CN113695256B - Power grid foreign matter detection and identification method and device - Google Patents
Power grid foreign matter detection and identification method and device Download PDFInfo
- Publication number
- CN113695256B CN113695256B CN202110950237.6A CN202110950237A CN113695256B CN 113695256 B CN113695256 B CN 113695256B CN 202110950237 A CN202110950237 A CN 202110950237A CN 113695256 B CN113695256 B CN 113695256B
- Authority
- CN
- China
- Prior art keywords
- frame
- foreign
- information
- foreign matter
- determining
- 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.)
- Active
Links
Images
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B07—SEPARATING SOLIDS FROM SOLIDS; SORTING
- B07C—POSTAL SORTING; SORTING INDIVIDUAL ARTICLES, OR BULK MATERIAL FIT TO BE SORTED PIECE-MEAL, e.g. BY PICKING
- B07C5/00—Sorting according to a characteristic or feature of the articles or material being sorted, e.g. by control effected by devices which detect or measure such characteristic or feature; Sorting by manually actuated devices, e.g. switches
- B07C5/34—Sorting according to other particular properties
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B07—SEPARATING SOLIDS FROM SOLIDS; SORTING
- B07C—POSTAL SORTING; SORTING INDIVIDUAL ARTICLES, OR BULK MATERIAL FIT TO BE SORTED PIECE-MEAL, e.g. BY PICKING
- B07C5/00—Sorting according to a characteristic or feature of the articles or material being sorted, e.g. by control effected by devices which detect or measure such characteristic or feature; Sorting by manually actuated devices, e.g. switches
- B07C5/36—Sorting apparatus characterised by the means used for distribution
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B08—CLEANING
- B08B—CLEANING IN GENERAL; PREVENTION OF FOULING IN GENERAL
- B08B13/00—Accessories or details of general applicability for machines or apparatus for cleaning
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B08—CLEANING
- B08B—CLEANING IN GENERAL; PREVENTION OF FOULING IN GENERAL
- B08B7/00—Cleaning by methods not provided for in a single other subclass or a single group in this subclass
- B08B7/0035—Cleaning by methods not provided for in a single other subclass or a single group in this subclass by radiant energy, e.g. UV, laser, light beam or the like
- B08B7/0042—Cleaning by methods not provided for in a single other subclass or a single group in this subclass by radiant energy, e.g. UV, laser, light beam or the like by laser
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (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)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Optics & Photonics (AREA)
- Probability & Statistics with Applications (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Image Analysis (AREA)
- Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
Abstract
The application discloses a method and a device for detecting and identifying foreign matters in a power grid, which are characterized in that firstly, regional suggestion frames corresponding to foreign matters in a one-to-one mode are determined, and positive samples and negative samples are divided. And extracting the region suggestion frame of the positive sample, and generating a mask result of the region suggestion frame of the positive sample. The foreign object image is input into the loss function to determine position information of the candidate frame, category information of the candidate frame, and mask information of the foreign object. The region proposal frame is acquired, and an offset of the region proposal frame to the actual position of the foreign object image is determined. Screening the candidate frames, determining target candidate frames, and outputting foreign matter information, cable classification information and pixel area information. The feature images of different scales are extracted from the foreign object image for detecting the foreign object, and a segmentation operation is performed. Since the dimensions and feelings of each feature map are different, foreign matter of different sizes can be detected by processing and analyzing each feature map, and more accurate foreign matter region information can be extracted.
Description
Technical Field
The application relates to the technical field of foreign matter identification of laser removers, in particular to a method and a device for detecting and identifying foreign matters in a power grid.
Background
In recent years, as the pace of power grid construction increases, more overhead power transmission lines are built in suburban towns. Overhead transmission lines are easily entangled by foreign matter such as kites, kite lines, agricultural plastic cloths, sun-shading nets, etc. Foreign matter on the overhead transmission line is easy to cause tripping accidents such as single-phase grounding, interphase short circuit and the like of the ground wire, and the safe and stable operation of a power grid is seriously influenced.
In the prior art, a video monitoring mode is generally adopted to identify the foreign matters of the power grid in the overhead transmission line. The existing video monitoring method comprises the steps of firstly obtaining foreign matter images, then modeling a foreground and a background, then pertinently extracting power grid foreign matter and background characteristics with distinction, and finally classifying the extracted characteristics through a classifier, so that detection and identification of the power grid foreign matter are realized.
However, the video monitoring mode has poor robustness, and the accuracy of detecting and identifying the foreign matters in the power grid is low when the use scene is changed. For example, when the usage scene is changed from a transformer substation to other scenes, the detection precision of the method is reduced sharply, and when the picture acquired by the camera is changed from a near scene to a far scene, the recognition precision is reduced.
Disclosure of Invention
The application discloses a method and a device for detecting and identifying foreign matters in a power grid, which are used for solving the problems that in the detection and identification of the foreign matters in the power grid in the prior art, the robustness of a video monitoring mode is poor, and the accuracy of the detection and identification of the foreign matters in the power grid is low when a use scene is changed. For example, when a use scene is changed from a transformer substation to other scenes, the detection precision of the method is reduced sharply, and when a picture acquired by a camera is changed from a near scene to a far scene, the recognition precision is reduced.
