recognition result.
In one embodiment, the loss function is defined as:
wherein, N is the number of positive samples of the prior frame, c is the identification value of the category confidence coefficient,/is the identification value of the position of the boundary frame corresponding to the prior frame, g is the position parameter of the real frame, LconfAs confidence error, LlocAs the position error, α is the learning rate.
In one embodiment, the data enhancement processing includes at least one of horizontal flipping, random cropping, color warping, and randomly acquiring block fields.
An embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored, which, when executed by a processor, implements the steps of the method for bird identification of power transmission lines as described above.
The foregoing is a preferred embodiment of the present invention, and it should be noted that it would be apparent to those skilled in the art that various modifications and enhancements can be made without departing from the principles of the invention, and such modifications and enhancements are also considered to be within the scope of the invention.
Claims (10)
1. A bird identification method for a power transmission line is characterized by comprising the following steps:
acquiring a bird image data set of a power transmission line, and performing data enhancement processing on the bird image data set to obtain a training image set;
adding an occlusion label to the training image set, and extracting a characteristic diagram of the training image set by adopting a convolutional neural network;
inputting the characteristic diagram into a main stream SSD model to calculate a loss function, and combining the loss function with the shielding label to perform model training to obtain a bird recognition model;
acquiring image data to be recognized of the power transmission line, inputting the image data to be recognized into the bird recognition model, outputting a plurality of recognition frames, and selecting an optimal recognition frame from the recognition frames as a final recognition result.
2. The bird identification method of claim 1, wherein the adding of the occlusion tag to the bird image dataset comprises:
setting a corresponding shielding level according to the proportion of the shielded area of the birds in the bird image data set, and adding a shielding label to the bird image data set according to the shielding level, wherein the larger the proportion of the shielded area of the birds is, the higher the corresponding shielding level is.
3. The method of claim 2, wherein training the loss function in combination with the occlusion tag comprises:
detecting whether a shielding label exists in a labeling frame in the feature map, if so, acquiring a corresponding shielding grade according to the shielding label, and calculating to obtain a shielding coefficient corresponding to the shielding label according to the shielding grade;
and combining the shielding coefficient with a loss function to obtain a shielding compensation coefficient, and performing model training according to the shielding compensation coefficient.
4. The bird identification method for the power transmission line according to claim 3, wherein the higher the occlusion level is, the larger the occlusion compensation coefficient is.
5. The bird recognition method for the power transmission line according to claim 1, wherein the model training combining the loss function and the occlusion tag to obtain a bird recognition model comprises:
generating a prior frame on the feature map through the SSD model, taking the prior frame with the intersection ratio with the real target being greater than a preset threshold value as a training positive sample, and defining the prior frame with the intersection ratio with the real target being less than or equal to the preset threshold value as a candidate negative sample;
sampling the candidate negative samples, sorting the sampled candidate negative samples in a descending order according to confidence errors, and selecting the candidate negative samples meeting the quantity requirement as training negative samples according to the quantity of the training positive samples and the quantity proportion of the training sample positive samples to the training negative samples and the maximum and descending order of errors;
and carrying out model training according to the training positive sample and the training negative sample to obtain a bird recognition model.
6. The method for identifying birds on power transmission lines according to claim 1, wherein the step of selecting the optimal identification frame from the plurality of identification frames as the final identification result comprises the steps of:
and filtering the recognition frames with the overlapping degree higher than a preset threshold value in the recognition frames by adopting a non-maximum value suppression algorithm, and selecting the optimal recognition frame from the filtered recognition frames as a final recognition result.
7. The method of bird identification for transmission lines according to claim 1, wherein the loss function is defined as:
wherein N is the number of positive samples of the prior frame, c is the category confidence recognition value, L is the position recognition value of the corresponding boundary frame of the prior frame, g is the position parameter of the real frame, and LconfAs confidence error, LlocAs the position error, α is the learning rate.
8. The method of claim 1, wherein the data enhancement process comprises at least one of horizontal flipping, random cropping, color warping, and randomly collecting block fields.
9. The utility model provides a transmission line birds recognition device which characterized in that includes:
the data preprocessing module is used for acquiring a bird image data set of the power transmission line and performing data enhancement processing on the bird image data set to obtain a training image set;
the characteristic image extraction module is used for adding an occlusion label to the training image set and extracting a characteristic image of the training image set by adopting a convolutional neural network;
the model training module is used for inputting the characteristic diagram into a main stream SSD model to calculate a loss function, and combining the loss function with the shielding label to perform model training to obtain a bird recognition model;
and the bird recognition module is used for acquiring the image data to be recognized of the power transmission line, inputting the image data to be recognized into the bird recognition model, outputting a plurality of recognition frames, and selecting the optimal recognition frame from the recognition frames as a final recognition result.
