CN115147377A - Training method and device for CycleGAN model for generating defect images of photovoltaic panel - Google Patents

Training method and device for CycleGAN model for generating defect images of photovoltaic panel Download PDF

Info

Publication number
CN115147377A
CN115147377A CN202210790763.5A CN202210790763A CN115147377A CN 115147377 A CN115147377 A CN 115147377A CN 202210790763 A CN202210790763 A CN 202210790763A CN 115147377 A CN115147377 A CN 115147377A
Authority
CN
China
Prior art keywords
image
defect
photovoltaic panel
candidate image
normal
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202210790763.5A
Other languages
Chinese (zh)
Inventor
吴昊
叶林
李东辉
张时
周盛龙
杨和康
李霖
常梦星
张晓萱
任鑫
王�华
王恩民
武青
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Huaneng Clean Energy Research Institute
Huaneng Dali Wind Power Co Ltd Eryuan Branch
Original Assignee
Huaneng Clean Energy Research Institute
Huaneng Dali Wind Power Co Ltd Eryuan Branch
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Huaneng Clean Energy Research Institute, Huaneng Dali Wind Power Co Ltd Eryuan Branch filed Critical Huaneng Clean Energy Research Institute
Priority to CN202210790763.5A priority Critical patent/CN115147377A/en
Publication of CN115147377A publication Critical patent/CN115147377A/en
Priority to PCT/CN2023/079045 priority patent/WO2024007602A1/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0004—Industrial image inspection
    • G06T7/001—Industrial image inspection using an image reference approach
    • 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
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Biophysics (AREA)
  • Evolutionary Computation (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Quality & Reliability (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Image Analysis (AREA)
  • Photovoltaic Devices (AREA)

Abstract

The application provides a training method and a device of a cycleGAN model for generating a defect image of a photovoltaic panel, the method comprises the following steps: the method comprises the steps of obtaining a training sample, inputting the training sample into a to-be-trained cyclic generation countermeasure network cyclic GAN model for generating a photovoltaic panel defect image, obtaining a candidate image of the photovoltaic panel output by the cyclic GAN model, and training the cyclic generation countermeasure network cyclic GAN model according to the training sample and the candidate image to obtain a target cyclic GAN model for generating the photovoltaic panel defect image. Therefore, the target CycleGAN model for generating the defect images of the photovoltaic panel can convert the normal images of the photovoltaic panel into the defect images of the photovoltaic panels of different types, the problem that the defect images of the photovoltaic panel are few is solved, and the accuracy of defect detection of the photovoltaic panel is improved.

Description

Training method and device for CycleGAN model for generating defect image of photovoltaic panel
Technical Field
The invention relates to the technical field of artificial intelligence, in particular to a training method and a training device for a cycleGAN model for generating a defect image of a photovoltaic panel.
Background
With the rapid development of the photovoltaic industry in recent years, the scale of a photovoltaic panel is huge, as most of centralized photovoltaic stations are built in remote areas, related personnel are not easy to reach, the operation and maintenance of the photovoltaic panel become one of the main factors influencing the development of the photovoltaic industry, and in the operation process of the photovoltaic panel, the defects of surface damage, dust, shielding, dirt, hot spots and the like may exist, so that the power generation efficiency of a photovoltaic module is not high, the photovoltaic panel can be damaged seriously, the safe operation of the photovoltaic station is influenced, and certain economic loss can be caused.
In the related art, the defect of the photovoltaic panel can be detected by an artificial intelligence image recognition method, however, the defect detection accuracy of the photovoltaic panel is low due to the fact that the number of the defect images of the photovoltaic panel is small. Therefore, how to solve the problem of few defect images of the photovoltaic panel and further improve the accuracy of defect detection of the photovoltaic panel becomes a problem to be solved urgently.
Disclosure of Invention
The application provides a training method of a cycleGAN model for generating a defect image of a photovoltaic panel, which can convert a normal image of the photovoltaic panel into different types of defect images of the photovoltaic panel through a target cycleGAN model for generating the defect image of the photovoltaic panel, solve the problem of few defect images of the photovoltaic panel and improve the accuracy of defect detection of the photovoltaic panel.