The first aspect of the application discloses a method for detecting and identifying foreign matters in a power grid, which comprises the following steps:
acquiring a power grid suspension foreign matter data set, wherein the power grid suspension foreign matter data set comprises a plurality of foreign matter images;
acquiring a plurality of feature images of any foreign object image aiming at the any foreign object image;
determining a region suggestion frame corresponding to any foreign object image according to the plurality of feature images;
acquiring a preset default frame, acquiring a preset real sample value, and determining a positive sample and a negative sample according to the power grid hanging foreign matter data set, the default frame and the real sample value; the default frame refers to frames of different sizes and aspect ratios generated at fixed positions of the foreign object image;
acquiring a region suggestion frame of the positive sample and acquiring a feature map of the positive sample;
determining a mask result of the region suggestion frame of the positive sample according to the region suggestion frame of the positive sample and the feature map of the positive sample;
determining position information of a candidate frame, category information of the candidate frame and mask information of foreign matters according to mask results of the positive sample, the negative sample and the region suggestion frame of the positive sample;
acquiring a preset area proposal frame, and determining the offset from the area proposal frame to the actual position of the foreign object image according to the area proposal frame;
determining a target candidate frame according to the position information of the candidate frame, the category information of the candidate frame, the mask information of the foreign matter and the offset;
determining foreign matter information, classification information of cables and pixel area information according to the target candidate frame;
determining a target firing line according to the foreign matter information, the classification information of the cable and the pixel area information, wherein the target firing line is a firing reference line which is parallel to the cable and passes through a foreign matter area on the target candidate frame;
and burning and removing the foreign matters according to the target burning line.
Optionally, the determining, according to the plurality of feature maps, an area suggestion box corresponding to the any foreign object image includes:
determining a plurality of pending area suggestion boxes corresponding to the feature maps according to the feature maps;
and determining the region suggestion frame corresponding to any foreign object image according to the plurality of the region suggestion frames to be determined.
Optionally, the determining the positive sample and the negative sample according to the grid hanging foreign matter dataset, the default frame and the preset real sample value includes:
acquiring a preset cross ratio threshold;
judging whether the intersection ratio of the default frame and the real sample value exceeds the intersection ratio threshold value or not according to the power grid suspended foreign matter data set, and if yes, determining to be a positive sample; if not, the negative sample is determined.
Optionally, the burning and cleaning the foreign matter according to the target burning line includes:
determining the minimum laser power density required by fusing the foreign matters according to the foreign matters information;
and burning and removing the foreign matters according to the target burning line and the minimum laser power density required by fusing the foreign matters.
Optionally, the offset includes: the abscissa of the upper left corner of the region proposal frame, the ordinate of the upper left corner of the region proposal frame, the width of the region proposal frame, and the height of the region proposal frame.
The second aspect of the application discloses a power grid foreign matter detection and identification device, the power grid foreign matter detection and identification device is applied to the power grid foreign matter detection and identification method disclosed in the first aspect of the application, the power grid foreign matter detection and identification device includes:
the foreign matter data set acquisition module is used for acquiring a power grid hanging foreign matter data set, wherein the power grid hanging foreign matter data set comprises a plurality of foreign matter images;
the characteristic map acquisition module is used for acquiring a plurality of characteristic maps of any foreign object image aiming at any foreign object image;
the region suggestion frame determining module is used for determining a region suggestion frame corresponding to any foreign object image according to the plurality of feature images;
the sample dividing module is used for acquiring a preset default frame, acquiring a preset real sample value and determining a positive sample and a negative sample according to the power grid suspended foreign matter data set, the default frame and the real sample value; the default frame refers to frames of different sizes and aspect ratios generated at fixed positions of the foreign object image;
the positive sample parameter module is used for acquiring a region suggestion frame of the positive sample and acquiring a feature map of the positive sample;
a mask result determining module, configured to determine a mask result of the region suggestion frame of the positive sample according to the region suggestion frame of the positive sample and the feature map of the positive sample;
the information acquisition module is used for determining the position information of the candidate frame, the category information of the candidate frame and the mask information of the foreign matters according to the positive sample, the negative sample and the mask result of the region suggestion frame of the positive sample;
the offset determining module is used for acquiring a preset area proposal frame and determining the offset from the area proposal frame to the actual position of the foreign object image according to the area proposal frame;
a target candidate frame determining module, configured to determine a target candidate frame according to the position information of the candidate frame, the category information of the candidate frame, the mask information of the foreign object, and the offset;
the target information acquisition module is used for determining foreign matter information, cable classification information and pixel area information according to the target candidate frame;
the reference line determining module is used for determining a target burning line according to the foreign matter information, the classification information of the cable and the pixel area information, wherein the target burning line is a burning reference line which is parallel to the cable and passes through a foreign matter area on the target candidate frame;
and the foreign matter removing module is used for burning and removing the foreign matters according to the target burning line.
Optionally, the area suggestion box determining module includes:
a pending area suggestion frame acquisition unit configured to determine a plurality of pending area suggestion frames corresponding to the plurality of feature maps according to the plurality of feature maps;
and the region suggestion frame acquisition unit is used for determining the region suggestion frame corresponding to any foreign object image according to the plurality of pending region suggestion frames.
Optionally, the sample dividing module includes:
the threshold value acquisition unit is used for acquiring a preset cross ratio threshold value;
the sample dividing unit is used for judging whether the cross ratio of the default frame to the real sample value exceeds the cross ratio threshold value according to the power grid hanging foreign matter data set, and if so, determining the cross ratio as a positive sample; if not, the negative sample is determined.
Optionally, the foreign matter removal module includes:
the density acquisition unit is used for determining the minimum laser power density required by fusing the foreign matters according to the foreign matters information;
the burning clearing unit is used for burning and clearing the foreign matters according to the target burning line and the minimum laser power density required by the fusing of the foreign matters.
