CN111353413A - Low-missing-report-rate defect identification method for power transmission equipment - Google Patents
Low-missing-report-rate defect identification method for power transmission equipment Download PDFInfo
- Publication number
- CN111353413A CN111353413A CN202010117318.3A CN202010117318A CN111353413A CN 111353413 A CN111353413 A CN 111353413A CN 202010117318 A CN202010117318 A CN 202010117318A CN 111353413 A CN111353413 A CN 111353413A
- Authority
- CN
- China
- Prior art keywords
- power transmission
- transmission line
- network
- defect
- picture
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- 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
- 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
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Molecular Biology (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Probability & Statistics with Applications (AREA)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a digital image identification technology, in particular to a method for identifying defects of low missing report rate of power transmission equipment, which comprises the following steps: collecting a power transmission line inspection image through an unmanned aerial vehicle or a helicopter; marking the power transmission line inspection image, and establishing a data set; establishing a power transmission line defect identification model based on a fast RCNN network and establishing a power transmission line defect identification model based on a YOLO v3 network respectively, and training on a power transmission line inspection image data set; optimizing the trained fast RCNN model and the YOLO v3 model, and then combining the models through a self-adaptive discriminator to carry out joint training; identifying the inspection image of the power transmission line by using the jointly trained low-missing report rate defect detection model, acquiring the state information of the power equipment, and judging whether the power equipment has defects or not; and carrying out batch end-to-end processing on the inspection images. The method can enlarge the adaptability of the model to the complex environment and reduce the omission factor.
Description
Technical Field
The invention belongs to the technical field of digital image recognition, and particularly relates to a low-missing report rate defect recognition method for power transmission equipment.
Background
With the development of the long-distance power transmission technology, the power grids in all regions are gradually connected into a whole, and the stable operation of the power grids becomes more and more important. On one hand, with the continuous extension of the operation line, the contradiction between the increase of the line patrol workload and the shortage of line patrol personnel is increasingly prominent; on the other hand, most of the current basic level line patrol work is organized by referring to operation experience, and the requirements of refinement and intelligent operation and maintenance are difficult to meet. Therefore, the refined line patrol management of the power transmission line is realized, the operation and maintenance efficiency of the power transmission line is improved, and the power transmission line is widely concerned by power operation units.
In 2015, the novel inspection mode of the power transmission line combined with a helicopter and an unmanned aerial vehicle is greatly popularized; still other companies plan to basically realize the cooperative routing inspection target of 'machine routing inspection is main + person routing inspection is auxiliary' in 2020. Some areas have been provided with the "people patrol and machine patrol" operation and maintenance guidance suggestions "of overhead transmission lines, and the technical characteristics and advantages of the" people patrol "and the" machine patrol "are brought into full play, so that the purposes of reasonably controlling the operation and maintenance cost, and improving the operation and maintenance efficiency and the patrol quality are achieved. The intelligent processing of the routing inspection data is required to be enhanced, the routing inspection data is deeply mined and subjected to multi-dimensional analysis, the potential rules among the data are explored, and technical support is provided for improving the health level and the operation and maintenance management level of line equipment.
The daily-developed patrol operation of the power transmission line machine generates a large amount of patrol images and videos, the workload of manually interpreting the image data is huge, and the missing interpretation often occurs, so that the patrol operation is difficult to be applied to the actual operation and maintenance work, the interpretation of a small amount of data suspected of having obvious defects can only be completed, and a large amount of data can only be placed in a hard disk and cannot be applied. The main problems judged by manual experience are: on one hand, the operation condition of the power transmission channel is complex, and the operation and maintenance personnel have limited information which can be acquired by judging the operation condition of the equipment and the channel condition on site, so that risk assessment deviation is easily caused; on the other hand, guidance cannot be provided for the standardized process of the inspection operation of the power transmission line machine, and situations such as insufficient inspection, missing of important inspection items and the like can be caused.
By deeply mining and utilizing a large amount of machine patrol image data, the main defects of the power transmission line such as appearance, operating environment and element abnormity of the power transmission line can be effectively found, and reference is provided for equipment management and operation maintenance. However, such data is fast in production speed and less in effective information, and identification of obvious defects of the power transmission line can be realized through manual inspection and identification, but much labor and time are needed, the analysis efficiency is low, and the given result has the problems of subjectivity, fuzziness, incompleteness, easiness in missing detection and false detection, and the best discovery and processing time may be lost.
At present, the research aiming at automatic identification of the inspection image is mainly based on a general algorithm model, the requirement of low missing report rate of the inspection of a power system is not considered, the problem of high defect missing detection rate exists when the actual inspection image is processed, and the popularization of the automatic identification of the defects in the inspection work of the power transmission line is hindered.
Disclosure of Invention
The invention aims to provide a method for effectively identifying defects of power transmission equipment by analyzing power transmission line inspection images shot by an unmanned aerial vehicle or a helicopter, so as to provide reference for power detection personnel and guarantee the reliability of power transmission.
In order to achieve the purpose, the invention adopts the technical scheme that: a low false negative rate defect identification method for power transmission equipment comprises the following steps:
step 1, collecting a power transmission line inspection image through an unmanned aerial vehicle or a helicopter;
step 2, marking the power transmission line inspection image and establishing a data set;
step 3, establishing a power transmission line defect identification model based on a fast RCNN network, and training on a power transmission line inspection image data set;
step 4, establishing a power transmission line defect identification model based on a YOLO v3 network, and training on a power transmission line inspection image data set;
step 5, optimizing the trained fast RCNN model and the YOLO v3 model, combining the optimized models through a self-adaptive discriminator, and performing combined training;
step 6, identifying the inspection image of the power transmission line by using the low missing report rate defect detection model jointly trained in the step 5, acquiring the state information of the power equipment, and judging whether the power equipment has defects or not;
and 7, constructing an end-to-end transmission line inspection image intelligent defect identification system, and carrying out batch end-to-end processing on inspection images.
