CN114913376A

CN114913376A - Image-based defect automatic identification method, device and system and storage medium

Info

Publication number: CN114913376A
Application number: CN202210539735.6A
Authority: CN
Inventors: 葛馨远; 王照; 陈金梅
Original assignee: Guangzhou Power Supply Bureau of Guangdong Power Grid Co Ltd
Current assignee: Guangzhou Power Supply Bureau of Guangdong Power Grid Co Ltd
Priority date: 2022-05-18
Filing date: 2022-05-18
Publication date: 2022-08-16

Abstract

The invention discloses a method, a device, a system and a storage medium for automatically identifying defects based on images, wherein the method comprises the following steps: acquiring image data to be identified, and classifying the image data to be identified according to a preset type to obtain classified image data; the image data to be identified is a plurality of original image data which are subjected to cruise shooting by the unmanned aerial vehicle in a preset area; carrying out intelligent defect identification on the classified image data through the trained algorithm model to generate a defect frame, and filling corresponding defect information in the defect frame; and auditing the defect information according to the prior knowledge and/or the auditing operation instruction, and outputting a defect identification report of the identified image data according to an auditing result. According to the invention, the trained algorithm model is used for intelligently identifying the defects of the image data cruising by the unmanned aerial vehicle, so that the defect identification rate and the working efficiency are improved, and the critical defects are found and processed, so that the power failure loss can be avoided, and a solid guarantee is provided for ensuring the reliable power supply of a power distribution system.

Description

Image-based defect automatic identification method, device and system and storage medium

Technical Field

The invention relates to the technical field of patrol data processing of power distribution network machines, in particular to a method, a device and a system for automatically identifying defects based on images and a storage medium.

Background

In the low latitude subtropical monsoon climate area, the extreme climate such as thunder and lightning, typhoon, rainstorm and the like appear in summer. The overhead line of the power distribution network is influenced by typhoon, rainstorm, thunder and lightning and the like, so that the defects of strand loosening and strand breaking of a lead, lightning stroke ablation of an insulator, damage of the insulator and the like are easily caused, and the typical defects are hidden dangers which are not negligible when the overhead line of the power distribution network is safely and reliably operated. Therefore, the fact that typical defects of the distribution network are discovered and eliminated in time is very important for guaranteeing safe and stable operation of the distribution network. The traditional power distribution network defect finding mainly depends on visual identification of operation and maintenance personnel in the periodic inspection process or identification by means of a telescope, and the method is low in efficiency and high in labor cost, and the defect position is not easy to find due to the height problem.

In recent years, the application of unmanned aerial vehicle inspection operation in a domestic power system is rapidly developed, and a coordinated inspection mode of 'mainly machine inspection and assisting human inspection' is also explicitly proposed in 'thirteen-five' planning by southern power grid companies, so that the inspection quality and efficiency of a power line are continuously improved. Taking the Guangdong power grid as an example, nearly 2PB machine patrol data has been accumulated by the Guangdong power grid by 10 months of 2020, wherein the original image is as high as 4000 ten thousand. At present, the defect discovery of the power distribution system mostly depends on the follow-up manual picture inspection of a filing machine and the manual identification of the defect picture. However, as the number of machine patrol pictures increases, the time and effort for identifying defects in the machine patrol pictures by manpower is increased, which brings new working difficulties for operators. In addition, the condition of missing detection and false detection also exists in the process of identifying the defects in the inspection pictures of the machine by operators with different working experiences, and the accuracy of defect identification is influenced.

Therefore, the prior art has yet to be improved.

Disclosure of Invention

The invention aims to solve the technical problem that the defects in the prior art are overcome, and provides an image-based defect automatic identification method, device, system and storage medium, so that the defect identification efficiency and accuracy of a power distribution network are improved.