Optionally, the offset includes: the abscissa of the upper left corner of the region proposal frame, the ordinate of the upper left corner of the region proposal frame, the width of the region proposal frame, and the height of the region proposal frame.
The application relates to the field of foreign matter identification of laser cleaners and discloses a method and a device for detecting and identifying foreign matters in a power grid. Positive and negative samples in the grid hanging foreign object data set are then determined. And further extracting a region suggestion frame of the positive sample, and generating a mask result of the region suggestion frame of the positive sample. After the selection of the positive and negative samples is completed, the positive and negative samples are input into a loss function for calculation, and the position information of the candidate frame, the category information of the candidate frame and the mask information of the foreign matter are determined. And then acquiring a preset area proposal frame, determining the offset from the area proposal frame to the actual position of the foreign object image, further screening candidate frames, and determining target candidate frames. And finally, outputting foreign matter information, cable classification information and pixel area information according to the target candidate frame, and completing recognition and removal of the foreign matter according to the information.
The method and the device are used for detecting the foreign matters by extracting the feature images with different scales from the foreign matter images, and perform one-time segmentation operation after the detection result. Since the dimensions and feelings of each feature map are different, the processing and analysis of each feature map can detect foreign objects of different sizes and extract more accurate foreign object region information.
Drawings
In order to more clearly illustrate the technical solutions of the present application, the drawings that are needed in the embodiments will be briefly described below, and it will be obvious to those skilled in the art that other drawings can be obtained from these drawings without inventive effort.
Fig. 1 is a schematic workflow diagram of a method for detecting and identifying foreign matters in a power grid according to an embodiment of the present application;
fig. 2 is a schematic structural diagram of obtaining different laser power densities in a method for detecting and identifying a foreign object in a power grid according to an embodiment of the present application;
fig. 3 is a schematic structural diagram of a device for detecting and identifying foreign matters in a power grid according to an embodiment of the present application.
Detailed Description
In order to solve the problems that in the prior art, in detection and identification of the foreign matters of the power grid, the robustness of a video monitoring mode is poor, and when a use scene is changed, the detection and identification precision of the foreign matters of the power grid is low. For example, when a using scene is changed from a transformer substation to other scenes, the method detection precision can be reduced sharply, and when a picture acquired by a camera is changed from a near scene to a far scene, the technical problem of the reduction of the recognition precision can be also caused.
The first embodiment of the application discloses a method for detecting and identifying a foreign object on a power grid, referring to a workflow diagram shown in fig. 1, the method for detecting and identifying a foreign object on a power grid includes:
step S101, a grid hanging foreign matter data set is acquired, the grid hanging foreign matter data set comprising a plurality of foreign matter images.
In particular, during the testing of deep learning models, the data set used for training is as important as the training model, and a high quality data set tends to improve the quality of model training and the accuracy of prediction. Through finding out the data, the number of the existing power grid foreign matter pictures is small, the background interference is large, and the existing power grid foreign matter pictures cannot be used as data pictures in the project to participate in training, so that a power grid hanging foreign matter data set needs to be established in advance.
The phenomenon of foreign matter winding on the transmission line generally occurs in the field area far away from the urban area, so that when the power grid hanging foreign matter data set is constructed, a foreign matter erection and shooting foreign matter picture platform is selected to be constructed in the field, and the foreign matter shooting background is true as far as possible, and foreign matter images are shot at different distances. Because the place where the foreign matter is erected is troublesome to move, the electric wire and the foreign matter are fixed at a place with wide visual field, and a movable foreign matter picture shooting platform is built by adopting a notebook, a mobile power supply and a small trailer. When the data set is shot, the foreign matters are hung on the steel-cored aluminum stranded wire of the 300-meter outer roof platform, a 52-time camera is adopted for shooting, and the background firstly considers the ideal situation, namely, a sunny sky, windless or breeze environment.
Common line hanging foreign matters generally comprise branches, kites, balloons and film fragments of agricultural plastic greenhouses, when a power grid hanging foreign matter data set is constructed, common foreign matters such as advertisement banners, kites, cloth strips, plastic bags and the like are firstly selected, and because the morphological differences of different kites are large, four different types of kites are always selected when the data set is shot.
In order to realize the recognition of the net-hanging foreign matters in the electric power high-voltage line laser foreign matter removal scene and the detection of laser ignition points, the embodiment provides a power grid foreign matter detection recognition method based on Mask R-CNN instance segmentation, a lightweight deep learning neural network is adopted, different scale features are fully extracted from an image, an area proposal frame is generated, and a final result, foreign matter information, classification information of a cable and pixel area information are extracted through non-maximum suppression.
The embodiment mainly improves the traditional method from the aspect of feature extraction, extracts feature response graphs with different scales from images through a deep convolution layer structure for target detection, and performs one-time segmentation operation after the detection result. Since the scale size and receptive field of each feature map are different, processing and analysis of each feature map will detect objects of different sizes and extract more accurate object region information, which is difficult to achieve with conventional methods.
Step S102, a plurality of feature images of any foreign object image are acquired for any foreign object image.
Step S103, determining an area suggestion frame corresponding to any foreign object image according to the feature images.
In some embodiments of the present application, the determining, according to the plurality of feature maps, a region suggestion box corresponding to the any one foreign object image includes:
and determining a plurality of pending area suggestion boxes corresponding to the feature maps according to the feature maps.