In the method for identifying the defects of the power transmission equipment with the low missing report rate, the step 2 comprises the following steps:
step 2.1, marking the position and width of the top left vertex of the defect in the collected power transmission line inspection picture, classifying the defect type of the collected power transmission line inspection picture, and writing marking information into a text file according to the picture name, the defect type, the abscissa of the top left vertex of the defect, the ordinate of the top left vertex of the defect, the width of the defect and the height of the defect in sequence, wherein the text file name is consistent with the picture name;
2.2, randomly selecting 80% of pictures to form a training sample library for model learning and excavating the characteristics of typical defects of the power transmission line; the remaining 20% of pictures form a verification set and are used for evaluating the accuracy of the model in the model training process; when the obtained sample data is less, a verification set is not set; the validation set does not participate in the training of the model.
In the method for identifying the defect of the low false negative rate of the power transmission equipment, the step 3 is realized by the following steps:
step 3.1, the Faster R-CNN comprises a region generation network RPN and a region detection network, and the two networks form a two-stage end-to-end detection network through a shared convolution layer;
step 3.1.1, adjusting the size of the picture to a proper size, and then performing feature extraction on the input sample picture through the depth convolution layer to obtain a deep feature pyramid picture with rich semantic information;
step 3.1.2, inputting the characteristic pyramid map into a region generation network RPN for region extraction, and obtaining a candidate region containing a target by using an anchors mechanism and an NMS algorithm;
step 3.1.3, inputting the candidate area into an area detection network, utilizing a RoI pooling layer to down-sample RoI with different sizes to a fixed size, and judging the target category and determining a position boundary frame through a full connection layer;
3.2, forming K prediction frames by taking each pixel point as a center on a feature map extracted from the convolutional layer and using different scales and length-width ratios, inputting the K prediction frames into a network to carry out target and background scoring and bounding box definition, carrying out redundancy processing on the prediction frames by using a non-maximum suppression algorithm based on the scoring, further carrying out category scoring and bounding box fine tuning regression on the processed candidate region input region detection network, then comparing with an actual target region to calculate a loss function of the network, and carrying out gradient updating on parameters in the network by using a back propagation algorithm until the network converges;
and 3.3, introducing a feature pyramid network FPN into the fast RCNN model, wherein the feature pyramid network FPN adds the feature graph of each level on the original single network and the feature graph of the next level which is zoomed twice during feature extraction.
In the method for identifying the defect of the low false negative rate of the power transmission equipment, the step 4 comprises the following steps:
step 4.1, for any one electric transmission line inspection picture collected by the unmanned aerial vehicle or the helicopter, firstly adjusting the size to 608 × 608, and dividing the picture into 19 × 19 areas; extracting picture features through a DarkNet-53 convolutional neural network to obtain a deep feature map of the power transmission line inspection picture; then, the deep feature map is transmitted to a target prediction network, and feature information of different layers is fused through upsampling and cross-layer connection, so that prediction results of 19 × 19, 38 × 38 and 76 × 76 scales are obtained; if the center of the defect of the power equipment falls in a certain area, the area is responsible for predicting the defect; in a target prediction network, automatically selecting prediction results of different scales according to the sizes of defects, identifying the defects by adjusting the central positions and sizes of candidate anchor frames through regression, and respectively corresponding target confidence coefficients, target positions and target types of each region in different scales by three matrixes of 19 × 9, 38 × 9 and 76 × 9 as output results of model prediction;
and 4.2, processing the prediction result of each scale by the model by adopting a maximum suppression method, eliminating repeated frame selection of the same predicted target, and finally obtaining the corresponding type and position of the power equipment defect actually contained in the inspection picture.
In the method for identifying the defect of the low false dismissal rate of the power transmission equipment, the step 7 comprises the following steps:
7.1, transmitting pictures of the power transmission equipment shot by the helicopter or the unmanned aerial vehicle to a server, reading all polling picture names under a folder by a system file list, and forming a to-be-detected polling picture list;
7.2, the defect identification system calls a low missing report rate defect identification model, the inspection picture list is used as input, and batch processing is carried out on the inspection pictures in the folder;
and 7.3, the system stores the inspection picture defect identification result into a log file, marks the defects contained in the inspection picture and stores the inspection picture containing the defects after detection into a result folder.
The invention has the beneficial effects that: the method comprises the steps of respectively extracting features through two improved target detection networks, and finally fusing the extraction results of the two networks by adopting an improved non-maximum suppression discriminator to finally realize the low false negative rate detection of the defects of the power transmission line. The method can fully utilize the image characteristics, enhances the adaptability of the model to the complex environment of the power transmission line, and can effectively avoid the missing detection of the defect target.
Drawings
FIG. 1 is a general flow diagram of the fast RCNN model according to an embodiment of the present invention;
FIG. 2 is a diagram of an FPN network architecture according to one embodiment of the present invention;
FIG. 3 is a general flow diagram of the YOLO v3 model according to an embodiment of the invention;
FIG. 4 is a network structure for detecting defects of a transmission line with a low missing report rate according to an embodiment of the present invention;
fig. 5 is a comparison diagram of the defect identification result of the power transmission equipment according to one embodiment of the invention.
Detailed Description
Embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
The embodiment improves the existing general target detection model, reduces the missing rate of defect detection, establishes an end-to-end defect identification system, improves the effectiveness and the practicability of automatic defect identification, and has very important significance for improving the automation degree of power transmission line inspection.
The embodiment is realized by the following technical scheme, and the method for identifying the low false negative rate defect of the power transmission equipment based on the combined target detection framework comprises the following steps:
s1: collecting a power transmission line inspection image through an unmanned aerial vehicle or a helicopter;
s2: marking the power transmission line inspection image, and establishing a data set;
s3: establishing a power transmission line defect identification model based on a fast RCNN network, and training on a power transmission line inspection image data set;
s4: establishing a power transmission line defect identification model based on a YOLO v3 network, and training on a power transmission line inspection image data set;
s5: optimizing the trained fast RCNN model and the YOLO v3 model, combining the optimized models through a self-adaptive discriminator, and performing combined training;
s6: identifying the inspection image of the power transmission line by using the trained low-missing report rate defect detection model, acquiring the state information of the power equipment, and judging whether the equipment has defects or not;
s7: and constructing an end-to-end transmission line inspection image intelligent defect identification system, and carrying out batch end-to-end processing on inspection images.