The technical scheme adopted by the invention for solving the technical problem is as follows:

in a first aspect, the present invention provides an automatic defect identification method based on images, including:

acquiring image data to be identified, and classifying the image data to be identified according to a preset type to obtain classified image data; the image data to be identified is a plurality of original image data which are obtained by cruising and shooting of the unmanned aerial vehicle in a preset area;

performing intelligent defect identification on the classified image data through a trained algorithm model to generate a defect frame, and filling corresponding defect information in the defect frame;

and auditing the defect information according to the prior knowledge and/or the auditing operation instruction, and outputting a defect identification report of the identified image data according to an auditing result.

In one implementation, the acquiring image data to be recognized previously includes:

obtaining tagged image data from a source data layer; the labeled image data is sample image data of a labeled defect area and defect information;

and constructing an algorithm model, and performing model training according to the labeled image data.

In one implementation, the model training from the labeled image data includes:

preprocessing the labeled image data, and classifying the preprocessed image data to obtain a training set and a test set; wherein the pre-processing comprises: data conversion, data normalization, data cleaning and data encoding;

and training the constructed algorithm model through the training set, and testing the algorithm model through the testing set to obtain the trained algorithm model.

In one implementation manner, the acquiring image data to be recognized and classifying the image data to be recognized according to a preset type to obtain classified image data includes:

acquiring the image data to be identified;

and classifying the image data to be identified according to the power network feeder line and the tower coordinates of the line ledger to obtain the classified image data.

In one implementation, the classifying the image data to be recognized according to tower coordinates of a power grid feeder and a line ledger further includes:

selecting a corresponding line folder according to the classified image data;

inputting corresponding defect identification project information according to the feeder line name and the line name in the line folder; the defect identification project information comprises power supply operation and maintenance information, machine patrol information and audit information.

In an implementation manner, the intelligently identifying the defect of the classified image data through the trained algorithm model to generate a defect frame, and filling the defect frame with corresponding defect information includes:

selecting the trained algorithm model according to the classified image data;

inputting the classified image data to the trained algorithm model input layer;

determining the intelligently identified defect part, defect type and defect appearance according to the image data output by the trained algorithm model;

selecting corresponding defect library data; the defect library data comprises tower data, overhead conductor data and insulator data;

and generating the defect frame according to the defect part, the defect type, the defect representation and the defect library data, and filling corresponding defect information in the defect frame.

In one implementation, the reviewing the defect information according to the priori knowledge and/or the review operation instruction, and outputting the defect identification report of the identified image data according to the review result includes:

auditing the defect information according to the prior knowledge or the auditing operation instruction, and labeling atypical defects in the defect information;

and calling the auxiliary drawing to edit, delete, modify and add the identified defect marking frame, and generating a defect identification report of the identified image data according to a report form format.

In a second aspect, the present invention further provides an apparatus for automatically identifying defects based on images, comprising: a processor and a memory, the memory storing an image-based defect automatic identification program, the image-based defect automatic identification program when executed by the processor being for implementing the image-based defect automatic identification method according to the first aspect.

In a third aspect, the present invention further provides an image-based automatic defect identification system, including: a server and the automatic defect identification device according to the second aspect;

the server is used for providing database data and computing power service required in automatic defect identification for the automatic defect identification device;

the automatic defect identification device is used for executing the following operations:

In a fourth aspect, the present invention further provides a storage medium, which is a computer-readable storage medium, and the storage medium stores an image-based automatic defect identification program, and the image-based automatic defect identification program is used to implement the image-based automatic defect identification method according to the first aspect when executed by a processor.

The invention adopts the technical scheme and has the following effects:

the invention provides a method, a device, a system and a storage medium for automatically identifying defects based on images, wherein the method comprises the following steps: acquiring image data to be identified, and classifying the image data to be identified according to a preset type to obtain classified image data; the image data to be identified is a plurality of pieces of original image data which are obtained by cruising and shooting of the unmanned aerial vehicle in a preset area; performing intelligent defect identification on the classified image data through the trained algorithm model to generate a defect frame, and filling corresponding defect information in the defect frame; and auditing the defect information according to the prior knowledge and/or the auditing operation instruction, and outputting a defect identification report of the identified image data according to an auditing result. According to the method, the trained algorithm model is used for intelligently identifying the defects of the image data cruising by the unmanned aerial vehicle, so that the defect identification rate and the working efficiency are improved, the power failure loss caused by the discovery and the processing of the critical defects can be avoided, and a solid guarantee is provided for ensuring the reliable power supply of a power distribution system.