And determining the region suggestion frame corresponding to any foreign object image according to the plurality of the region suggestion frames to be determined.
Specifically, step S102 and step S103 are based on the generation of a multi-layer feature map region suggestion frame of a feature pyramid, and after an image is read in, a series of feature maps with different depths are generated through multi-layer convolution, and on the series of feature maps, a series of region suggestion frames with the same size and different depths are generated by using an ROI alignment technology, and a final region suggestion frame is obtained through a non-maximum suppression algorithm.
Step S104, a preset default frame is obtained, a preset real sample value is obtained, and a positive sample and a negative sample are determined according to the power grid suspended foreign matter data set, the default frame and the real sample value. The default frame refers to frames of different sizes and aspect ratios generated at fixed positions of the foreign object image.
In some embodiments of the present application, the determining the positive and negative samples according to the grid hanging foreign matter dataset, the default box and the preset real sample value includes:
and acquiring a preset cross ratio threshold.
And judging whether the cross ratio of the default frame to the real sample value exceeds the cross ratio threshold value according to the power grid suspended foreign matter data set, and if so, determining the cross ratio as a positive sample. If not, the negative sample is determined.
Specifically, in default boxes of various dimensions and aspect ratios, there will be a small portion of the alien image that partially coincides with the true sample value (GroundTruth). At this time, the coincidence sample is set as a positive sample, and the rest of the foreign matter image is regarded as a negative sample. Wherein the coincidence of the default box and the true sample value is measured by the cross ratio.
If the A frame and the B frame exist, the cross ratio calculation mode is as follows: the intersection ratio is the ratio of the intersection range of A and B to the union of A and B. The algorithm takes the cross ratio threshold value of 0.5 in positive and negative sample judgment, namely, the cross ratio of the real sample value and the cross ratio of the real sample value exceeds 0.5 in all default frames, and the algorithm is regarded as a positive sample and is regarded as a negative sample. During training, all positive samples will participate in the training. The number of negative samples is far more than that of positive samples, and the excessive negative samples are unfavorable for training of the network, so that the number of the negative samples and the positive samples is randomly taken to be 3:1.
Step S105, acquiring a region suggestion box of the positive sample, and acquiring a feature map of the positive sample.
And S106, determining a mask result of the region suggestion frame of the positive sample according to the region suggestion frame of the positive sample and the feature map of the positive sample.
Specifically, a region suggestion frame of the positive sample is extracted, a feature map is generated after ROI alignment and primary deconv processing, sigmoid processing is carried out on each category, and finally a mask result of the region suggestion frame is generated.
Step S107, determining the position information of the candidate frame, the category information of the candidate frame and the mask information of the foreign matter according to the mask results of the positive sample, the negative sample and the region suggestion frame of the positive sample.
Specifically, after the selection of positive and negative samples is completed, the foreign object images are respectively input into a loss function for calculation, so that a basis is provided for network parameter adjustment. Mask RCNN is an end-to-end deep learning network, and finally, the position information of the candidate frame, the category represented by the candidate frame and the Mask information of the foreign matter are output. The position information, the category information and the mask information are listed in the same loss function, and the loss function is:
L＝L_cls+L_box+L_mask；
lcls is a Softmax loss function, and the working principle of Lcls is the same as that of a common image classification network, and is used for calculating the classification accuracy of an algorithm.
Lbox is the position Loss function Smooth L1 Loss, which is responsible for the Loss of regression position.
Lmask is a loss function of Mask, and for each ROI, mask branches have outputs with Km m dimension, which encodes K masks with m, i.e. K masks with m resolution binary values.
After the calculation of the loss function is completed, the back propagation can be started to update the parameters.
Step S108, acquiring a preset area proposal frame, and determining the offset of the area proposal frame to the actual position of the foreign object image according to the area proposal frame.
Specifically, the main role of the region proposal box in the recognition process is to provide the deep neural network with an initial value. Because the convolutional neural network itself has difficulty in generating a corresponding detection frame only through convolution, pooling and other operations without an anchor frame providing an initial value.
In some embodiments of the present application, the offset includes: the abscissa of the upper left corner of the region proposal frame, the ordinate of the upper left corner of the region proposal frame, the width of the region proposal frame, and the height of the region proposal frame.
Specifically, after the convolutional network obtains the initial value, the category and the position of the default frame are predicted. Since the default frame is simply a frame of different size and aspect ratio generated at a fixed location and the targets of different locations and dimensions cannot be accurately predicted, the algorithm needs to predict the offset of the region proposal frame to the actual location. The offset of this position includes 4 values, the abscissa corresponding to the upper left corner of the box, and the width and height of the box, respectively.
Step S109, determining a target candidate frame according to the position information of the candidate frame, the category information of the candidate frame, the mask information of the foreign object, and the offset.
Specifically, redundant candidate frames for identifying the unified target are removed through non-maximum suppression, and finally the left candidate frames serve as targets to be identified, namely target candidate frames.
And step S110, determining foreign matter information, cable classification information and pixel area information according to the target candidate frame.
And step S111, determining a target firing line according to the foreign matter information, the classification information of the cable and the pixel area information, wherein the target firing line is a firing reference line which is parallel to the cable and passes through a foreign matter area on the target candidate frame.
Specifically, according to the foreign object information, the classification information of the cable and the pixel area information output in step S110, a firing reference line parallel to the cable and passing through the foreign object area is drawn on the image corresponding to the target candidate frame, and two intersection points of the firing reference line and the foreign object boundary are adopted as firing starting points.