In step 7, the specific design method of the system is as follows:
s7.1: the power patrol personnel operate the helicopter or the unmanned aerial vehicle to shoot the images of the power transmission equipment, and the collected images are uniformly transmitted to the server. And reading all image names under the folder by the system file list to form a to-be-detected inspection image list.
S7.2: and the defect identification system calls a low missing report rate defect identification model, takes the inspection image list as input, and performs batch processing on the inspection images in the folder.
S7.3: the system stores the inspection picture defect identification result into a log file, marks the defects contained in the inspection image and stores the inspection picture containing the defects after detection into a result folder.
In specific implementation, the method and the system for identifying the defects of the low missing report rate of the power transmission equipment comprise the following steps:
the method comprises the steps that an unmanned aerial vehicle or a helicopter is adopted to carry an image acquisition device, main power transmission equipment is shot in the process of patrolling a power transmission line, a visible light image of the power equipment is obtained, wherein each defect type comprises enough sample quantity as original data.
Analyzing common defects of the power transmission line, marking the acquired power transmission line inspection pictures, and constructing a power transmission line defect sample library;
1) after the power transmission line inspection image is acquired, manually marking the position and the width and the height of the top left vertex of the defect in the inspection image, classifying the defect type of the inspection image, and writing marking information into a text file according to the format of the image name, the defect type, the abscissa of the top left vertex of the defect, the ordinate of the top left vertex of the defect, the width of the defect and the height of the defect in sequence, wherein the text file name is consistent with the image name.
2) And randomly selecting 80% of pictures to form a training sample library for model learning and mining the characteristics of typical defects of the power transmission line. The remaining 20% of the pictures form a validation set for evaluating the accuracy of the model in the model training process. When the obtained sample data is less, a verification set is not required to be set, and the loss function value is not reduced in the independent model training. The validation set does not participate in the training of the model.
Establishing a power transmission line defect identification model based on a fast RCNN network, and training on a power transmission line inspection image data set;
1) the Faster R-CNN is composed of a region generation network RPN and a region detection network, the two networks form a two-stage end-to-end detection network through a shared convolution layer, and the overall flow of the network is shown in FIG. 1. Firstly, for an input picture, adjusting the size of the picture to a proper size to prevent the collapse caused by network overload due to overlarge picture size, then performing feature extraction on the input sample picture through a depth convolution layer to obtain a deep feature pyramid map with rich semantic information, then inputting the feature map into an RPN network to perform region extraction, obtaining a candidate region containing a target by using an anchors mechanism and an NMS algorithm, then inputting the candidate region generated by the RPN network into a detection network, down-sampling RoIs with different sizes to a fixed size by using a RoI pooling layer, and then determining the category of the target and determining a position boundary box through a full connection layer.
2) The candidate region generation of the Faster RCNN network is mainly implemented by the RPN network. The RPN network forms K (the K in the embodiment is 2000) prediction frames by taking each pixel point as a center on a feature map extracted from the convolutional layer and taking different scales and length-width ratios, then inputs the prediction frames into the network to carry out target and background scoring and bounding box definition, uses a non-maximum suppression algorithm to carry out redundancy processing on the prediction frames based on the scoring, inputs a processed candidate region into the detection network to carry out further category scoring and bounding box fine tuning regression, then compares the candidate region with an actual target region to calculate a loss function of the network, and carries out gradient updating on parameters in the network through a back propagation algorithm until the network converges.
3) For the characteristic of large difference in target size of the power transmission line, in this embodiment, FPN (Feature Pyramid Network) is introduced into the fast RCNN model, and the structure of the Network is shown in fig. 2. The FPN network adds the feature map of each stage of the original single network and the feature map of the next stage which is zoomed twice during feature extraction. Through the cross-layer connection, the influence of feature maps with different resolutions is considered in the prediction of each layer, the fusion of multilayer semantics is realized, and the detection precision of small object targets can be effectively improved.
Fourthly, establishing a power transmission line defect identification model based on a YOLO v3 network, and training on a power transmission line inspection image data set;
1) the core idea of YOLO is to use a full convolution structure and perform regression by using context information of a full graph, and the regression result, i.e., the category and position information of the target box, is shown in fig. 3. The YOLO v3 network abandons the link of area search, directly sets a preselected anchor frame, and finally obtains a target detection result by carrying out regression adjustment on the preselected anchor frame.
2) For any machine patrol picture, firstly adjusting the size to 608 × 608, dividing the picture into 19 × 19 areas, and extracting picture features through a DarkNet-53 convolutional neural network to obtain a deep feature map of the transmission line patrol picture. And then, transmitting the feature map to a target prediction network, and fusing feature information of different layers through upsampling and cross-layer connection to obtain prediction results of 19 × 19, 38 × 38 and 76 × 76. For each zone, if the center of a power equipment defect falls within the zone, the zone is responsible for predicting the defect. In the target prediction network, the prediction results of different scales are automatically selected according to the sizes of the defects, the central positions and the sizes of the candidate anchor frames are adjusted through regression, the defects can be accurately identified, and the output results of model prediction are three matrixes of 19 × 9, 38 × 9 and 76 × 9, which respectively correspond to the target confidence coefficient, the target positions and the target types of each region in different scales.
3) And processing the prediction result of each scale by the model by adopting a maximum suppression method, eliminating repeated frame selection for predicting the same target, and finally obtaining the corresponding type and position of the power equipment defect actually contained in the inspection picture.
Fifthly, optimizing the trained fast RCNN model and the YOLO v3 model, combining the optimized models through a self-adaptive discriminator, and performing combined training;
1) in the embodiment, the defect detection results of the fast RCNN model and the YOLOv3 model are fused, and a power transmission line defect low-missing report rate identification model based on a combined target detection framework is built. FIG. 4 is a combined convolutional neural network framework, which comprises a total of three parts: (1) a defect identification network based on fast RCNN; (2) a defect identification network based on YOLOv 3; (3) an adaptive discriminator based on an improved non-maxima suppression method.