Drawings

In order to more clearly illustrate the embodiments or technical solutions of the present invention, the drawings used in the embodiments or technical solutions of the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the structures shown in the drawings without creative efforts.

FIG. 1 is a flow chart of a method for automatic image-based defect identification in one implementation of the invention.

FIG. 2 is an architecture diagram of an automatic image-based defect recognition system in one implementation of the invention.

FIG. 3 is a schematic diagram of an image-based automatic defect identification system deployment in one implementation of the present invention.

FIG. 4 is a diagram of a built-in defect database in one implementation of the invention.

Fig. 5 is a functional schematic diagram of an automatic image-based defect recognition apparatus in one implementation of the present invention.

The implementation, functional features and advantages of the objects of the present invention will be further explained with reference to the accompanying drawings.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention clearer and clearer, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not limit the invention.

Exemplary method

The defect discovery of the power distribution system mostly depends on the follow-up manual picture archiving of a machine and manual identification of the defect picture. However, as the number of machine patrol pictures increases, the time and effort for identifying defects in the machine patrol pictures by manpower is increased, which brings new working difficulties for operators. In addition, the condition of missing detection and false detection also exists in the process of identifying the defects in the inspection pictures of the machine by operators with different working experiences, and the accuracy of defect identification is influenced.

In order to solve the technical problem, the embodiment provides an automatic defect identification method based on an image, and the trained algorithm model is used for intelligently identifying the defects of image data cruising by an unmanned aerial vehicle, so that the defect identification rate and the working efficiency are improved.

The embodiment of the invention is divided into a model training part and a defect automatic identification part.

In the model training part, the method comprises the following steps:

s001, acquiring labeled image data from a source data layer; wherein, the labeled image data is sample image data of a labeled defect area and defect information;

and S002, constructing an algorithm model, and performing model training according to the labeled image data.

In this embodiment, for the model training part, a cloud platform can be identified based on typical defects of network deployment machine patrol pictures, and the infrastructure of the platform takes Docker and Kubernetes technologies as the core, and a machine learning framework and an algorithm library are constructed by relying on the computing resources of the underlying infrastructure, the high-speed internet and a diversified file system, so as to provide upper-layer business application services.

In the present embodiment, the algorithm production architecture based on the above platform includes the following processes:

firstly, a machine learning platform acquires labeled image data from a source data layer;

secondly, after the platform acquires the image data, preprocessing the image data, including data conversion, data normalization, data cleaning and data encoding;

again, the pre-processed image data is divided into a training set/test set. Training the algorithm model by using the training set data, and testing the algorithm model by using the test set data;

finally, generating a trained algorithm model through multiple training iterations; when the algorithm is deployed, the algorithm model is managed and stored in the storage management library through the asset storage model.

Specifically, in one implementation manner of the present embodiment, the step S002 includes the following steps:

step S002a, preprocessing the labeled image data, and classifying the preprocessed image data to obtain a training set and a test set; wherein the pre-processing comprises: data conversion, data normalization, data cleaning and data encoding;

and S002b, training the constructed algorithm model through the training set, and testing the algorithm model through the testing set to obtain the trained algorithm model.

In the actual training process, image data annotation is a key link of the whole artificial intelligence model production process and is related to the identification accuracy of the finally output model. Because massive marking data are needed to participate in model training, the original data of the cleaned sample needs to be marked manually. In order to realize the unified management of the labeling personnel and the labeling data and further integrally improve the quality and the efficiency of data labeling, a set of online or offline data labeling tools is needed.