And step S112, burning and removing the foreign matters according to the target burning line.
In some embodiments of the present application, the burning and cleaning of the foreign matter according to the target burning line includes:
and determining the minimum laser power density required by fusing the foreign matters according to the foreign matters information.
And burning and removing the foreign matters according to the target burning line and the minimum laser power density required by fusing the foreign matters.
Specifically, after starting the burning, repeating the steps S101 to S110 to obtain updated foreign object areas and burning points until the foreign objects are removed and the burning is completed.
In some embodiments of the present application, the minimum laser power density required when the foreign object is fused is obtained by:
in combination with the actual working scene of laser cleaning operation and the operation range of a laser, some common foreign material is selected, wherein foreign materials with different colors and transparencies are divided into different foreign material types, 100% round light spots with different power percentages are used for burning and fusing the foreign materials respectively, the percentage of the maximum power corresponding to the laser when the foreign materials are fused is recorded through a laser burning experiment, and the power density of the optimal fusing and cleaning of the foreign material is calculated.
The adjustment of the laser power is generally difficult to accurately adjust, so that experiments can be performed by adopting a method of changing the size of the light spot, thereby obtaining different laser power densities. A focusing lens with a focal length of F is adopted and is placed in front of an output port of a laser tube, laser beams are scattered after focusing at a focus, different spot sizes can be obtained at different positions after the scattering, and the laser power density corresponding to the spot size at the position can be calculated according to a formula.
The experimental foreign matter material is placed on the optical axis, and moves from far to near in sequence, as shown in fig. 2, the experimental foreign matter material starts to move from pn to p1, when the foreign matter is fused or burnt at a certain position, the spot size of the position is recorded, and the minimum laser power density required when the foreign matter is fused can be obtained through calculation.
The air temperature in the experimental environment is 15.7 ℃ and the humidity is 61%, and the experimental results are shown in table 1:
TABLE 1
Aiming at a self-constructed power grid suspended foreign matter data set, in order to realize the identification of suspended net foreign matters in a power high-voltage line laser foreign matter removal scene and the detection of laser burning points, the embodiment provides a power grid foreign matter detection and identification method based on Mask R-CNN instance segmentation, a lightweight deep learning neural network is adopted, different scale features are fully extracted from an image, an area proposal frame is generated, and a final result, foreign matter information, cable classification information and pixel area information are extracted through non-maximum suppression. After the foreign matter type is identified, the recommended light-emitting power is given according to the optimal fusing density.
According to the method for detecting and identifying the foreign matters in the power grid disclosed by the embodiment of the application, firstly, a one-to-one corresponding region suggestion frame is determined according to any foreign matter image in the power grid hanging foreign matter data set. Positive and negative samples in the grid hanging foreign object data set are then determined. And further extracting a region suggestion frame of the positive sample, and generating a mask result of the region suggestion frame of the positive sample. After the selection of the positive and negative samples is completed, the positive and negative samples are input into a loss function for calculation, and the position information of the candidate frame, the category information of the candidate frame and the mask information of the foreign matter are determined. And then acquiring a preset area proposal frame, determining the offset from the area proposal frame to the actual position of the foreign object image, further screening candidate frames, and determining target candidate frames. And finally, outputting foreign matter information, cable classification information and pixel area information according to the target candidate frame, and completing recognition and removal of the foreign matter according to the information.
The present embodiment uses feature images of different scales for detection of foreign matter by extracting feature images from the foreign matter image, and performs a segmentation operation after the detection result. Since the dimensions and feelings of each feature map are different, the processing and analysis of each feature map can detect foreign objects of different sizes and extract more accurate foreign object region information.
The following are device embodiments of the present application, which may be used to perform method embodiments of the present application. For details not disclosed in the device embodiments of the present application, please refer to the method embodiments of the present application.
The second embodiment of the application discloses a power grid foreign matter detection and identification device, the power grid foreign matter detection and identification device is applied to the power grid foreign matter detection and identification method disclosed in the first embodiment of the application, see the structure schematic diagram shown in fig. 3, the power grid foreign matter detection and identification device includes:
the foreign matter data set acquisition module 201 is configured to acquire a grid hanging foreign matter data set, where the grid hanging foreign matter data set includes a plurality of foreign matter images.
The feature map obtaining module 202 is configured to obtain, for any foreign object image, a plurality of feature maps of the any foreign object image.
The region suggestion frame determining module 203 is configured to determine a region suggestion frame corresponding to the any foreign object image according to the plurality of feature maps.
Further, the region suggestion box determining module 203 includes:
and the undetermined area suggestion frame acquisition unit is used for determining a plurality of undetermined area suggestion frames corresponding to the plurality of feature images according to the plurality of feature images.
And the region suggestion frame acquisition unit is used for determining the region suggestion frame corresponding to any foreign object image according to the plurality of pending region suggestion frames.
The sample dividing module 204 is configured to obtain a default frame set in advance, obtain a real sample value set in advance, and determine a positive sample and a negative sample according to the grid hanging foreign matter dataset, the default frame, and the real sample value. The default frame refers to frames of different sizes and aspect ratios generated at fixed positions of the foreign object image.
Further, the sample dividing module 204 includes:
the threshold value acquisition unit is used for acquiring a preset cross ratio threshold value.
And the sample dividing unit is used for judging whether the cross ratio of the default frame to the real sample value exceeds the cross ratio threshold value according to the power grid suspended foreign matter data set, and if so, determining the cross ratio as a positive sample. If not, the negative sample is determined.