2) The adaptive arbiter is designed according to an improved non-maxima suppression algorithm. The non-maximum suppression algorithm (NMS) is a common duplication removing method in target detection, and can determine the accurate position of a target from a plurality of candidate positions, so that repeated detection is avoided. The current non-maximum suppression algorithm uses the confidence of the target as a criterion for judging whether the candidate frame is accurate, i.e. the candidate frame with high confidence is positioned with higher accuracy. In the algorithm, however, fast RCNN uses the target classification score as a confidence, while YOLO v3 integrates the classification score and the localization score using a full convolution network. Since the convergence of the target classification score is generally higher than the target location score, the prediction result of YOLO v3 is difficult to gain advantage in the non-maxima suppression process. Therefore, the embodiment introduces an improved non-maximum suppression algorithm, which introduces a positioning score into the confidence by calculating the distribution of each boundary of the target frame, and ensures the balance of the two algorithms in the model.
3) For the trained model, the present embodiment fixes the parameters of the feature extraction network, modifies the structure and the loss function of the target prediction layer, and the position information of each defective target is represented by the mean and the variance of the frame position:
B=[b1x,b1y,b2x,b2y]
V=[σ1x,σ1y,σ2x,σ2y]
for each target location distribution, the fitting can be performed by a gaussian function:
wherein x iseThe mean value of the predicted target position is represented, and σ represents the variance corresponding to each edge.
The variance of the boundary position of the marked picture can be considered as 0, and the marked picture is approximately regarded as a dirichlet function distribution:
PD(x)=δ(x-xg)
wherein x isgIndicating the actual location of the defect.
Correcting the positioning loss function:
4) after feature extraction and regression are carried out on the fast RCNN and the YOLO v3, N predicted targets are obtained together, the positions of the targets are adjusted according to the intersection ratio, the variance and the confidence coefficient among the targets by the improved non-maximum suppression method, and the accurate positions of the defects are obtained. The flow of implementation of the improved non-maxima suppression method is shown in table 1:
TABLE 1 improved non-maxima suppression Algorithm
Wherein, p is the influence weight of each adjacent frame:
the updated boundary positions are:
the discriminator combines the adjusted models, performs joint training on the patrol image data set, autonomously adapts to the threshold value required by the data set, and performs adaptive fusion on the feature extraction results of the two networks, thereby avoiding the deviation caused by manually set hyper-parameters.
5) The model first trained the fast RCNN model and the YOLO v3 model separately. The models are initialized by using parameters pre-trained on the ImageNet data set respectively, and because no power transmission line inspection picture exists in the public data set, the two models are trained on the power transmission line inspection image data set continuously until the loss function value of the models does not decrease, so that a preliminary power transmission line defect identification model is obtained.
6) Subsequently, the present embodiment improves the target prediction layers in the two models, improves the original fixed boundary position into a random coordinate distributed according to a normal distribution, and directly predicts the mean and variance of the boundary position. And the two networks carry out defect identification to obtain the position and the category information of the defect, then the position and the category information of the defect are input into the self-adaptive discriminator, and a certain weight is given to each target according to the variance of the frames in the detection result, so that the detection results of the two networks are fused, and the position of the defect target is adaptively adjusted. And finally, the discriminator inhibits a plurality of prediction results of the same target according to the fusion result to obtain the type and position information of the equipment defect in the picture.
Sixthly, identifying the inspection image of the power transmission line by using the trained low missing report rate defect detection model, acquiring the state information of the power equipment, and judging whether the equipment has defects or not;
constructing an end-to-end transmission line inspection image intelligent defect identification system, and carrying out batch end-to-end processing on inspection images;
the embodiment provides an end-to-end power transmission line defect identification method, which comprises the steps of firstly establishing a power transmission line inspection image defect identification system on a server platform, uploading the power transmission line inspection image to the server platform after the power transmission line inspection image is shot to obtain an inspection image, reading the name of the inspection image to be identified through a file system, reading the inspection image according to an image list by a defect identification algorithm, and processing the inspection image in batches to output an analysis result. For the pictures with defects, the algorithm outputs information such as picture names, types of each defect in the pictures, positions of the defects in the pictures and the like, and the information can be used as a maintenance reference opinion or data for further subsequent processing.
In the first step of the embodiment, four types of transmission line typical defects including tower defects, insulator defects, ground wire defects and large-size hardware defects can be performed according to unmanned aerial vehicle aerial data, and 5000 total images of routing inspection aerial images are acquired from a certain power-saving network company;
in the embodiment, the two pairs of routing inspection aerial images are marked, and the positions and types of the defects are identified and used as labels for training and learning of the deep learning model.
And step three in the embodiment, a power transmission line defect identification model based on the fast RCNN is built and trained, and the power transmission line defect identification model is used as a partial feature extraction network of the integral model.
In the fourth step of the embodiment, a transmission line defect identification model based on the YOLO v3 network is built and trained, and the model is used as a partial feature extraction network of the whole model.
In the fifth step of the embodiment, the fast RCNN and YOLO v3 networks are improved, the two networks are combined through the construction of the adaptive discriminator to obtain the combined target detection network, and the combined training is continued on the basis of the parameters obtained through the training in the third step and the fourth step to complete the construction and the training of the low false rejection rate defect identification model of the power transmission line.
Step six in this embodiment is the practical application of the present invention to the inspection image of the power transmission line, and the test result of the trained model on the inspection image is shown in fig. 5.
To analyze the effectiveness of this example, this example was tested in batches on a 1000-test sample set, and the results are shown in table 2.
TABLE 2 comparison of test results
As can be seen from table 2, the combined defect detection model provided in this embodiment achieves a good effect on the inspection image data set, and the average missed detection rate of the combined defect detection model is 0.207, which is reduced by 33.9% compared with the fast RCNN network and 46.1% compared with the YOLO v3 network. According to the test result, the missing detection rate in the power transmission line inspection image identification can be effectively reduced, and the requirement of power transmission line inspection work is met.
It should be understood that parts of the specification not set forth in detail are well within the prior art.
Although specific embodiments of the present invention have been described above with reference to the accompanying drawings, it will be appreciated by those skilled in the art that these are merely illustrative and that various changes or modifications may be made to these embodiments without departing from the principles and spirit of the invention. The scope of the invention is only limited by the appended claims.