The annotation result is in standard Pascal Voc format (the same format as ImageNet), namely JPG + XML; an online marking function of the Web end is constructed based on HTML + JavaScript + CSS + MYSQL, and marking modes such as classification marking, frame marking, point tracing marking and the like are provided. The module can support most current mainstream browsers, including IE10 and above, GoogleChromes 66.0 and above, Firefox 59.0.2 and above, and Windows 7.0.0 and above. The main labeling mode is that image data and requirements to be labeled are submitted to a platform, the platform distributes tasks in a crowdsourcing mode after auditing, and data labeling work can be carried out after the tasks are received.

Tag management provides a hierarchical tag management function divided by users. The user can classify the label according to the service requirement of the actual service scene, manage the relevant information of the label, and can be divided into: tag name, tag code, tag instance graph, tag attributes, tag description, etc.

In this embodiment, the model training process is implemented by an algorithm submodule, which can be trained according to the detection task and the recognition task; the algorithm submodule comprises an AI basic software platform, the AI basic software platform constructs an algorithm development module based on cloud computing power and applying technologies such as Linux + Tomcat + Python + Mysql + java + API and the like, can support an open source algorithm and a self-research algorithm, and realizes complete life cycle research and development management of algorithm models such as computing power scheduling, algorithm modeling, algorithm training, algorithm testing and packaging, algorithm deployment and the like. The AI basic software platform integrates a classical algorithm of machine learning, a mainstream deep learning neural network algorithm and the like to construct a basic algorithm ecosphere. The platform provides a function for a user to develop a model based on technical frameworks such as Tensorflow and the like.

The algorithm modeling provides a guide mode creating process, and the operation steps are as follows:

first, basic information of the model is written, and categories of problems are handled, for example: name of the model, image recognition or voice recognition;

then, selecting a data set for model training, preprocessing the data set, and segmenting the training data set;

secondly, writing in a training name, selecting an applicable algorithm according to actual requirements, and configuring relevant parameters, such as: selecting a convolutional neural network, and setting a neural network hierarchy;

thirdly, determining the calculation power and the use condition of the calculation resources;

and finally, completing the work of configuring and creating the model and starting training the model.

For the created model, the platform presents brief information of the model in a model list mode, and the brief information comprises the following steps: the method comprises the following steps of a technical framework, model names, algorithm parameters, latest model training information, training progress and the like. By clicking and checking the detailed contents of the model, the training records of the model, and the detailed information of the algorithm unit and the data set of each training can be checked.

After training is completed, different evaluation data sets can be selected to carry out multiple evaluations (including internal evaluation and external evaluation) on the trained model, and evaluation results are checked; after the evaluation is completed, the trained and evaluated algorithm model can be released and stored.

As shown in fig. 1, in the automatic defect identification section, the following steps are included:

step S100, acquiring image data to be identified, and classifying the image data to be identified according to a preset type to obtain classified image data; the image data to be recognized are a plurality of original image data shot by the unmanned aerial vehicle in a preset area in a cruising mode.

In this embodiment, the image-based automatic defect identification method is implemented through a pre-trained algorithm model, the algorithm model is deployed in an image-based automatic defect identification system, and the image-based automatic defect identification system is a typical defect identification system for a network distribution machine patrol image.

As shown in fig. 2, in this embodiment, the typical defect identification system for the patrol image of the distribution network machine mainly includes: a hardware layer, a network layer, an AI platform layer and a business layer.

The hardware layer comprises a GPU inference server and a workstation, and the system is based on the hardware layer and bears various function modules related to the AI platform and system services. The GPU reasoning server mainly bears all modules of the AI platform layer, is infrastructure of each module of the AI platform and is connected with the workstation through a network. The GPU reasoning server is a quick, stable and elastic computing service applied to various scenes such as image coding and decoding, deep learning, scientific computing and the like based on the GPU, excellent graphic processing capacity and high-performance computing capacity provide extreme computing performance, computing pressure is effectively relieved, and computing processing efficiency and competitiveness of products are improved. The workstation bears all contents of a system service layer, the using operation of the software client is carried out through the workstation, and meanwhile, each module of an AI platform in the GPU reasoning server can be called through a network to realize intelligent defect identification. The workstation has strong data processing capacity, has an intuitive user interface which is convenient for man-machine information exchange, can be connected with a computer network, and can intercommunicate information and share resources in a larger range. The workstation gives the professional workers a comprehensive help in programming, computing, document writing, archiving, communication, etc.