A positive sample parameter module 205, configured to obtain a region suggestion box of the positive sample, and obtain a feature map of the positive sample.
A mask result determining module 206, configured to determine a mask result of the region suggestion frame of the positive sample according to the region suggestion frame of the positive sample and the feature map of the positive sample.
An information obtaining module 207, configured to determine, according to the positive sample, the negative sample, and the mask result of the region suggestion frame of the positive sample, position information of the candidate frame, category information of the candidate frame, and mask information of the foreign object.
The offset determining module 208 is configured to obtain a preset area proposal frame, and determine an offset from the area proposal frame to an actual position of the foreign object image according to the area proposal frame.
The target candidate frame determining module 209 is configured to determine a target candidate frame according to the location information of the candidate frame, the category information of the candidate frame, the mask information of the foreign object, and the offset.
The target information obtaining module 210 is configured to determine foreign object information, classification information of the cable, and pixel area information according to the target candidate frame.
The reference line determining module 211 is configured to determine a target firing line according to the foreign object information, the classification information of the cable, and the pixel area information, where the target firing line is a firing reference line parallel to the cable and passing through the foreign object area on the target candidate frame.
The foreign matter removal module 212 is configured to burn and remove the foreign matters according to the target burn line.
Further, the foreign matter removal module 212 includes:
and the density acquisition unit is used for determining the minimum laser power density required by fusing the foreign matters according to the foreign matters information.
The burning clearing unit is used for burning and clearing the foreign matters according to the target burning line and the minimum laser power density required by the fusing of the foreign matters.
In some embodiments of the present application, the offset includes: the abscissa of the upper left corner of the region proposal frame, the ordinate of the upper left corner of the region proposal frame, the width of the region proposal frame, and the height of the region proposal frame.
The foregoing detailed description has been provided for the purposes of illustration in connection with specific embodiments and exemplary examples, but such description is not to be construed as limiting the application. Those skilled in the art will appreciate that various equivalent substitutions, modifications and improvements may be made to the technical solution of the present application and its embodiments without departing from the spirit and scope of the present application, and these all fall within the scope of the present application. The scope of the application is defined by the appended claims.
Claims (7)
1. The utility model provides a power grid foreign matter detection and identification method which is characterized by comprising the following steps:
acquiring a power grid suspension foreign matter data set, wherein the power grid suspension foreign matter data set comprises a plurality of foreign matter images;
acquiring a plurality of feature images of any foreign object image aiming at the any foreign object image;
determining a region suggestion frame corresponding to any foreign object image according to the plurality of feature images;
specifically, a multi-layer feature map region suggestion frame based on a feature pyramid is generated, and after an image is read in, multi-layer convolution is carried out; generating a series of feature maps with different depths; generating a series of region suggestion frames with the same size and different depths on the series of feature images by utilizing an ROI alignment technology, and obtaining a final region suggestion frame by a non-maximum suppression algorithm;
acquiring a preset default frame, acquiring a preset real sample value, and determining a positive sample and a negative sample according to the power grid hanging foreign matter data set, the default frame and the real sample value; the default frame refers to frames of different sizes and aspect ratios generated at fixed positions of the foreign object image;
the determining positive and negative samples according to the grid hanging foreign matter dataset, the default frame and a preset real sample value comprises the following steps:
acquiring a preset cross ratio threshold;
judging whether the intersection ratio of the default frame and the real sample value exceeds the intersection ratio threshold value or not according to the power grid suspended foreign matter data set, and if yes, determining to be a positive sample; if not, determining as a negative sample;
in the default frames with various dimensions and aspect ratios, a small part of foreign object images are partially overlapped with real sample values, a contact ratio sample is set as a positive sample, and the rest foreign object images are regarded as negative samples, wherein the contact ratio of the default frames and the real sample values is measured by adopting an overlap ratio;
acquiring a region suggestion frame of the positive sample and acquiring a feature map of the positive sample;
determining a mask result of the region suggestion frame of the positive sample according to the region suggestion frame of the positive sample and the feature map of the positive sample;
the method comprises the steps of extracting a region suggestion frame of a positive sample, generating a feature map after ROI alignment and primary deconv processing, and performing sigmoid processing on each category to finally generate a mask result of the region suggestion frame;
determining position information of a candidate frame, category information of the candidate frame and mask information of foreign matters according to mask results of the positive sample, the negative sample and the region suggestion frame of the positive sample;
after the selection of positive and negative samples is completed, the position information, the category information and the mask information of the foreign object image are listed into the same loss function for calculation, and after the calculation of the loss function is completed, the parameter can be updated by back propagation;
acquiring a preset area proposal frame, and determining the offset from the area proposal frame to the actual position of the foreign object image according to the area proposal frame;
the offset includes: an abscissa of an upper left corner of the region proposal frame, an ordinate of an upper left corner of the region proposal frame, a width of the region proposal frame, and a height of the region proposal frame;
determining a target candidate frame according to the position information of the candidate frame, the category information of the candidate frame, the mask information of the foreign matter and the offset;
removing redundant candidate frames for identifying unified targets through non-maximum suppression, wherein the finally left candidate frames are used as targets to be identified, namely target candidate frames;
determining foreign matter information, classification information of cables and pixel area information according to the target candidate frame;
determining a target firing line according to the foreign matter information, the classification information of the cable and the pixel area information, wherein the target firing line is a firing reference line which is parallel to the cable and passes through a foreign matter area on the target candidate frame;
specifically, two intersection points of a burning reference line and a foreign body boundary are adopted as starting points of burning;
burning and removing the foreign matters according to the target burning line;
the burning and cleaning of the foreign matters according to the target burning line comprises the following steps:
determining the minimum laser power density required by fusing the foreign matters according to the foreign matters information;
burning and removing the foreign matters according to the target burning line and the minimum laser power density required by fusing the foreign matters;
the minimum laser power density required when the foreign matter is fused is obtained by the following method:
the method is characterized in that common foreign material is selected in combination with the actual working scene of laser cleaning operation and the operation range of a laser, foreign materials with different colors and transparencies are divided into different foreign material types, 100% round light spots with different power percentages are used for burning and fusing the foreign materials respectively, the percentage of the maximum power of the corresponding laser when the foreign materials are fused is recorded through a laser burning experiment, and the optimal fusing and cleaning power density of the foreign material is calculated.