Claims (5)
1. A low false negative rate defect identification method for power transmission equipment is characterized by comprising the following steps:
step 1, collecting a power transmission line inspection image through an unmanned aerial vehicle or a helicopter;
step 2, marking the power transmission line inspection image and establishing a data set;
step 3, establishing a power transmission line defect identification model based on a fast RCNN network, and training on a power transmission line inspection image data set;
step 4, establishing a power transmission line defect identification model based on a YOLO v3 network, and training on a power transmission line inspection image data set;
step 5, optimizing the trained fast RCNN model and the YOLO v3 model, combining the optimized models through a self-adaptive discriminator, and performing combined training;
step 6, identifying the inspection image of the power transmission line by using the low missing report rate defect detection model jointly trained in the step 5, acquiring the state information of the power equipment, and judging whether the power equipment has defects or not;
and 7, constructing an end-to-end transmission line inspection image intelligent defect identification system, and carrying out batch end-to-end processing on inspection images.
2. The method for identifying low false negative rate defects of power transmission equipment according to claim 1, wherein the step 2 is implemented by the following steps:
step 2.1, marking the position and width of the top left vertex of the defect in the collected power transmission line inspection picture, classifying the defect type of the collected power transmission line inspection picture, and writing marking information into a text file according to the picture name, the defect type, the abscissa of the top left vertex of the defect, the ordinate of the top left vertex of the defect, the width of the defect and the height of the defect in sequence, wherein the text file name is consistent with the picture name;
2.2, randomly selecting 80% of pictures to form a training sample library for model learning and excavating the characteristics of typical defects of the power transmission line; the remaining 20% of pictures form a verification set and are used for evaluating the accuracy of the model in the model training process; when the obtained sample data is less, a verification set is not set; the validation set does not participate in the training of the model.
3. The method for identifying low false negative rate defects of power transmission equipment according to claim 1, wherein the step 3 is implemented by the following steps:
step 3.1, the Faster R-CNN comprises a region generation network RPN and a region detection network, and the two networks form a two-stage end-to-end detection network through a shared convolution layer;
step 3.1.1, adjusting the size of the picture to a proper size, and then performing feature extraction on the input sample picture through the depth convolution layer to obtain a deep feature pyramid picture with rich semantic information;
step 3.1.2, inputting the characteristic pyramid map into a region generation network RPN for region extraction, and obtaining a candidate region containing a target by using an anchors mechanism and an NMS algorithm;
step 3.1.3, inputting the candidate area into an area detection network, utilizing a RoI pooling layer to down-sample RoI with different sizes to a fixed size, and judging the target category and determining a position boundary frame through a full connection layer;
3.2, forming K prediction frames by taking each pixel point as a center on a feature map extracted from the convolutional layer and using different scales and length-width ratios, inputting the K prediction frames into a network to carry out target and background scoring and bounding box definition, carrying out redundancy processing on the prediction frames by using a non-maximum suppression algorithm based on the scoring, further carrying out category scoring and bounding box fine tuning regression on the processed candidate region input region detection network, then comparing with an actual target region to calculate a loss function of the network, and carrying out gradient updating on parameters in the network by using a back propagation algorithm until the network converges;
and 3.3, introducing a feature pyramid network FPN into the fast RCNN model, wherein the feature pyramid network FPN adds the feature graph of each level on the original single network and the feature graph of the next level which is zoomed twice during feature extraction.
4. The method for identifying low false negative rate defects of power transmission equipment according to claim 1, wherein the step 4 is implemented by the following steps:
step 4.1, for any one electric transmission line inspection picture collected by the unmanned aerial vehicle or the helicopter, firstly adjusting the size to 608 × 608, and dividing the picture into 19 × 19 areas; extracting picture features through a DarkNet-53 convolutional neural network to obtain a deep feature map of the power transmission line inspection picture; then, the deep feature map is transmitted to a target prediction network, and feature information of different layers is fused through upsampling and cross-layer connection, so that prediction results of 19 × 19, 38 × 38 and 76 × 76 scales are obtained; if the center of the defect of the power equipment falls in a certain area, the area is responsible for predicting the defect; in a target prediction network, automatically selecting prediction results of different scales according to the sizes of defects, identifying the defects by adjusting the central positions and sizes of candidate anchor frames through regression, and respectively corresponding target confidence coefficients, target positions and target types of each region in different scales by three matrixes of 19 × 9, 38 × 9 and 76 × 9 as output results of model prediction;
and 4.2, processing the prediction result of each scale by the model by adopting a maximum suppression method, eliminating repeated frame selection of the same predicted target, and finally obtaining the corresponding type and position of the power equipment defect actually contained in the inspection picture.