10. A computer-readable storage medium, characterized in that the computer-readable storage medium has stored thereon a computer program which, when being executed by a processor, carries out the steps of the method for bird identification of electric transmission lines according to any one of claims 1 to 8.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111578096.6A CN114387510A (en) | 2021-12-22 | 2021-12-22 | Bird identification method and device for power transmission line and storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111578096.6A CN114387510A (en) | 2021-12-22 | 2021-12-22 | Bird identification method and device for power transmission line and storage medium |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114387510A true CN114387510A (en) | 2022-04-22 |
Family
ID=81197487
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111578096.6A Pending CN114387510A (en) | 2021-12-22 | 2021-12-22 | Bird identification method and device for power transmission line and storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114387510A (en) |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160343146A1 (en) * | 2015-05-22 | 2016-11-24 | International Business Machines Corporation | Real-time object analysis with occlusion handling |
CN206023133U (en) * | 2016-08-20 | 2017-03-15 | 安徽工贸职业技术学院 | The anti-bird from building nest mechanism of electric wire based on lever construction |
CN111241905A (en) * | 2019-11-21 | 2020-06-05 | 南京工程学院 | Power transmission line nest detection method based on improved SSD algorithm |
CN111476089A (en) * | 2020-03-04 | 2020-07-31 | 上海交通大学 | Pedestrian detection method, system and terminal based on multi-mode information fusion in image |
CN111709374A (en) * | 2020-06-18 | 2020-09-25 | 深圳市赛为智能股份有限公司 | Bird condition detection method and device, computer equipment and storage medium |
CN113076871A (en) * | 2021-04-01 | 2021-07-06 | 华南理工大学 | Fish shoal automatic detection method based on target shielding compensation |
CN113255691A (en) * | 2021-04-15 | 2021-08-13 | 南昌大学 | Method for detecting and identifying harmful bird species target of bird-involved fault of power transmission line |
US20210350162A1 (en) * | 2020-05-07 | 2021-11-11 | Skydio, Inc. | Visual observer for unmanned aerial vehicles |
-
2021
- 2021-12-22 CN CN202111578096.6A patent/CN114387510A/en active Pending
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160343146A1 (en) * | 2015-05-22 | 2016-11-24 | International Business Machines Corporation | Real-time object analysis with occlusion handling |
CN206023133U (en) * | 2016-08-20 | 2017-03-15 | 安徽工贸职业技术学院 | The anti-bird from building nest mechanism of electric wire based on lever construction |
CN111241905A (en) * | 2019-11-21 | 2020-06-05 | 南京工程学院 | Power transmission line nest detection method based on improved SSD algorithm |
CN111476089A (en) * | 2020-03-04 | 2020-07-31 | 上海交通大学 | Pedestrian detection method, system and terminal based on multi-mode information fusion in image |
US20210350162A1 (en) * | 2020-05-07 | 2021-11-11 | Skydio, Inc. | Visual observer for unmanned aerial vehicles |
CN111709374A (en) * | 2020-06-18 | 2020-09-25 | 深圳市赛为智能股份有限公司 | Bird condition detection method and device, computer equipment and storage medium |
CN113076871A (en) * | 2021-04-01 | 2021-07-06 | 华南理工大学 | Fish shoal automatic detection method based on target shielding compensation |
CN113255691A (en) * | 2021-04-15 | 2021-08-13 | 南昌大学 | Method for detecting and identifying harmful bird species target of bird-involved fault of power transmission line |
Non-Patent Citations (2)
Title |
---|
宋威: "自然环境下水果单体目标检测算法的研究与优化", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
苏蒙: "基于改进SSD的多尺度目标检测算法研究", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110705405B (en) | Target labeling method and device | |
CN114119676B (en) | Target detection tracking identification method and system based on multi-feature information fusion | |
CN115797798A (en) | Ecological restoration effect evaluation method based on abandoned mine remote sensing image | |
CN111337789A (en) | Method and system for detecting fault electrical element in high-voltage transmission line | |
CN115359239A (en) | Wind power blade defect detection and positioning method and device, storage medium and electronic equipment | |
CN114998576B (en) | Method, device, equipment and medium for detecting loss of cotter pin of power transmission line | |
CN113515655A (en) | Fault identification method and device based on image classification | |
CN113095441A (en) | Pig herd bundling detection method, device, equipment and readable storage medium | |
CN114359619A (en) | Incremental learning-based power grid defect detection method, device, equipment and medium | |
CN115908786A (en) | Electrical cabinet grounding cable abnormity identification method and system based on deep learning | |
CN114494695A (en) | Intelligent water conservancy urban and rural waterlogging level monitoring and early warning method and device | |
CN111241905A (en) | Power transmission line nest detection method based on improved SSD algorithm | |
CN114581419A (en) | Transformer insulating sleeve defect detection method, related equipment and readable storage medium | |
CN116958841B (en) | Unmanned aerial vehicle inspection system for power distribution line based on image recognition | |
CN111950606B (en) | Knife switch state identification method, device, equipment and storage medium | |
CN112016641A (en) | Method and device for alarming line short circuit fault caused by foreign matter | |
CN114387510A (en) | Bird identification method and device for power transmission line and storage medium | |
CN112669302A (en) | Dropper defect detection method and device, electronic equipment and storage medium | |
CN116229419B (en) | Pedestrian detection method and device | |
CN109993071B (en) | Method and system for automatically identifying and investigating color-changing forest based on remote sensing image | |
CN115908999B (en) | Method for detecting rust of top hardware fitting of distribution pole tower, medium and edge terminal equipment | |
CN106803290A (en) | Transmission line makes an inspection tour recording method and device | |
CN110674827A (en) | Equipment state visual detection method integrating deep learning and morphological filtering | |
CN114970694B (en) | Network security situation assessment method and model training method thereof | |
CN115661117A (en) | Contact net insulator visible light image detection method |
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: 20220422 |