2. The method for detecting and identifying a foreign object on a power grid according to claim 1, wherein determining an area suggestion box corresponding to the any foreign object image according to the feature maps comprises:
determining a plurality of pending area suggestion boxes corresponding to the feature maps according to the feature maps;
and determining the region suggestion frame corresponding to any foreign object image according to the plurality of the region suggestion frames to be determined.
3. A power grid foreign matter detection and recognition device, characterized in that the power grid foreign matter detection and recognition device is applied to the power grid foreign matter detection and recognition method according to any one of claims 1 or 2, the power grid foreign matter detection and recognition device comprising:
the foreign matter data set acquisition module is used for acquiring a power grid hanging foreign matter data set, wherein the power grid hanging foreign matter data set comprises a plurality of foreign matter images;
the characteristic map acquisition module is used for acquiring a plurality of characteristic maps of any foreign object image aiming at any foreign object image;
the region suggestion frame determining module is used for determining a region suggestion frame corresponding to any foreign object image according to the plurality of feature images;
the sample dividing module is used for acquiring a preset default frame, acquiring a preset real sample value and determining a positive sample and a negative sample according to the power grid suspended foreign matter data set, the default frame and the real sample value; the default frame refers to frames of different sizes and aspect ratios generated at fixed positions of the foreign object image;
the positive sample parameter module is used for acquiring a region suggestion frame of the positive sample and acquiring a feature map of the positive sample;
a mask result determining module, configured to determine a mask result of the region suggestion frame of the positive sample according to the region suggestion frame of the positive sample and the feature map of the positive sample;
the information acquisition module is used for determining the position information of the candidate frame, the category information of the candidate frame and the mask information of the foreign matters according to the positive sample, the negative sample and the mask result of the region suggestion frame of the positive sample;
the offset determining module is used for acquiring a preset area proposal frame and determining the offset from the area proposal frame to the actual position of the foreign object image according to the area proposal frame;
a target candidate frame determining module, configured to determine a target candidate frame according to the position information of the candidate frame, the category information of the candidate frame, the mask information of the foreign object, and the offset;
the target information acquisition module is used for determining foreign matter information, cable classification information and pixel area information according to the target candidate frame;
the reference line determining module is used for determining a target burning line according to the foreign matter information, the classification information of the cable and the pixel area information, wherein the target burning line is a burning reference line which is parallel to the cable and passes through a foreign matter area on the target candidate frame;
and the foreign matter removing module is used for burning and removing the foreign matters according to the target burning line.
4. The grid foreign object detection and identification device of claim 3, wherein the region suggestion box determination module includes:
a pending area suggestion frame acquisition unit configured to determine a plurality of pending area suggestion frames corresponding to the plurality of feature maps according to the plurality of feature maps;
and the region suggestion frame acquisition unit is used for determining the region suggestion frame corresponding to any foreign object image according to the plurality of pending region suggestion frames.
5. The grid foreign object detection and identification device of claim 3, wherein the sample division module comprises:
the threshold value acquisition unit is used for acquiring a preset cross ratio threshold value;
the sample dividing unit is used for judging whether the cross ratio of the default frame to the real sample value exceeds the cross ratio threshold value according to the power grid hanging foreign matter data set, and if so, determining the cross ratio as a positive sample; if not, the negative sample is determined.
6. The grid foreign object detection and identification device of claim 3, wherein the foreign object removal module comprises:
the density acquisition unit is used for determining the minimum laser power density required by fusing the foreign matters according to the foreign matters information;
the burning clearing unit is used for burning and clearing the foreign matters according to the target burning line and the minimum laser power density required by the fusing of the foreign matters.