5. The method for identifying low false negative rate defects of power transmission equipment according to claim 1, wherein the step 7 is implemented by the following steps:
7.1, transmitting pictures of the power transmission equipment shot by the helicopter or the unmanned aerial vehicle to a server, reading all polling picture names under a folder by a system file list, and forming a to-be-detected polling picture list;
7.2, the defect identification system calls a low missing report rate defect identification model, the inspection picture list is used as input, and batch processing is carried out on the inspection pictures in the folder;
and 7.3, the system stores the inspection picture defect identification result into a log file, marks the defects contained in the inspection picture and stores the inspection picture containing the defects after detection into a result folder.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010117318.3A CN111353413B (en) | 2020-02-25 | 2020-02-25 | Low-missing-report-rate defect identification method for power transmission equipment |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010117318.3A CN111353413B (en) | 2020-02-25 | 2020-02-25 | Low-missing-report-rate defect identification method for power transmission equipment |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111353413A true CN111353413A (en) | 2020-06-30 |
CN111353413B CN111353413B (en) | 2022-04-15 |
Family
ID=71195802
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010117318.3A Active CN111353413B (en) | 2020-02-25 | 2020-02-25 | Low-missing-report-rate defect identification method for power transmission equipment |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111353413B (en) |
Cited By (35)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111784692A (en) * | 2020-08-11 | 2020-10-16 | 国网内蒙古东部电力有限公司 | Method and device for detecting insulator defects in power system and electronic equipment |
CN111860439A (en) * | 2020-07-31 | 2020-10-30 | 广东电网有限责任公司 | Unmanned aerial vehicle inspection image defect detection method, system and equipment |
CN112001902A (en) * | 2020-08-19 | 2020-11-27 | 上海商汤智能科技有限公司 | Defect detection method and related device, equipment and storage medium |
CN112036463A (en) * | 2020-08-26 | 2020-12-04 | 国家电网有限公司 | Power equipment defect detection and identification method based on deep learning |
CN112069894A (en) * | 2020-08-03 | 2020-12-11 | 许继集团有限公司 | Wire strand scattering identification method based on fast-RCNN model |
CN112132826A (en) * | 2020-10-12 | 2020-12-25 | 国网河南省电力公司濮阳供电公司 | Pole tower accessory defect inspection image troubleshooting method and system based on artificial intelligence |
CN112229845A (en) * | 2020-10-12 | 2021-01-15 | 国网河南省电力公司濮阳供电公司 | Unmanned aerial vehicle high-precision winding tower intelligent inspection method based on visual navigation technology |
CN112308850A (en) * | 2020-11-09 | 2021-02-02 | 国网山东省电力公司威海供电公司 | Multi-scale feature fusion power transmission line detection method and system |
CN112330915A (en) * | 2020-10-29 | 2021-02-05 | 五邑大学 | Unmanned aerial vehicle forest fire prevention early warning method and system, electronic equipment and storage medium |
CN112365491A (en) * | 2020-11-27 | 2021-02-12 | 上海市计算技术研究所 | Method for detecting welding seam of container, electronic equipment and storage medium |
CN112487916A (en) * | 2020-11-25 | 2021-03-12 | 上海电力大学 | Binocular three-dimensional identification system for electrical equipment |
CN112561872A (en) * | 2020-12-09 | 2021-03-26 | 福州大学 | Corrosion defect segmentation method for tower crane |
CN113012107A (en) * | 2021-02-22 | 2021-06-22 | 江苏方天电力技术有限公司 | Power grid defect detection method and system |
CN113052820A (en) * | 2021-03-25 | 2021-06-29 | 贵州电网有限责任公司 | Circuit equipment defect identification method based on neural network technology |
CN113111875A (en) * | 2021-04-02 | 2021-07-13 | 广州地铁集团有限公司 | Seamless steel rail weld defect identification device and method based on deep learning |
CN113160184A (en) * | 2021-04-26 | 2021-07-23 | 贵州电网有限责任公司 | Unmanned aerial vehicle intelligent inspection cable surface defect detection method based on deep learning |
CN113191389A (en) * | 2021-03-31 | 2021-07-30 | 中国石油大学(华东) | Submarine pipeline autonomous inspection method and device based on optical vision technology |
CN113361520A (en) * | 2021-06-01 | 2021-09-07 | 南京南瑞信息通信科技有限公司 | Transmission line equipment defect detection method based on sample offset network |
CN113378918A (en) * | 2021-06-09 | 2021-09-10 | 武汉大学 | Insulator binding wire state detection method based on metric learning |
CN113409249A (en) * | 2021-05-17 | 2021-09-17 | 上海电力大学 | Insulator defect detection method based on end-to-end algorithm |
CN113436184A (en) * | 2021-07-15 | 2021-09-24 | 南瑞集团有限公司 | Power equipment image defect judging method and system based on improved twin network |
CN113642486A (en) * | 2021-08-18 | 2021-11-12 | 国网江苏省电力有限公司泰州供电分公司 | Unmanned aerial vehicle distribution network inspection method with airborne front-end identification model |
CN113657280A (en) * | 2021-08-18 | 2021-11-16 | 广东电网有限责任公司 | Power transmission line target defect detection warning method and system |
CN113971666A (en) * | 2021-10-29 | 2022-01-25 | 贵州电网有限责任公司 | Power transmission line machine inspection image self-adaptive identification method based on depth target detection |
CN114359281A (en) * | 2022-03-17 | 2022-04-15 | 南方电网数字电网研究院有限公司 | Electric power component identification method and device based on hierarchical ensemble learning |
CN114549512A (en) * | 2022-03-01 | 2022-05-27 | 成都数之联科技股份有限公司 | Circuit board defect detection method, device, equipment and medium |
CN114627360A (en) * | 2020-12-14 | 2022-06-14 | 国电南瑞科技股份有限公司 | Substation equipment defect identification method based on cascade detection model |
CN114724091A (en) * | 2022-06-07 | 2022-07-08 | 智洋创新科技股份有限公司 | Method and device for identifying foreign matters on transmission line wire |
CN114972721A (en) * | 2022-06-13 | 2022-08-30 | 中国科学院沈阳自动化研究所 | Power transmission line insulator string recognition and positioning method based on deep learning |
WO2022241784A1 (en) * | 2021-05-21 | 2022-11-24 | 京东方科技集团股份有限公司 | Defect detection method and apparatus, storage medium, and electronic device |
CN116071773A (en) * | 2023-03-15 | 2023-05-05 | 广东电网有限责任公司东莞供电局 | Method, device, medium and equipment for detecting form in power grid construction type archive |
CN116309564A (en) * | 2023-05-17 | 2023-06-23 | 厦门微图软件科技有限公司 | Method and system for detecting appearance defects of battery cells based on artificial intelligent image recognition |
CN116797949A (en) * | 2023-06-21 | 2023-09-22 | 广东电网有限责任公司汕尾供电局 | Convolutional neural network-based power tower acceptance state sensing method |
CN117253365A (en) * | 2023-11-17 | 2023-12-19 | 上海伯镭智能科技有限公司 | Automatic detection method and related device for vehicle traffic condition |
CN113436184B (en) * | 2021-07-15 | 2024-05-24 | 南瑞集团有限公司 | Power equipment image defect discriminating method and system based on improved twin network |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20170323163A1 (en) * | 2016-05-06 | 2017-11-09 | City Of Long Beach | Sewer pipe inspection and diagnostic system and method |
CN110310261A (en) * | 2019-06-19 | 2019-10-08 | 河南辉煌科技股份有限公司 | A kind of Contact Net's Suspension Chord defects detection model training method and defect inspection method |
CN110335270A (en) * | 2019-07-09 | 2019-10-15 | 华北电力大学(保定) | Transmission line of electricity defect inspection method based on the study of hierarchical regions Fusion Features |
CN110618129A (en) * | 2019-07-24 | 2019-12-27 | 安徽南瑞继远电网技术有限公司 | Automatic power grid wire clamp detection and defect identification method and device |
CN110689531A (en) * | 2019-09-23 | 2020-01-14 | 云南电网有限责任公司电力科学研究院 | Automatic power transmission line machine inspection image defect identification method based on yolo |
US20200026257A1 (en) * | 2018-07-23 | 2020-01-23 | Accenture Global Solutions Limited | Augmented reality (ar) based fault detection and maintenance |
-
2020
- 2020-02-25 CN CN202010117318.3A patent/CN111353413B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20170323163A1 (en) * | 2016-05-06 | 2017-11-09 | City Of Long Beach | Sewer pipe inspection and diagnostic system and method |
US20200026257A1 (en) * | 2018-07-23 | 2020-01-23 | Accenture Global Solutions Limited | Augmented reality (ar) based fault detection and maintenance |
CN110310261A (en) * | 2019-06-19 | 2019-10-08 | 河南辉煌科技股份有限公司 | A kind of Contact Net's Suspension Chord defects detection model training method and defect inspection method |
CN110335270A (en) * | 2019-07-09 | 2019-10-15 | 华北电力大学(保定) | Transmission line of electricity defect inspection method based on the study of hierarchical regions Fusion Features |
CN110618129A (en) * | 2019-07-24 | 2019-12-27 | 安徽南瑞继远电网技术有限公司 | Automatic power grid wire clamp detection and defect identification method and device |
CN110689531A (en) * | 2019-09-23 | 2020-01-14 | 云南电网有限责任公司电力科学研究院 | Automatic power transmission line machine inspection image defect identification method based on yolo |
Non-Patent Citations (2)
Title |
---|
JING ZHANG 等: "An automatic diagnostic method of abnormal heat defect in transmission lines based on infrared video", 《IEEE》 * |
赵振兵 等: "基于深度学习的输电线路视觉检测研究综述", 《广东电力》 * |
Cited By (44)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111860439A (en) * | 2020-07-31 | 2020-10-30 | 广东电网有限责任公司 | Unmanned aerial vehicle inspection image defect detection method, system and equipment |
CN112069894A (en) * | 2020-08-03 | 2020-12-11 | 许继集团有限公司 | Wire strand scattering identification method based on fast-RCNN model |
CN111784692A (en) * | 2020-08-11 | 2020-10-16 | 国网内蒙古东部电力有限公司 | Method and device for detecting insulator defects in power system and electronic equipment |
CN112001902A (en) * | 2020-08-19 | 2020-11-27 | 上海商汤智能科技有限公司 | Defect detection method and related device, equipment and storage medium |
CN112036463A (en) * | 2020-08-26 | 2020-12-04 | 国家电网有限公司 | Power equipment defect detection and identification method based on deep learning |
CN112132826A (en) * | 2020-10-12 | 2020-12-25 | 国网河南省电力公司濮阳供电公司 | Pole tower accessory defect inspection image troubleshooting method and system based on artificial intelligence |
CN112229845A (en) * | 2020-10-12 | 2021-01-15 | 国网河南省电力公司濮阳供电公司 | Unmanned aerial vehicle high-precision winding tower intelligent inspection method based on visual navigation technology |
CN112330915A (en) * | 2020-10-29 | 2021-02-05 | 五邑大学 | Unmanned aerial vehicle forest fire prevention early warning method and system, electronic equipment and storage medium |
CN112330915B (en) * | 2020-10-29 | 2023-02-28 | 五邑大学 | Unmanned aerial vehicle forest fire prevention early warning method and system, electronic equipment and storage medium |
CN112308850A (en) * | 2020-11-09 | 2021-02-02 | 国网山东省电力公司威海供电公司 | Multi-scale feature fusion power transmission line detection method and system |
CN112487916A (en) * | 2020-11-25 | 2021-03-12 | 上海电力大学 | Binocular three-dimensional identification system for electrical equipment |
CN112487916B (en) * | 2020-11-25 | 2022-08-23 | 上海电力大学 | Binocular three-dimensional identification system for electrical equipment |
CN112365491A (en) * | 2020-11-27 | 2021-02-12 | 上海市计算技术研究所 | Method for detecting welding seam of container, electronic equipment and storage medium |
CN112561872A (en) * | 2020-12-09 | 2021-03-26 | 福州大学 | Corrosion defect segmentation method for tower crane |
CN112561872B (en) * | 2020-12-09 | 2022-05-24 | 福州大学 | Corrosion defect segmentation method for tower crane |
CN114627360A (en) * | 2020-12-14 | 2022-06-14 | 国电南瑞科技股份有限公司 | Substation equipment defect identification method based on cascade detection model |
CN113012107A (en) * | 2021-02-22 | 2021-06-22 | 江苏方天电力技术有限公司 | Power grid defect detection method and system |
CN113012107B (en) * | 2021-02-22 | 2022-07-08 | 江苏方天电力技术有限公司 | Power grid defect detection method and system |
CN113052820A (en) * | 2021-03-25 | 2021-06-29 | 贵州电网有限责任公司 | Circuit equipment defect identification method based on neural network technology |
CN113191389A (en) * | 2021-03-31 | 2021-07-30 | 中国石油大学(华东) | Submarine pipeline autonomous inspection method and device based on optical vision technology |
CN113111875A (en) * | 2021-04-02 | 2021-07-13 | 广州地铁集团有限公司 | Seamless steel rail weld defect identification device and method based on deep learning |
CN113160184B (en) * | 2021-04-26 | 2022-06-07 | 贵州电网有限责任公司 | Unmanned aerial vehicle intelligent inspection cable surface defect detection method based on deep learning |
CN113160184A (en) * | 2021-04-26 | 2021-07-23 | 贵州电网有限责任公司 | Unmanned aerial vehicle intelligent inspection cable surface defect detection method based on deep learning |
CN113409249A (en) * | 2021-05-17 | 2021-09-17 | 上海电力大学 | Insulator defect detection method based on end-to-end algorithm |
WO2022241784A1 (en) * | 2021-05-21 | 2022-11-24 | 京东方科技集团股份有限公司 | Defect detection method and apparatus, storage medium, and electronic device |
CN113361520A (en) * | 2021-06-01 | 2021-09-07 | 南京南瑞信息通信科技有限公司 | Transmission line equipment defect detection method based on sample offset network |
CN113378918A (en) * | 2021-06-09 | 2021-09-10 | 武汉大学 | Insulator binding wire state detection method based on metric learning |
CN113378918B (en) * | 2021-06-09 | 2022-06-07 | 武汉大学 | Insulator binding wire state detection method based on metric learning |
CN113436184B (en) * | 2021-07-15 | 2024-05-24 | 南瑞集团有限公司 | Power equipment image defect discriminating method and system based on improved twin network |
CN113436184A (en) * | 2021-07-15 | 2021-09-24 | 南瑞集团有限公司 | Power equipment image defect judging method and system based on improved twin network |
CN113657280A (en) * | 2021-08-18 | 2021-11-16 | 广东电网有限责任公司 | Power transmission line target defect detection warning method and system |
CN113642486A (en) * | 2021-08-18 | 2021-11-12 | 国网江苏省电力有限公司泰州供电分公司 | Unmanned aerial vehicle distribution network inspection method with airborne front-end identification model |
CN113971666A (en) * | 2021-10-29 | 2022-01-25 | 贵州电网有限责任公司 | Power transmission line machine inspection image self-adaptive identification method based on depth target detection |
CN114549512A (en) * | 2022-03-01 | 2022-05-27 | 成都数之联科技股份有限公司 | Circuit board defect detection method, device, equipment and medium |
CN114359281A (en) * | 2022-03-17 | 2022-04-15 | 南方电网数字电网研究院有限公司 | Electric power component identification method and device based on hierarchical ensemble learning |
CN114724091A (en) * | 2022-06-07 | 2022-07-08 | 智洋创新科技股份有限公司 | Method and device for identifying foreign matters on transmission line wire |
CN114972721A (en) * | 2022-06-13 | 2022-08-30 | 中国科学院沈阳自动化研究所 | Power transmission line insulator string recognition and positioning method based on deep learning |
CN116071773A (en) * | 2023-03-15 | 2023-05-05 | 广东电网有限责任公司东莞供电局 | Method, device, medium and equipment for detecting form in power grid construction type archive |
CN116309564A (en) * | 2023-05-17 | 2023-06-23 | 厦门微图软件科技有限公司 | Method and system for detecting appearance defects of battery cells based on artificial intelligent image recognition |
CN116309564B (en) * | 2023-05-17 | 2023-08-11 | 厦门微图软件科技有限公司 | Method and system for detecting appearance defects of battery cells based on artificial intelligent image recognition |
CN116797949A (en) * | 2023-06-21 | 2023-09-22 | 广东电网有限责任公司汕尾供电局 | Convolutional neural network-based power tower acceptance state sensing method |
CN116797949B (en) * | 2023-06-21 | 2024-03-05 | 广东电网有限责任公司汕尾供电局 | Convolutional neural network-based power tower acceptance state sensing method |
CN117253365A (en) * | 2023-11-17 | 2023-12-19 | 上海伯镭智能科技有限公司 | Automatic detection method and related device for vehicle traffic condition |
CN117253365B (en) * | 2023-11-17 | 2024-02-02 | 上海伯镭智能科技有限公司 | Automatic detection method and related device for vehicle traffic condition |
Also Published As
Publication number | Publication date |
---|---|
CN111353413B (en) | 2022-04-15 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111353413B (en) | Low-missing-report-rate defect identification method for power transmission equipment | |
CN109118479B (en) | Capsule network-based insulator defect identification and positioning device and method | |
CN107609525B (en) | Remote sensing image target detection method for constructing convolutional neural network based on pruning strategy | |
CN110084165B (en) | Intelligent identification and early warning method for abnormal events in open scene of power field based on edge calculation | |
CN110163213B (en) | Remote sensing image segmentation method based on disparity map and multi-scale depth network model | |
CN104992223A (en) | Dense population estimation method based on deep learning | |
CN113205063A (en) | Visual identification and positioning method for defects of power transmission conductor | |
CN111222478A (en) | Construction site safety protection detection method and system | |
CN111160407A (en) | Deep learning target detection method and system | |
CN111160432A (en) | Automatic classification method and system for panel production defects | |
CN111126278A (en) | Target detection model optimization and acceleration method for few-category scene | |
CN110599458A (en) | Underground pipe network detection and evaluation cloud system based on convolutional neural network | |
CN111667461A (en) | Method for detecting abnormal target of power transmission line | |
CN113971666A (en) | Power transmission line machine inspection image self-adaptive identification method based on depth target detection | |
CN114037895A (en) | Unmanned aerial vehicle pole tower inspection image identification method | |
CN113536944A (en) | Distribution line inspection data identification and analysis method based on image identification | |
CN115984158A (en) | Defect analysis method and device, electronic equipment and computer readable storage medium | |
CN115830302B (en) | Multi-scale feature extraction fusion power distribution network equipment positioning identification method | |
CN113095160A (en) | Power system personnel safety behavior identification method and system based on artificial intelligence and 5G | |
CN115937079A (en) | YOLO v 3-based rapid detection method for defects of power transmission line | |
CN112508946B (en) | Cable tunnel anomaly detection method based on antagonistic neural network | |
CN112380985A (en) | Real-time detection method for intrusion foreign matters in transformer substation | |
CN113780462A (en) | Vehicle detection network establishment method based on unmanned aerial vehicle aerial image and application thereof | |
CN113689439A (en) | Unmanned aerial vehicle image capturing method based on reinforcement learning image processing technology | |
CN111860332A (en) | Dual-channel electrokinetic diagram part detection method based on multi-threshold cascade detector |
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 |