The network layer is used for providing basic network data connection, and can simultaneously connect a plurality of workstations to the same group of GPU reasoning servers, so that defect intelligent analysis can be simultaneously carried out by a plurality of people and a plurality of machines.

The AI platform layer is provided with an AI model library, an AI engine, an algorithm self-training module and an algorithm artificial training module. The AI platform layer is used as the core content of the system technical architecture and carries the training and recognition calling of the algorithm model. The platform trains and updates the model base through an algorithm artificial training module; the AI engine responds to the requirements of the front-end client to call a model library and an algorithm; the AI model library is important content for supporting the identification of the client model algorithm, and the customization of different models is the greatest technical characteristic.

The service layer mainly refers to a defect identification system client, and comprises various functions: intelligent mapping, intelligent defect identification, manual defect review, one-key report generation and the like.

As shown in fig. 3, in this embodiment, when the typical defect identification system in this embodiment is deployed, the GPU inference workstation and the client application workstation may be expanded horizontally as the amount of traffic data increases.

Specifically, in one implementation manner of the present embodiment, the step S100 includes the following steps:

step S101, acquiring the image data to be identified;

and S102, classifying the image data to be recognized according to the power grid feeder line and the tower coordinates of the line ledger to obtain the classified image data.

In practical application, the main business process of a typical defect identification system includes: firstly, a client acquires original machine patrol data on a workstation, and automatically classifies the data by using an intelligent image splitting function; then, data are uploaded to a server through a client (the server has a breakpoint continuous transmission function), an AI model is selected for defect intelligent identification, a defect frame is generated, and partial defect information is automatically filled; then, examining the intelligent defect identification result, performing operations such as addition, deletion, modification and the like on the defects, and performing one or two times of examination as required; finally, a report or report is generated by one key.

In practical application, the typical defect recognition system has the function of intelligent image segmentation, and the intelligent image segmentation module can classify the original photos of the unmanned aerial vehicle according to the tower coordinates of the feeder lines and the line ledgers, so that a large amount of workload can be saved.

Specifically, in an implementation manner of the present embodiment, the step S100 further includes the following steps:

step S103, selecting a corresponding line folder according to the classified image data;

step S104, inputting corresponding defect identification project information according to the feeder line name and the line name in the line folder; the defect identification project information comprises power supply operation and maintenance information, machine patrol information and audit information.

In practical application, the typical defect identification system also has a function of quickly uploading data, a 'newly added' key can be clicked to enter a project creation interface through the line information of a built-in power supply office, and corresponding defect identification project information is automatically input according to formats by selecting a standard named line folder and according to a feeder line name and a line name named by the folder under the interface; the information of the power supply bureau, the operation and maintenance team and the like is searched for the ledger database according to the name of the feeder line and the name of the line, and is automatically input according to the search result; and selecting or writing information such as tour teams, analysts, auditors and the like according to actual conditions.

As shown in fig. 1, in an implementation manner of the embodiment of the present invention, the method for automatically identifying defects based on images further includes the following steps:

and S200, intelligently identifying defects of the classified image data through the trained algorithm model to generate a defect frame, and filling corresponding defect information in the defect frame.

In this embodiment, after the image data is classified and the defect identification item information is entered, a trained algorithm model can be selected from a typical defect identification system, and a task of defect intelligent identification can be completed by matching with the data of a defect library.

Specifically, in one implementation manner of the present embodiment, the step S200 includes the following steps:

step S201, selecting the trained algorithm model according to the classified image data;

step S202, inputting the classified image data into the trained algorithm model input layer;

step S203, determining the intelligently identified defect part, defect type and defect appearance according to the image data output by the trained algorithm model;

step S204, selecting corresponding defect library data; the defect library data comprises tower data, overhead conductor data and insulator data;

step S205, generating the defect frame according to the defect portion, the defect type, the defect representation, and the defect library data, and filling corresponding defect information in the defect frame.

In the practical application process, an issued algorithm model can be selected in the typical defect recognition system, namely a trained algorithm model is selected; then, selecting the classified image data as input, inputting the input data into an input layer of the trained algorithm model, and identifying and outputting the corresponding image data through a feature extraction function of the trained algorithm model; the output image data comprises the marked defect part, the defect type and the defect representation.

The defect position, defect type and defect representation of the input image can be determined through the output image data, and the intelligent defect identification process of the input image is realized.

In this embodiment, in addition to the intelligent identification process of the algorithm model, data in the corresponding defect library needs to be selected, as shown in fig. 4, where the data in the defect library includes tower data, overhead conductor data, and insulator data.

In the embodiment, the defect library is a customized database which is pre-built in the system, and the defect description is called in a hierarchy reduction mode by adopting the customized distribution network defect library, so that the defects can be quickly and accurately identified; and generating a defect frame by combining the defect database data according to the defect part, the defect type and the defect representation identified by the algorithm model, and filling corresponding defect information in the defect frame so as to supplement the information of the defect identified by the algorithm model.

and step S300, auditing the defect information according to the prior knowledge and/or the auditing operation instruction, and outputting a defect identification report of the identified image data according to an auditing result.

In this embodiment, after the defect intelligent identification is completed, the identified defect needs to enter the check interface to perform secondary determination, and in the check interface, the real-time check can be performed in a manual check mode, or the automatic check can be performed through the input priori knowledge (i.e., the check experience manually input).

Meanwhile, in the checking process, the marking work can be carried out on the atypical defects. When the marking work is carried out on the atypical defect, CAD auxiliary drawing hormone can be called, and the operations of editing, deleting, modifying, adding and the like can be carried out on the identified defect marking box.

Specifically, in one implementation manner of the present embodiment, the step S300 includes the following steps:

step S301, examining and verifying the defect information according to the prior knowledge and/or the examination operation instruction, and labeling atypical defects in the defect information;

and step S302, calling the auxiliary drawing to edit, delete, modify and add the identified defect labeling frame, and generating a defect identification report of the identified image data according to a report form format.

During checking, the shortcut key set by the system can be quickly checked, wherein the shortcut key is modified in the setting interface; in the nuclear detection interface, the left list is a list of defects identified by intelligent identification; the lower part is a photo list of the current tower (switchable); the right side is the detail information of the current defect frame, and a distribution network line defect library is arranged in the window.

And for the defects which cannot be identified and the defects which are missed to be detected, the artificial picture frame can be used for repairing the defects.

After the checking work is finished, a defect report can be generated, a Word version inspection report and an Excel version inspection statistic can be derived according to requirements, and a plurality of items can be combined to generate a report and a report.

The present embodiment is described below by way of practical examples:

in the embodiment, the developed typical defect identification system for the patrol image of the distribution network machine is used for carrying out patrol image processing on a plurality of lines, the number of processed pictures per hour is 2335 on average, the recall rate and the accuracy rate of a distribution network bird nest, insulator lightning stroke ablation, insulator damage and lead appearance defect (broken strand and scattered strand ablation) model on a test set reach over 80%, the defect identification rate and the working efficiency are improved, the power failure loss can be avoided through the discovery and the processing of critical defects, a solid guarantee is provided for ensuring the reliable power supply of a power distribution system, and the application prospect is wide.

Description of the test: and testing the recognition rate and the accuracy rate of the hidden defects of insulator damage, insulator arc burn, tower foreign matter and broken strand of the lead in the scene of the unmanned aerial vehicle line patrol picture by using the hidden defect analysis software.

The preset conditions are as follows: presetting four types of pictures to be identified with defect hidden danger and pictures to be identified without defect hidden danger, wherein the number of each type of pictures is as follows:

picture type	Breakage of insulator	Insulator arc burn	Foreign body on pole tower	Broken strand of wire	Defect free
						Quantity (Zhang)	100	100	100	16	980

The testing steps are as follows: and respectively importing the pictures to be identified into the device for identification according to the following picture combination mode.

The insulator breakage test picture combination mode is as follows:

/	for the first time	For the second time	The third time
				With insulator breakage (open)	50	100	100
No insulator damage (tension)	980	500	980

The combination mode of the arc burn test pictures of the insulator is as follows:

the tower foreign matter test picture combination mode is as follows:

/	for the first time	For the second time	The third time
				Foreign matter with pole tower (Zhang)	50	100	100
Foreign matter (Zhang) of rodless tower	980	500	980

The combination mode of the broken strand test pictures of the conducting wires is as follows:

/	for the first time	For the second time	The third time
				With wire broken strand (sheet)	8	16	16
Strand break without conductor	980	500	980

And (3) testing results: through the test, the system is to the damaged defect hidden danger identification rate of insulator in the picture: 83.60%, 82.53% and 82.53%, the accuracy is 87.38%, 86.67% and 81.20%, the pass standard is met, and the test result is passed. The identification rate of the hidden danger of the arc burning defect of the insulator is 81.90%, 82.56% and 82.56%, the accuracy rate is 89.22%, 85.5% and 88.24%, the passing standard is met, and the test result is passed. The identification rate of the potential hazards of foreign body defects of the tower is 92.0%, 95.0% and 95.0%, the accuracy rate is 95.82%, 97.50% and 96.85%, the standard is met, and the test result is passed. The identification rate of the hidden danger of the broken strand defect of the lead is 87.5 percent, 81.25 percent and 81.25 percent, the accuracy rate is 82.99 percent, 82.17 percent and 82.22 percent, the pass standard is met, and the test result is passed.

In this embodiment, a typical defect data management system for picture inspection by a network distributor is used to perform standardized management and storage on defect data. Defect data are marked on a typical defect recognition cloud platform of a picture of the network configuration machine, and model optimization training is carried out, so that the recall rate and the accuracy rate of the model are further improved.

The intelligent defect identification is carried out on the image data of unmanned aerial vehicle cruising through the trained algorithm model, the defect identification rate and the working efficiency are improved, the power failure loss caused by emergency defect discovery and treatment can be avoided, and a solid guarantee is provided for ensuring reliable power supply of a power distribution system.

Exemplary device

Based on the above embodiments, the present invention further provides an automatic defect identification apparatus based on images, and a schematic block diagram thereof may be as shown in fig. 5.

The image-based defect automatic identification device comprises: the system comprises a processor, a memory, an interface, a display screen and a communication module which are connected through a system bus; wherein, the processor of the image-based defect automatic identification device is used for providing calculation and control capability; the memory of the automatic defect identifying device based on the image comprises a storage medium and an internal memory; the storage medium stores an operating system and a computer program; the internal memory provides an environment for the operation of an operating system and a computer program in the storage medium; the interface is used for connecting external equipment, such as mobile terminals, computers and the like; the display screen is used for displaying corresponding image-based defect automatic identification information; the communication module is used for communicating with a cloud server or a mobile terminal.

The computer program is executed by a processor to implement an image-based automatic defect identification method.

It will be understood by those skilled in the art that the schematic block diagram shown in fig. 5 is only a block diagram of a partial structure related to the present invention, and does not constitute a limitation of the image-based defect automatic identification apparatus to which the present invention is applied, and a specific image-based defect automatic identification apparatus may include more or less components than those shown in the figure, or combine some components, or have different component arrangements.

In one embodiment, an automatic defect identification device based on images is provided, which includes: the automatic defect identification device comprises a processor and a memory, wherein the memory stores an automatic defect identification program based on an image, and the automatic defect identification program based on the image is used for realizing the automatic defect identification method based on the image when being executed by the processor.

Based on the above embodiment, the present invention further provides an image-based automatic defect identification system, including: the server and the automatic defect identification device are used for identifying the defects;

the automatic defect recognition device is used for performing the following operations:

carrying out intelligent defect identification on the classified image data through a trained algorithm model to generate a defect frame, and filling corresponding defect information in the defect frame;

In one embodiment, a storage medium is provided, wherein the storage medium stores an image-based automatic defect identification program, and the image-based automatic defect identification program is used for realizing the above image-based automatic defect identification method when being executed by a processor.

It will be understood by those skilled in the art that all or part of the processes of the methods of the above embodiments may be implemented by hardware related to instructions of a computer program, which may be stored in a non-volatile storage medium, and when executed, may include the processes of the embodiments of the methods described above. Any reference to memory, storage, databases, or other media used in embodiments provided herein may include non-volatile and/or volatile memory.

In summary, the present invention provides a method, an apparatus, a system and a storage medium for automatically identifying defects based on images, wherein the method comprises: acquiring image data to be recognized, and classifying the image data to be recognized according to a preset type to obtain classified image data; the image data to be identified is a plurality of original image data which are subjected to cruise shooting by the unmanned aerial vehicle in a preset area; performing intelligent defect identification on the classified image data through the trained algorithm model to generate a defect frame, and filling corresponding defect information in the defect frame; and auditing the defect information according to the prior knowledge and/or the auditing operation instruction, and outputting a defect identification report of the identified image data according to an auditing result. According to the invention, the trained algorithm model is used for intelligently identifying the defects of the image data cruising by the unmanned aerial vehicle, so that the defect identification rate and the working efficiency are improved, and the critical defects are found and processed, so that the power failure loss can be avoided, and a solid guarantee is provided for ensuring the reliable power supply of a power distribution system.

It is to be understood that the invention is not limited to the examples described above, but that modifications and variations may be effected thereto by those of ordinary skill in the art in light of the foregoing description, and that all such modifications and variations are intended to be within the scope of the invention as defined by the appended claims.

Claims

1. An automatic defect identification method based on images is characterized by comprising the following steps:

2. The method according to claim 1, wherein the acquiring image data to be identified comprises:

obtaining tagged image data from a source data layer; wherein, the labeled image data is sample image data of a labeled defect area and defect information;

3. The method of claim 2, wherein the model training according to the labeled image data comprises:

4. The method according to claim 1, wherein the obtaining image data to be recognized and classifying the image data to be recognized according to a preset type to obtain classified image data comprises:

acquiring the image data to be identified;

and classifying the image data to be recognized according to the power grid feeder line and the tower coordinates of the line ledger to obtain the classified image data.

5. The method according to claim 4, wherein the image data to be recognized is classified according to tower coordinates of a power grid feeder and a line ledger, and then the method further comprises:

selecting a corresponding line folder according to the classified image data;

6. The method of claim 1, wherein the intelligent defect recognition is performed on the classified image data through the trained algorithm model to generate a defect frame, and the defect frame is filled with corresponding defect information, and the method includes:

selecting the trained algorithm model according to the classified image data;

inputting the classified image data to the trained algorithm model input layer;

7. The method for automatically identifying defects based on images according to claim 1, wherein the examining the defect information according to the prior knowledge and/or the examining operation instruction and outputting the defect identification report of the identified image data according to the examining result comprises:

8. An image-based automatic defect recognition apparatus, comprising: a processor and a memory, the memory storing an image-based defect automatic identification program, the image-based defect automatic identification program when executed by the processor being configured to implement the image-based defect automatic identification method according to any one of claims 1 to 7.

9. An image-based automatic defect identification system, comprising: a server and the defect automatic identification device according to claim 8;

10. A storage medium, characterized in that the storage medium is a computer-readable storage medium, and the storage medium stores an image-based automatic defect identification program, and the image-based automatic defect identification program is used for implementing the automatic defect identification method according to any one of claims 1 to 7 when executed by a processor.