7. The grid foreign matter detection and identification device of claim 3, wherein the offset includes: the abscissa of the upper left corner of the region proposal frame, the ordinate of the upper left corner of the region proposal frame, the width of the region proposal frame, and the height of the region proposal frame.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110950237.6A CN113695256B (en) | 2021-08-18 | 2021-08-18 | Power grid foreign matter detection and identification method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110950237.6A CN113695256B (en) | 2021-08-18 | 2021-08-18 | Power grid foreign matter detection and identification method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113695256A CN113695256A (en) | 2021-11-26 |
| CN113695256B true CN113695256B (en) | 2023-05-23 |
Family
ID=78653352
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110950237.6A Active CN113695256B (en) | 2021-08-18 | 2021-08-18 | Power grid foreign matter detection and identification method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113695256B (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110200598A (en) * | 2019-06-12 | 2019-09-06 | 天津大学 | A kind of large-scale plant that raises sign exception birds detection system and detection method |
| CN110503097A (en) * | 2019-08-27 | 2019-11-26 | 腾讯科技（深圳）有限公司 | Training method, device and the storage medium of image processing model |
| CN111008567A (en) * | 2019-11-07 | 2020-04-14 | 郑州大学 | Driver behavior identification method |
| CN111346842A (en) * | 2018-12-24 | 2020-06-30 | 顺丰科技有限公司 | Coal gangue sorting method, device, equipment and storage medium |
| CN111695622A (en) * | 2020-06-09 | 2020-09-22 | 全球能源互联网研究院有限公司 | Identification model training method, identification method and device for power transformation operation scene |
| WO2020216008A1 (en) * | 2019-04-25 | 2020-10-29 | 腾讯科技（深圳）有限公司 | Image processing method, apparatus and device, and storage medium |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US4692623A (en) * | 1986-02-13 | 1987-09-08 | The United States Of America As Represented By The Secretary Of The Army | Precision laser beam positioner and spatially resolved laser beam sampling meter |
| US10679351B2 (en) * | 2017-08-18 | 2020-06-09 | Samsung Electronics Co., Ltd. | System and method for semantic segmentation of images |
| US11170508B2 (en) * | 2018-01-03 | 2021-11-09 | Ramot At Tel-Aviv University Ltd. | Systems and methods for the segmentation of multi-modal image data |
| CN109191438B (en) * | 2018-08-17 | 2021-10-08 | 中科光绘（上海）科技有限公司 | Foreign matter identification method for laser foreign matter cleaner |
| CN109829908B (en) * | 2019-01-31 | 2023-04-14 | 广东电网有限责任公司 | Binocular image-based method and device for detecting safety distance of ground object below power line |
| CN111814867B (en) * | 2020-07-03 | 2024-06-18 | 浙江大华技术股份有限公司 | Training method of defect detection model, defect detection method and related device |
| CN112950634B (en) * | 2021-04-22 | 2023-06-30 | 内蒙古电力(集团)有限责任公司内蒙古电力科学研究院分公司 | Unmanned aerial vehicle inspection-based wind turbine blade damage identification method, equipment and system |
-
2021
- 2021-08-18 CN CN202110950237.6A patent/CN113695256B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111346842A (en) * | 2018-12-24 | 2020-06-30 | 顺丰科技有限公司 | Coal gangue sorting method, device, equipment and storage medium |
| WO2020216008A1 (en) * | 2019-04-25 | 2020-10-29 | 腾讯科技（深圳）有限公司 | Image processing method, apparatus and device, and storage medium |
| CN110200598A (en) * | 2019-06-12 | 2019-09-06 | 天津大学 | A kind of large-scale plant that raises sign exception birds detection system and detection method |
| CN110503097A (en) * | 2019-08-27 | 2019-11-26 | 腾讯科技（深圳）有限公司 | Training method, device and the storage medium of image processing model |
| CN111008567A (en) * | 2019-11-07 | 2020-04-14 | 郑州大学 | Driver behavior identification method |
| CN111695622A (en) * | 2020-06-09 | 2020-09-22 | 全球能源互联网研究院有限公司 | Identification model training method, identification method and device for power transformation operation scene |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113695256A (en) | 2021-11-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111784685B (en) | Power transmission line defect image identification method based on cloud edge cooperative detection | |
| CN110148196B (en) | Image processing method and device and related equipment | |
| CN111047554B (en) | Composite insulator overheating defect detection method based on instance segmentation | |
| CN109840889A (en) | High-precision vision measurement method, device and system based on bionic Algorithm | |
| CN103235830A (en) | Unmanned aerial vehicle (UAV)-based electric power line patrol method and device and UAV | |
| CN110610483B (en) | Crack image acquisition and detection method, computer equipment and readable storage medium | |
| CN114677554A (en) | Statistical filtering infrared small target detection tracking method based on YOLOv5 and Deepsort | |
| CN107689157B (en) | Traffic intersection passable road planning method based on deep learning | |
| CN111476314B (en) | Fuzzy video detection method integrating optical flow algorithm and deep learning | |
| CN112528979A (en) | Transformer substation inspection robot obstacle distinguishing method and system | |
| Zhang et al. | Application research of YOLO v2 combined with color identification | |
| Ma et al. | Image-based air pollution estimation using hybrid convolutional neural network | |
| CN114283137A (en) | Photovoltaic module hot spot defect detection method based on multi-scale characteristic diagram inference network | |
| CN111145222A (en) | Fire detection method combining smoke movement trend and textural features | |
| Schiopu et al. | Pothole detection and tracking in car video sequence | |
| CN115984543A (en) | Target detection algorithm based on infrared and visible light images | |
| Goga et al. | Fusing semantic labeled camera images and 3D LiDAR data for the detection of urban curbs | |
| CN114972177A (en) | Road disease identification management method and device and intelligent terminal | |
| Tutzauer et al. | Semantic urban mesh enhancement utilizing a hybrid model | |
| CN115965934A (en) | Parking space detection method and device | |
| CN113695256B (en) | Power grid foreign matter detection and identification method and device | |
| CN106228560B (en) | A kind of demographic method under complex scene | |
| CN112529917A (en) | Three-dimensional target segmentation method, device, equipment and storage medium | |
| CN116189136A (en) | Deep learning-based traffic signal lamp detection method in rainy and snowy weather | |
| CN106886609B (en) | Block type rural residential area remote sensing quick labeling 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |