CN114325856A - Power transmission line foreign matter monitoring method based on edge calculation - Google Patents
Power transmission line foreign matter monitoring method based on edge calculation Download PDFInfo
- Publication number
- CN114325856A CN114325856A CN202111449106.6A CN202111449106A CN114325856A CN 114325856 A CN114325856 A CN 114325856A CN 202111449106 A CN202111449106 A CN 202111449106A CN 114325856 A CN114325856 A CN 114325856A
- Authority
- CN
- China
- Prior art keywords
- image
- power transmission
- transmission line
- frame
- key frame
- 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
Abstract
The invention belongs to the technical field of power transmission line monitoring methods, and particularly relates to a power transmission line foreign matter monitoring method based on edge calculation, which comprises the following steps: presetting SOC chips for camera equipment arranged on a high-voltage transmission cable and a tower; obtaining a target detection network model by adopting a YOLOV5 network structure and deploying the target detection network model into an SOC chip; acquiring video data through an SOC chip arranged on the camera equipment and extracting a key frame image from a current video frame sequence; carrying out image preprocessing operation on the extracted key frame image, transmitting the processed image into an SOC chip as the input of a convolutional neural network, carrying out identification detection through a trained network model, and transmitting an identification result to a background; the detection result is displayed in the background and early warning is carried out when foreign matters exist, time delay caused by transmission of a large amount of data in the network can be greatly reduced by using the method, and the real-time requirement in practical application is met.
Description
Technical Field
The invention belongs to the technical field of power transmission line monitoring methods, and particularly relates to a power transmission line foreign matter monitoring method based on edge calculation.
Background
When the transmission line, especially the high-voltage transmission cable, is electrified to work, a strong electromagnetic field is formed in the surrounding area. If the foreign matter enters into the safety protection distance of the power transmission line, the strong electromagnetic field of the power transmission line is discharged, and then the power transmission line is damaged and large-area power failure accidents are caused.
In order to prevent power failure accidents and casualty accidents caused by foreign matters entering the safety protection distance of the power transmission line, the power transmission line protection monitoring device combining an image acquisition monitoring technology and a prompting technology is adopted. The monitoring device comprises an image acquisition module, an image content analysis module and an alarm module; the image content analysis module determines a protection area according to the position relation of the image acquisition module relative to the power transmission line and the safety protection distance of the power transmission line, wherein the specific protection area is a circumference protection area or an arc protection area which takes the power transmission line as the center and takes the safety protection distance as the radius; considering that the foreign object is mostly close to the power transmission line from the ground side (i.e. close to the power transmission line from the lower side of the high-voltage power transmission cable), the protection zone is often arranged at the lower side of the power transmission line. In the protection monitoring process, the image content analysis module extracts the features of the image collected by the image collection module, judges whether foreign matters enter the protection area range according to the extracted feature positions, and starts the alarm module to alarm when judging that the foreign matters are in the protection area range.
However, such a power transmission line protection monitoring device has a high requirement on the background environment of a monitoring area, so as to ensure that an image acquired by the image acquisition module has good contrast and definition. Under the environment with low visibility such as rain fog and night, the foreign matter characteristics in the image collected by the image collecting module are not obvious, the background noise is large, the judgment result of the image content analyzing module is not the same as the state of the foreign matter in the protection area, the misjudgment probability is high, the protection monitoring device cannot reach the effective prompt alarm effect, and even the protection monitoring device cannot work normally.
In the prior art, detection processing is performed on collected data, a network load problem in a data transmission process and a real-time problem caused by transmission delay are less considered, and research and solution methods for the problems also exist in the prior art, but certain problems and defects exist.
Disclosure of Invention
The invention aims to provide a transformer substation power transmission line foreign matter monitoring method based on edge calculation, aiming at solving the problems in the prior art, and the method can greatly reduce the time delay caused by the transmission of a large amount of data in a network and meet the real-time requirement in practical application.
The technical scheme of the invention is as follows:
the transformer substation power transmission line foreign matter monitoring method based on edge calculation comprises the following steps:
s1, presetting SOC chips for camera equipment arranged on a high-voltage transmission cable and a tower;
s2, training through a large number of data sets by adopting a YOLOV5 network structure to obtain a target detection network model with weight, and deploying the target detection network model into an SOC chip;
s3, the camera equipment acquires video data, acquires the video data through an SOC chip arranged on the camera equipment and extracts a key frame image from the current video frame sequence;
s4, performing image preprocessing operation on the key frame image extracted in the step S3, firstly performing smooth denoising processing on the extracted key frame image by adopting median filtering, and then performing Laplace sharpening processing on the denoised image to improve the image contrast and enhance the target edge information in the image;
s5, transmitting the image processed in the S4 to an SOC chip as the input of a convolutional neural network, carrying out recognition detection through a trained network model, and transmitting a recognition result to a background;
and S6, displaying in the background according to the detection result and early warning when foreign matters exist.
Specifically, the step S3 of extracting the key frame image includes performing difference calculation on two adjacent frames, recording a difference S between a current video frame and a previous frame, setting a threshold T (threshold) in consideration of natural variation of an environment, and skipping the current video frame when the difference d is less than T; and when d > is T, taking the frame as a key frame image to carry out subsequent processing.
Specifically, the difference calculation process includes the following steps:
s1, recording the image of the nth frame and the image of the (n-1) th frame in a video sequence as fnAnd fn-1；
S2, carrying out gray level processing on the image, and recording the gray value of the corresponding pixel point as fn(x, y) and fn-1(x，y)；
S3, carrying out normalization processing on the gray value,
s4, subtracting the gray values of the corresponding pixels of the two images to obtain the absolute value Dn(x，y)＝|f’n(x，y)-f’n-1(x，y)|；
S5, summing absolute values of difference values of all pixel points, wherein D is ═ Sigma Dn(x，y)；
And S6, representing the image change condition through the sum d of the absolute values of the difference values, comparing the image change condition with a threshold value T, and judging whether the image change condition can be used as a key frame.
Specifically, the input end of the YOLOV5 network structure adopts a Mosaic data enhancement mode to randomly scale, randomly cut, randomly arrange and splice images; the output end adopts GIOU _ Loss as the Loss function of the Bounding box.
The method is suitable for monitoring the foreign matters in the power transmission line of the edge computing, and aims to realize real-time monitoring on the foreign matters in the power transmission line of the transformer substation by adapting an Soc chip with computing capability in edge equipment (a network camera) and deploying a neural network model of the Soc chip. At present, a plurality of problems about the foreign matter identification of the power transmission line of the transformer substation are not effectively solved: (1) the problem of network load in the data transmission process is not considered, data acquired by the data acquisition end need to be transmitted back to the background for processing, and transmission, processing and feedback of mass data inevitably cause large time delay and are difficult to meet the real-time requirement. (2) Most of the prior patents adopt the traditional convolutional neural network model, have large volume and are difficult to balance the detection speed and the detection precision. Aiming at the problems, the invention provides a power transmission line foreign matter monitoring improvement method suitable for edge calculation, a neural network model with better balance of detection speed and precision is selected, and after a large number of data sets are trained, the neural network model is deployed in edge equipment in combination with an Soc chip.
With the development of artificial intelligence and edge calculation, the real-time monitoring of foreign matters in the power transmission line becomes possible by deploying intelligent monitoring on the power transmission line and the periphery, and the environmental protection and management are more and more emphasized. Massive data processing is faced in massive deployment of intelligent monitoring, the background data processing capability is also challenged in a mode of transmitting to background processing, and a certain network delay problem is caused. While the traditional data processing mode is broken through edge calculation, the balance between the model size and the speed precision is considered more for the selection of the neural network model.
The invention provides a method for monitoring and improving foreign matters of a power transmission line, which is suitable for edge calculation and has the following beneficial effects compared with the prior patent: 1. the Soc chip with certain calculation power is deployed in the network camera, the acquired data stream is directly subjected to real-time calculation processing at the front end, only the detection result is returned to the background for checking, and massive data processing is distributed to each edge device through edge calculation, so that data transmission is reduced, the processing pressure of a large amount of data in the background can be effectively relieved, and the real-time performance of monitoring is ensured; 2. the image preprocessing is carried out by adopting median filtering denoising and Laplace sharpening to highlight the edge information of the target aiming at the specific environment of the power transmission line, so that the detection precision can be effectively improved; the adaptation problem of the neural network model to the edge calculation is also considered, the calculation force of the edge device has certain limitation, the selected YOLOV5 network model is improved on the basis of the previous version, the calculation speed and the detection precision can be well balanced, and therefore the overall stability is improved, and the balance among the model volume, the detection precision and the detection speed is considered more in the edge calculation mode.
Drawings
FIG. 1 is a schematic process flow diagram of the present invention;
fig. 2 is a schematic diagram of a data transmission processing mode according to the present invention.
Detailed Description
The technical solution of the present invention is described in detail below with reference to the accompanying drawings and the detailed description.
Fig. 1 is a schematic processing flow diagram of a transformer substation power transmission line foreign matter monitoring method based on edge calculation, which specifically includes the following steps:
the method comprises the steps that S1, SOC chips are preset for camera equipment arranged on a high-voltage transmission cable and a tower, the SOC chips adopt Haisi Hi35191A V100 and integrate a powerful programmable neural network inference engine and a vector DSP, application of various intelligent algorithms is supported, 2.0 TOPS computing performance is achieved, various applications such as complete api and tool chain-face detection/recognition and target detection/tracking can be achieved, and high-pass, Botong, Union, Yiyi, Hua Haisi, Meiman and Intel can also be used;
s2, training through a large number of data sets by adopting a YOLOV5 network structure to obtain a target detection network model with weight, and deploying the target detection network model into an SOC chip;
s3, video data are obtained by the camera equipment, the video data are obtained by an SOC chip arranged on the camera equipment, a key frame image is extracted from a current video frame sequence, mainly aiming at river foreign matter detection, the condition that the key frame is selected to reflect increase, decrease, displacement, change and the like of suspicious foreign matters in a river channel, the key frame image extraction comprises the steps of carrying out difference calculation on two adjacent frames, recording the difference value s between the current video frame and the previous frame, considering natural change of the environment, setting a threshold value T (threshold), and skipping the frame when the difference value d is less than T; when d ═ T, the frame is taken as a key frame image for subsequent processing; s4, performing image preprocessing operation on the key frame image extracted in the step S3, firstly performing smooth denoising processing on the extracted key frame image by adopting median filtering, and protecting target edge information in the image while denoising, and then performing Laplace sharpening processing on the denoised image, so that the image contrast is improved, the target edge information in the image is enhanced, and the identification of a convolutional neural network is facilitated;
s5, transmitting the image processed in the S4 to an SOC chip as the input of a convolutional neural network, carrying out recognition detection through a trained network model, and transmitting a recognition result to a background;
and S6, displaying in the background according to the detection result and early warning when foreign matters exist.
The difference calculation process described in step S3 includes the steps of:
s1, recording the image of the nth frame and the image of the (n-1) th frame in a video sequence as fnAnd fn-1；
S2, carrying out gray level processing on the image, and recording the gray value of the corresponding pixel point as fn(x, y) and fn-1(x，y)；
S3, carrying out normalization processing on the gray value,
s4, subtracting the gray values of the corresponding pixels of the two images to obtain the absolute value Dn(x，y)＝|f’n(x，y)-f’n-1(x，y)|；
S5, summing absolute values of difference values of all pixel points, wherein D is ═ Sigma Dn(x，y)；
And S6, representing the image change condition through the sum d of the absolute values of the difference values, comparing the image change condition with a threshold value T, and judging whether the image change condition can be used as a key frame.
The input end of the YOLOV5 network structure adopts a Mosaic data enhancement mode to randomly zoom, randomly cut, randomly arrange and splice images, so that the detection effect on small targets can be effectively improved, the self-adaptive anchor frame calculation is realized, the initial length and width anchor frame is set, the prediction frame is output on the basis of the initial anchor frame in the training process and is compared with the real frame, the difference value between the initial anchor frame and the real frame is calculated, the reverse update is realized, and the network parameters are iterated.
Backbone: the Focus structure adopts a slicing operation to slice an original 608 × 608 × 3 image into a 304 × 304 × 12 feature map, and then the feature map is changed into a 304 × 304 × 32 feature map after 32 times of convolution operations.
And (6) selecting Neck: the FPN + PAN structure is adopted, the FPN layer conveys strong semantic features from top to bottom, the feature pyramid containing the two PAN structures conveys strong positioning features from bottom to top, and feature fusion is carried out on different detection layers from different backbone layers.
Output of the YOLOV5 network structure: the GIOU _ Loss is used as a Loss function of the Bounding box, so that the problem that the Loss function cannot be led when a prediction frame and a target frame in a DIOU _ Loss Loss function of the YOLOV4 are not intersected and the problems that the two prediction frames are the same in size, the two IOUs are the same and the IOU _ Loss cannot distinguish different intersection conditions are effectively solved.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention and not to limit it; although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art will understand that: modifications to the specific embodiments of the invention or equivalent substitutions for parts of the technical features may be made; without departing from the spirit of the present invention, it is intended to cover all aspects of the invention as defined by the appended claims.
Claims (4)
1. A power transmission line foreign matter monitoring method based on edge calculation is characterized by comprising the following steps:
s1, presetting SOC chips for camera equipment arranged on a high-voltage transmission cable and a tower;
s2, training through a large number of data sets by adopting a YOLOV5 network structure to obtain a target detection network model with weight, and deploying the target detection network model into an SOC chip;
s3, the camera equipment acquires video data, acquires the video data through an SOC chip arranged on the camera equipment and extracts a key frame image from the current video frame sequence;
s4, performing image preprocessing operation on the key frame image extracted in the step S3, firstly performing smooth denoising processing on the extracted key frame image by adopting median filtering, and then performing Laplace sharpening processing on the denoised image to improve the image contrast and enhance the target edge information in the image;
s5, transmitting the image processed in the S4 to an SOC chip as the input of a convolutional neural network, carrying out recognition detection through a trained network model, and transmitting a recognition result to a background;
and S6, displaying in the background according to the detection result and early warning when foreign matters exist.
2. The method for monitoring foreign matters in power transmission lines of transformer substations based on edge calculation as claimed in claim 1, wherein the step S3 of extracting the key frame image includes performing difference calculation on two adjacent frames, recording the difference S between the current video frame and the previous frame, setting a threshold T (threshold) in consideration of natural variation of the environment, and skipping the frame when the difference d is less than T; when d ═ T, the frame is subjected to subsequent processing as a key frame image.
3. The method for monitoring foreign matters on the power transmission line of the transformer substation based on the edge calculation as claimed in claim 2, wherein the differential calculation process comprises the following steps:
s1, recording the image of the nth frame and the image of the (n-1) th frame in a video sequence as fnAnd fn-1；
S2, carrying out gray level processing on the image, and recording the gray value of the corresponding pixel point as fn(x, y) and fn-1(x，y)；
S3, making a return to the gray valueA treatment of normalization is carried out,
s4, subtracting the gray values of the corresponding pixels of the two images to obtain the absolute value Dn(x，y)＝|f′n(x，y)-f′n-1(x，y)|；
S5, summing absolute values of difference values of all pixel points, wherein D is ═ Sigma Dn(x，y)；
And S6, representing the image change condition through the sum d of the absolute values of the difference values, comparing the image change condition with a threshold value T, and judging whether the image change condition can be used as a key frame.
4. The method for monitoring the foreign matters in the power transmission lines of the transformer substation based on the edge calculation is characterized in that a Mosaic data enhancement mode is adopted at the input end of the YOLOV5 network structure, and images are randomly zoomed, randomly cut, randomly arranged and spliced; the output end adopts GIOU _ Loss as the Loss function of the Bounding box.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111449106.6A CN114325856A (en) | 2021-11-30 | 2021-11-30 | Power transmission line foreign matter monitoring method based on edge calculation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111449106.6A CN114325856A (en) | 2021-11-30 | 2021-11-30 | Power transmission line foreign matter monitoring method based on edge calculation |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114325856A true CN114325856A (en) | 2022-04-12 |
Family
ID=81048456
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111449106.6A Pending CN114325856A (en) | 2021-11-30 | 2021-11-30 | Power transmission line foreign matter monitoring method based on edge calculation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114325856A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114626024A (en) * | 2022-05-12 | 2022-06-14 | 北京吉道尔科技有限公司 | Internet infringement video low-consumption detection method and system based on block chain |
| CN115115822A (en) * | 2022-06-30 | 2022-09-27 | 小米汽车科技有限公司 | Vehicle-end image processing method and device, vehicle, storage medium and chip |
-
2021
- 2021-11-30 CN CN202111449106.6A patent/CN114325856A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114626024A (en) * | 2022-05-12 | 2022-06-14 | 北京吉道尔科技有限公司 | Internet infringement video low-consumption detection method and system based on block chain |
| CN115115822A (en) * | 2022-06-30 | 2022-09-27 | 小米汽车科技有限公司 | Vehicle-end image processing method and device, vehicle, storage medium and chip |
| CN115115822B (en) * | 2022-06-30 | 2023-10-31 | 小米汽车科技有限公司 | Vehicle-end image processing method and device, vehicle, storage medium and chip |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110232380B (en) | Fire night scene restoration method based on Mask R-CNN neural network | |
| WO2021088300A1 (en) | Rgb-d multi-mode fusion personnel detection method based on asymmetric double-stream network | |
| KR20200007084A (en) | Ship detection method and system based on multi-dimensional features of scene | |
| CN111209810A (en) | Bounding box segmentation supervision deep neural network architecture for accurately detecting pedestrians in real time in visible light and infrared images | |
| CN111046880A (en) | Infrared target image segmentation method and system, electronic device and storage medium | |
| CN114325856A (en) | Power transmission line foreign matter monitoring method based on edge calculation | |
| CN111797712B (en) | Remote sensing image cloud and cloud shadow detection method based on multi-scale feature fusion network | |
| CN110852179B (en) | Suspicious personnel invasion detection method based on video monitoring platform | |
| CN114463677B (en) | Safety helmet wearing detection method based on global attention | |
| CN114648714A (en) | YOLO-based workshop normative behavior monitoring method | |
| CN116846059A (en) | Edge detection system for power grid inspection and monitoring | |
| CN111667655A (en) | Infrared image-based high-speed railway safety area intrusion alarm device and method | |
| Zhao et al. | Image dehazing based on haze degree classification | |
| Aarathi et al. | Vehicle color recognition using deep learning for hazy images | |
| CN113177439B (en) | Pedestrian crossing road guardrail detection method | |
| CN114399734A (en) | Forest fire early warning method based on visual information | |
| CN102324033B (en) | Image processing method of intelligent early warning and emergency response system for wind power safety | |
| CN117423157A (en) | Mine abnormal video action understanding method combining migration learning and regional invasion | |
| CN113052139A (en) | Deep learning double-flow network-based climbing behavior detection method and system | |
| Xu et al. | Research on pedestrian detection algorithm based on deep learning | |
| Chen et al. | Facemask Detection Based on Double Convolutional Neural Networks | |
| Green et al. | The detection and quantification of persons in cluttered beach scenes using neural network-based classification | |
| CN109766763A (en) | A kind of forest fire detection method and system | |
| Yamaguchi et al. | Road crack detection interpreting background images by convolutional neural networks and a self‐organizing map | |
| CN212873683U (en) | High-speed railway safety zone intrusion alarm device based on infrared image |
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 |