According to a first aspect of the application, a training method of a CycleGAN model for generating a defect image of a photovoltaic panel is provided, comprising the following steps: acquiring a training sample, wherein the training sample comprises a normal sample image and a defect sample image of a photovoltaic panel; inputting the training sample into a cyclic generation countermeasure network CycleGAN model to be trained for generating a defect image of the photovoltaic panel, and acquiring a candidate image of the photovoltaic panel output by the CycleGAN model, wherein the candidate image comprises a normal candidate image of the photovoltaic panel and a defect candidate image of the photovoltaic panel; and training the cycleGAN model according to the training sample and the candidate image to obtain a target cycleGAN model for generating a defect image of the photovoltaic panel.
In addition, the training method of the CycleGAN model for generating the defect image of the photovoltaic panel according to the above embodiment of the present application may further have the following additional technical features:
according to an embodiment of the present application, the inputting the training sample into a CycleGAN model to be trained for generating a defect image of a photovoltaic panel, and acquiring a candidate image of the photovoltaic panel output by the CycleGAN model further includes: generating a first defect candidate image and a third defect candidate image corresponding to the normal sample image, and generating a first normal candidate image corresponding to the first defect candidate image; and generating a corresponding second normal candidate image and a third normal candidate image according to the defect sample image, and generating a corresponding second defect candidate image according to the second normal candidate image.
According to one embodiment of the application, the CycleGAN model comprises a first generator and a second generator, the method further comprising:
inputting the normal sample image into the first generator, outputting the first defect candidate image from the normal sample image by the first generator, and inputting the first defect candidate image into the second generator, outputting the first normal candidate image from the first defect candidate image by the second generator, and inputting the normal sample image into the second generator, outputting the third defect candidate image from the normal sample image by the second generator.
Inputting the defect sample image into the second generator, outputting the second normal candidate image from the defect sample image by the second generator, inputting the second normal candidate image into the first generator, outputting the second defect candidate image from the second normal candidate image by the first generator, and inputting the defect sample image into the first generator, outputting the third normal candidate image from the defect sample image by the first generator.
According to one embodiment of the application, the CycleGAN model comprises a first arbiter and a second arbiter, the method further comprising: according to the first discriminator, discriminating the normal sample image of the photovoltaic panel and the normal candidate image of the photovoltaic panel; and judging the defect sample image of the photovoltaic panel and the defect candidate image of the photovoltaic panel according to the second discriminator.
According to an embodiment of the application, training the cycle generation countermeasure network CycleGAN model according to the training sample and the candidate image to obtain a target CycleGAN model for generating a defect image of the photovoltaic panel, further comprising: acquiring a total loss function of the cycleGAN model based on the discrimination results of the first discriminator and the second discriminator; and adjusting model parameters of the CycleGAN model based on the total loss function, and continuing training the adjusted CycleGAN model for the next time until a training end condition is met to obtain the target CycleGAN model.
According to an embodiment of the present application, the obtaining a loss function of the CycleGAN model based on the discrimination results of the first and second discriminators includes: acquiring a first loss function of the cycleGAN model according to the normal sample image, the second normal candidate image and the discrimination result of the defect sample image and the first defect candidate image; acquiring a second loss function of the cycleGAN model according to the normal sample image, the first normal candidate image and the judgment result of the defect sample image and the second defect candidate image; acquiring a third loss function of the cycleGAN model according to the normal sample image, the third normal candidate image and the judgment result of the defect sample image and the third defect candidate image; and acquiring the total loss function of the cycleGAN model according to the first loss function, the second loss function and the third loss function.
According to an embodiment of the application, the method further comprises: obtaining a target optimization function; and adjusting the model parameters of the cycleGAN model according to the target optimization function and the total loss function.
According to a second aspect of the application, a photovoltaic panel defect image generation method based on a CycleGAN model is provided, and comprises the following steps: acquiring a photovoltaic panel image to be processed; and inputting the photovoltaic panel image to be processed into a target cycleGAN model for generating a photovoltaic panel defect image so as to obtain the defect image of the photovoltaic panel.
According to a third aspect of the present application, there is provided a training apparatus for generating a CycleGAN model of a defect image of a photovoltaic panel, comprising: the device comprises a first acquisition module, a second acquisition module and a third acquisition module, wherein the first acquisition module is used for acquiring a training sample, and the training sample comprises a normal sample image and a defect sample image of a photovoltaic panel; the output module is used for inputting the training sample into a to-be-trained cyclic generation countermeasure network (CycleGAN) model for generating a photovoltaic panel defect image, and acquiring a candidate image of the photovoltaic panel output by the CycleGAN model, wherein the candidate image comprises a normal candidate image of the photovoltaic panel and a defect candidate image of the photovoltaic panel; and the training module is used for training the cycleGAN model according to the training sample and the candidate image so as to obtain a target cycleGAN model for generating a defect image of the photovoltaic panel.
The training device for the CycleGAN model for generating the defect images of the photovoltaic panel according to the above embodiment of the present application may further have the following additional technical features:
according to an embodiment of the present application, the output module is further configured to: generating a first defect candidate image and a third defect candidate image corresponding to the normal sample image, and generating a first normal candidate image corresponding to the first defect candidate image; and generating a corresponding second normal candidate image and a third normal candidate image according to the defect sample image, and generating a corresponding second defect candidate image according to the second normal candidate image.
According to an embodiment of the present application, the CycleGAN model includes a first generator and a second generator, and the output module is further configured to: inputting the normal sample image into the first generator, outputting the first defect candidate image by the first generator based on the normal sample image, and inputting the first defect candidate image into the second generator, outputting the first normal candidate image by the second generator based on the first defect candidate image, and inputting the defect sample image into the first generator, outputting the third defect candidate image by the first generator based on the defect sample image. Inputting the defect sample image into the second generator, outputting the second normal candidate image from the defect sample image by the second generator, inputting the second normal candidate image into the first generator, outputting the second defect candidate image from the second normal candidate image by the first generator, and inputting the normal sample image into the second generator, outputting the third normal candidate image from the normal sample image by the second generator.
According to an embodiment of the application, the CycleGAN model includes a first discriminator and a second discriminator, and the apparatus is further configured to: and judging the normal sample image of the photovoltaic panel and the normal candidate image of the photovoltaic panel according to the first discriminator. And judging the defect sample image of the photovoltaic panel and the defect candidate image of the photovoltaic panel according to a second judging device.
In an embodiment of the present application, the training module is further configured to: acquiring a total loss function of the cycleGAN model based on the discrimination results of the first discriminator and the second discriminator; and adjusting model parameters of the CycleGAN model based on the total loss function, and continuing to train the adjusted CycleGAN model for the next time until a training end condition is met to obtain the target CycleGAN model.
In an embodiment of the present application, the training module is further configured to: acquiring a first loss function of the cycleGAN model according to the normal sample image, the second normal candidate image and the discrimination result of the defect sample image and the first defect candidate image; acquiring a second loss function of the cycleGAN model according to the normal sample image, the first normal candidate image and the discrimination result of the defect sample image and the second defect candidate image; acquiring a third loss function of the cycleGAN model according to the normal sample image, the third normal candidate image and the discrimination result of the defect sample image and the third defect candidate image; and acquiring the total loss function of the cycleGAN model according to the first loss function, the second loss function and the third loss function.
In one embodiment of the present application, the apparatus is further configured to: obtaining a target optimization function; and adjusting the model parameters of the cycleGAN model according to the target optimization function and the total loss function.
According to a fourth aspect of the present application, there is provided a photovoltaic panel defect image generating apparatus based on a CycleGAN model, comprising: the acquisition module is used for acquiring a photovoltaic panel image to be processed; and the input module is used for inputting the photovoltaic panel defect image to be processed into a target cycleGAN model for generating the photovoltaic panel defect image so as to obtain the photovoltaic panel defect image.
In order to achieve the above object, a fifth aspect of the present application provides an electronic device, comprising: the device comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein when the processor executes the program, the training method of the CycleGAN model for generating the defect image of the photovoltaic panel according to the first aspect or the method for generating the defect image of the photovoltaic panel based on the CycleGAN model according to the second aspect is realized.
In order to achieve the above object, a sixth aspect of the present application proposes a non-transitory computer-readable storage medium storing computer instructions for causing the computer to execute the training method of the CycleGAN model for generating a defect image of a photovoltaic panel according to the first aspect or the method of generating a defect image of a photovoltaic panel based on the CycleGAN model according to the second aspect.
In order to achieve the above object, a seventh aspect of the present application proposes a computer program product, which comprises a computer program that, when being executed by a processor, implements the method for training a CycleGAN model for generating a defect image of a photovoltaic panel according to the first aspect or the method for generating a defect image of a photovoltaic panel based on a CycleGAN model according to the second aspect.
The technical scheme provided by the embodiment of the application at least comprises the following beneficial effects:
the application provides a training method of a cycleGAN model for generating a defect image of a photovoltaic panel, which can convert a normal image of the photovoltaic panel into different types of defect images of the photovoltaic panel through a target cycleGAN model for generating the defect image of the photovoltaic panel, solve the problem of few defect images of the photovoltaic panel and improve the accuracy of defect detection of the photovoltaic panel.
It should be understood that the statements in this section do not necessarily identify key or critical features of the embodiments of the present application, nor do they limit the scope of the present application. Other features of the present application will become apparent from the following description.
Drawings
The drawings are included to provide a better understanding of the present solution and are not to be considered limiting of the present application. Wherein:
fig. 1 is a schematic flowchart of a training method of a CycleGAN model for generating a defect image of a photovoltaic panel according to an embodiment of the present disclosure;
FIG. 2 is a schematic flowchart of another training method for generating a cycleGAN model of a defect image of a photovoltaic panel according to an embodiment of the present disclosure;
FIG. 3 is a schematic flowchart of another training method for generating a cycleGAN model of a defect image of a photovoltaic panel according to an embodiment of the present disclosure;
fig. 4 is a schematic structural diagram of a CycleGAN model provided in an embodiment of the present application;
fig. 5 is a schematic structural diagram of a CycleGAN model provided in an embodiment of the present application;
FIG. 6 is a schematic flowchart of another training method for generating a cycleGAN model of a defect image of a photovoltaic panel according to an embodiment of the present disclosure;
FIG. 7 is a schematic flowchart of another training method for generating a CycleGAN model of a defect image of a photovoltaic panel according to an embodiment of the present disclosure;
fig. 8 is a schematic flowchart of a method for generating a defect image of a photovoltaic panel based on a CycleGAN model according to an embodiment of the present application;
FIG. 9 is a schematic structural diagram of a model training apparatus according to an embodiment of the present disclosure;
fig. 10 is a schematic structural diagram of a photovoltaic panel defect image generating apparatus based on a CycleGAN model according to an embodiment of the present application;
fig. 11 is a schematic structural diagram of an electronic device according to an embodiment of the present application.
Detailed Description
The following description of the exemplary embodiments of the present application, taken in conjunction with the accompanying drawings, includes various details of the embodiments of the application for the understanding of the same, which are to be considered exemplary only. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present application. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.
The following embodiment is used to describe the training method of the CycleGAN model for generating the defect image of the photovoltaic panel in detail.
Fig. 1 is a schematic flowchart of a training method of a CycleGAN model for generating a defect image of a photovoltaic panel according to an embodiment of the present disclosure.
As shown in fig. 1, the training method for a CycleGAN model for generating a defect image of a photovoltaic panel provided by this embodiment specifically includes the following steps:
s101, obtaining a training sample, wherein the training sample comprises a normal sample image and a defect sample image of the photovoltaic panel.
It should be noted that after the training sample is obtained, the training sample may be labeled, and the training sample is divided into a normal sample image and a defect sample image of the photovoltaic panel.
Further, the defect sample image of the photovoltaic panel may be a damaged photovoltaic panel surface, a dust on the photovoltaic panel surface, a sheltered photovoltaic panel surface, a smudge on the photovoltaic panel surface, etc. Further, the type that photovoltaic panel surface was sheltered from can be divided into the photovoltaic panel surface and sheltered from by the shadow, the photovoltaic panel surface is sheltered from by the plant, the photovoltaic panel surface is sheltered from by the building, the photovoltaic panel surface is sheltered from by other subassemblies etc. further there is dirty type in photovoltaic panel surface can be divided into the photovoltaic panel surface and have bird excrement, there is dirt in photovoltaic panel surface, there is inorganic salt scale deposit etc. in the photovoltaic panel surface.
S102, inputting a training sample into a to-be-trained cyclic generation countermeasure network CycleGAN model for generating a defect image of the photovoltaic panel, and acquiring a candidate image of the photovoltaic panel output by the CycleGAN model, wherein the candidate image comprises a normal candidate image of the photovoltaic panel and the defect candidate image of the photovoltaic panel.
In the embodiment of the application, after the training sample is obtained, the training sample can be input into a CycleGAN model, and then the normal candidate image of the photovoltaic panel and the defect candidate image of the photovoltaic panel output by the CycleGAN model are obtained.
Wherein, the CycleGAN model comprises 2 generators, namely a first generator and a second generator.
For example, a normal sample image x of a photovoltaic panel in a training sample is input into a first generator in a CycleGAN model, a first defect candidate image x 'corresponding to the normal sample image x may be obtained, a defect sample image y of the photovoltaic panel in the training sample is input into a second generator in the CycleGAN model, and a second normal candidate image y' corresponding to the defect sample image y may be obtained.
Further, after the first defect candidate image and the second normal candidate image are acquired, the first defect candidate image and the second normal candidate image may be input into a CycleGAN model to acquire a first normal candidate image corresponding to the first defect candidate image and a second defect candidate image corresponding to the second normal candidate image.
For example, the first defect candidate image x 'may be input to the second generator in the CycleGAN model to obtain the first normal candidate image x "corresponding to the first defect candidate image x', and the second normal candidate image y 'may be input to the first generator in the CycleGAN model to obtain the second defect candidate image y" corresponding to the second normal candidate image y'.
S103, training the cycleGAN model according to the training sample and the candidate image to obtain a target cycleGAN model for generating a defect image of the photovoltaic panel.
It should be noted that, in the related art, the defect images of the photovoltaic panel are fewer, and the method and the device can convert a large number of normal images of the photovoltaic panel into different types of defect candidate images based on the target CycleGAN model, thereby solving the problem of fewer defect images of the photovoltaic panel in the prior art.
In the embodiment of the application, after the training sample and the candidate image are obtained, the cycleGAN model can be trained according to the training sample and the candidate image, and then the target cycleGAN model for generating the defect image of the photovoltaic panel is obtained.
Further, after the target CycleGAN model used for generating the defect images of the photovoltaic panel is obtained, the normal sample images of the photovoltaic panel can be converted into different types of defect images of the photovoltaic panel based on the target CycleGAN model.
According to the training method for the cycleGAN model for generating the defect images of the photovoltaic panel, the training samples are obtained and input into the to-be-trained cyclic generation countermeasure network cycleGAN model for generating the defect images of the photovoltaic panel, the candidate images of the photovoltaic panel output by the cycleGAN model are obtained, wherein the candidate images comprise the normal candidate images of the photovoltaic panel and the defect candidate images of the photovoltaic panel, and the cycleGAN model is trained according to the training samples and the candidate images to obtain the target cycleGAN model for generating the defect images of the photovoltaic panel. Therefore, the target CycleGAN model for generating the defect images of the photovoltaic panel can convert the normal images of the photovoltaic panel into the defect images of the photovoltaic panel of different types, the problem that the defect images of the photovoltaic panel are few is solved, and the defect detection accuracy of the photovoltaic panel is improved.
In the present application, when attempting to acquire a candidate image of a photovoltaic panel output by a CycleGAN model, a normal candidate image and a defect candidate image of the photovoltaic panel output by the CycleGAN model may be acquired based on the first generator and the second generator in the CycleGAN model.
As a possible implementation manner, as shown in fig. 2, on the basis of the above steps, a specific process of inputting the training sample into the CycleGAN model in the above step S101 and acquiring the candidate image of the photovoltaic panel output by the CycleGAN model includes the following steps:
s201, generating a corresponding first defect candidate image and a corresponding third normal candidate image according to the normal sample image, and generating a corresponding first normal candidate image according to the first defect candidate image.
In the embodiment of the present application, the normal sample image may be input to the first generator, the first generator outputs the first defect candidate image based on the normal sample image, and the first defect candidate image may be input to the second generator, and the second generator outputs the second normal candidate image based on the first defect candidate image.
Further, in the embodiment of the present application, the normal sample image may also be input to the second generator, and the third normal candidate image may be output by the second generator according to the normal sample image.
S202, generating a corresponding first normal candidate image and a corresponding third defect candidate image according to the defect sample image, and generating a corresponding second defect candidate image according to the first normal candidate image.
In the embodiment of the present application, the defect sample image may be input to the second generator, the first normal candidate image may be output by the second generator based on the defect sample image, the first normal candidate image may be input to the first generator, and the second defect candidate image may be output by the first generator based on the first normal candidate image.
Further, in the embodiment of the present application, the defect sample image may be input to the first generator, and the third defect candidate image may be output by the first generator according to the defect sample image.
It should be noted that, in the present application, the CycleGAN model includes two discriminators, that is, a first discriminator and a second discriminator, and after the candidate image of the photovoltaic panel output by the CycleGAN model is obtained, the normal sample image of the photovoltaic panel and the normal candidate image of the photovoltaic panel may be discriminated based on the first discriminator and the second discriminator, so as to discriminate the defect sample image of the photovoltaic panel and the defect candidate image of the photovoltaic panel.
As a possible implementation manner, as shown in fig. 3, on the basis of the above steps, a specific process of distinguishing a normal candidate image and a defect candidate image of a photovoltaic panel in the above steps includes the following steps:
s301, according to the first discriminator, discriminating the normal sample image of the photovoltaic panel and the normal candidate image of the photovoltaic panel.
It should be noted that the first discriminator may be configured to distinguish between the normal sample image and the normal candidate image.
S302, judging the defect sample image of the photovoltaic panel and the defect candidate image of the photovoltaic panel according to a second judging device.
It should be noted that the second discriminator may be used to distinguish the defect sample image from the defect candidate image.
The following explains a process of obtaining a candidate image of a photovoltaic panel based on a CycleGAN model, discriminating a normal sample image of the photovoltaic panel from a normal candidate image of the photovoltaic panel, and discriminating a defect sample image of the photovoltaic panel from a defect candidate image of the photovoltaic panel.
For example, the generator G is a first generator, the generator F is a second generator, the discriminator Dx is a first discriminator, and the discriminator Dy is a second discriminator.
As shown in fig. 4, for the normal sample image x, the defect sample image y, the normal sample image x is input to the generator G (first generator), the first defect candidate image x ' may be acquired, the first defect candidate image x ' is input to the generator F (second generator), the first normal candidate image x "may be acquired, the normal sample image x is input to the generator F (second generator), the third normal candidate image x '" may be acquired, the discriminator Dx (first discriminator) may be used to distinguish the normal sample image x from the normal candidate images (the first normal candidate image x ", the second normal candidate image y ', and the third normal candidate image x '"), and the discriminator Dy (second discriminator) may be used to distinguish the defect sample image y from the defect candidate images (the first defect candidate image x ', the second defect candidate image y ", and the third defect candidate image y '").
As shown in fig. 5, for the normal sample image x, the defect sample image y is input to the generator F (second generator), the second normal candidate image y ' may be obtained, the second normal candidate image y ' is input to the generator G (first generator), the second defect candidate image y "may be obtained, the defect sample y image is input to the generator G (first generator), the third defect candidate image y '" may be obtained, the discriminator Dx (first discriminator) may be used to distinguish the normal sample image x from the normal candidate images (first normal candidate image x ", second normal candidate image y ', and third normal candidate image x '"), and the discriminator Dy (second discriminator) may be used to distinguish the defect sample image y from the defect candidate images (first defect candidate image x ', second defect candidate image y ", and third defect candidate image y '").
Further, a total loss function of the cycleGAN model can be obtained according to the judgment result, and then model parameters of the cycleGAN model are adjusted to obtain the target cycleGAN model.
As a possible implementation manner, as shown in fig. 6, on the basis of the above steps, a specific process of training the CycleGAN model according to the training sample and the candidate image in step S103 to obtain the target CycleGAN model for generating the defect image of the photovoltaic panel includes the following steps:
s601, acquiring a total loss function of the cycleGAN model based on the discrimination results of the first discriminator and the second discriminator.
As a possible implementation manner, as shown in fig. 7, in addition to the above steps, a specific process of acquiring a total loss function of the CycleGAN model based on the determination results of the first and second discriminators in step S601 includes the following steps:
s701, acquiring a first loss function of the cycleGAN model according to the discrimination results of the normal sample image and the normal candidate image and the discrimination results of the defect sample image and the defect candidate image.
Wherein the first Loss function of the CycleGAN model, loss GAN The defect candidate image generated by the generator G (first generator) is represented by the discriminator Dy (second generator) as much as possibleA discriminator) judges as a defective sample image; the normal candidate image generated by the generator F (second generator) is determined as a normal sample image by the discriminator Dx (first discriminator) as much as possible.
Wherein, the first Loss function Loss of the cycleGAN model GAN The following formula can be used for the acquisition:
Figure BDA0003733835530000091
s702, acquiring a second loss function of the cycleGAN model according to the discrimination result of the normal sample image and the first normal candidate image and the discrimination result of the defect sample image and the second defect candidate image.
The second loss function of the CycleGAN model indicates that the normal sample image x passes through the generator G (the first generator) to generate the first defect candidate image x ', and then passes through the generator F (the second generator) to generate the first normal candidate image x', so that the generated normal candidate image and the normal sample image are as similar as possible.
Wherein, the second Loss function Loss of the cycleGAN model Cycle The following formula can be used for the acquisition:
Figure BDA0003733835530000101
and S703, acquiring a third loss function of the cycleGAN model according to the similarity of the normal sample image, the first normal candidate image and the second normal candidate image and the similarity of the defect sample image, the first defect candidate and the second defect candidate image.
Wherein, the third Loss function Loss of the cycleGAN model Identity Representing the content basis of the images generated by the warranty generators G, F and the input imageThe method is consistent, and only the conversion of normal and defective images is realized. That is, the purpose of the generator F (second generator) is to generate a normal candidate image, the normal sample image x is input, the generated image F (x) should be as similar as possible to the normal sample image x, and similarly, the purpose of the generator G (second generator) is to generate a defect image, the defect sample image y is input, and the generated image G (y) should be as similar as possible to the defect sample image y.
The third loss function of the CycleGAN model can be obtained by the following formula:
Figure BDA0003733835530000102
and S704, acquiring a total loss function of the cycleGAN model according to the first loss function, the second loss function and the third loss function.
It should be noted that, according to the first Loss function Loss GAN Second Loss function Loss CycleGAN And a third Loss function Loss Identity The total Loss function Loss of the cycleGAN model can be obtained CycleGAN 。
For example, the Loss-total function Loss CycleGAN ＝Loss GAN +λ i Loss Cycle +λ 2 Loss Identity 。
Wherein λ is 1 And λ 2 The degree of influence on the overall loss function of the CycleGAN model can be set.
And S602, adjusting model parameters of the CycleGAN model based on the total loss function, and continuing training the adjusted CycleGAN model for the next time until the training end condition is met to obtain the target CycleGAN model.
In the embodiment of the application, after the total loss function is obtained, the target optimization function can be obtained, and the CycleGAN is adjusted to generate the model parameters of the confrontation network model according to the target optimization function and the total loss function.
For example, in attempting to adjust the CycleGAN to generate model parameters for the antagonistic network model based on the objective optimization function and the total loss function, the adjustments can be made according to the following formula:
Figure BDA0003733835530000103
wherein the content of the first and second substances,
Figure BDA0003733835530000111
optimizing function, loss, for an object CycleGAN As a function of the total loss.
Further, the adjusted cycleGAN model can be trained for the next time until the training end condition is met, and the target cycleGAN model is obtained.
The training end condition may be set according to an actual situation, and the present application is not limited.
Alternatively, the training end condition may be set such that the number of training times reaches a preset number of training times.
For example, the training end condition may be set to set the number of times of training to 10000.
Further, after the target CycleGAN model for generating the defect image of the photovoltaic panel is obtained, the mapping relation between the normal image of the photovoltaic panel and the defect images of different types can be obtained according to the target CycleGAN model for generating the defect image of the photovoltaic panel.
Further, after the mapping relation between the normal images of the photovoltaic panel and the defect images of different types is obtained, the normal images of the photovoltaic panel can be converted into the defect images of different types, and a large number of defect images of the photovoltaic panel are obtained.
The training method for the cycleGAN model for generating the defect images of the photovoltaic panel can acquire the mapping relation between the normal images of the photovoltaic panel and the defect images of different types through the target cycleGAN model for generating the defect images of the photovoltaic panel, further convert the normal images of the photovoltaic panel into the defect images of different types, obtain a large number of defect images of the photovoltaic panel, solve the problem that the defect images of the photovoltaic panel are few, improve the accuracy of defect detection of the photovoltaic panel, and further improve the power generation efficiency of a photovoltaic module.
The method for generating a defect image of a photovoltaic panel based on a CycleGAN model according to the present application is described in detail below with reference to examples.
Fig. 8 is a schematic flow chart of a method for generating a defect image of a photovoltaic panel based on a CycleGAN model according to an embodiment of the present disclosure.
S801, acquiring a photovoltaic panel image to be processed.
Alternatively, the photovoltaic panel image to be processed may be a normal image of the photovoltaic panel.
S802, inputting the photovoltaic panel defect image to be processed into a target cycleGAN model for generating the photovoltaic panel defect image so as to obtain the photovoltaic panel defect image.
In the embodiment of the application, after the to-be-processed image is obtained, the to-be-processed defect image of the photovoltaic panel is input into a target CycleGAN model for generating the defect image of the photovoltaic panel, and then a large number of defect images of the photovoltaic panel can be obtained.
According to the photovoltaic panel defect image generation method based on the cycleGAN model, the photovoltaic panel defect image to be processed is input into the target cycleGAN model for generating the photovoltaic panel defect image by acquiring the photovoltaic panel image to be processed, so that the photovoltaic panel defect image is obtained. Therefore, the target CycleGAN model based on generation of the photovoltaic panel defect images can convert normal images of the photovoltaic panel into photovoltaic panel defect images of different types, the problem that the number of the photovoltaic panel defect images is small is solved, and the accuracy of photovoltaic panel defect detection is improved.
In order to implement the foregoing embodiment, this embodiment provides a training device for a CycleGAN model used for generating a defect image of a photovoltaic panel, and fig. 9 is a schematic structural diagram of the training device for a model provided in this embodiment.
As shown in fig. 9, the training apparatus 1000 for generating a CycleGAN model of a defect image of a photovoltaic panel includes: a first acquisition module 110, an output module 120 and a second acquisition module 130.
A first obtaining module 110, configured to obtain a training sample, where the training sample includes a normal sample image and a defect sample image of a photovoltaic panel;
the output module 120 is configured to input the training sample into a to-be-trained cyclic generation countermeasure network CycleGAN model for generating a photovoltaic panel defect image, and acquire a candidate image of the photovoltaic panel output by the CycleGAN model, where the candidate image includes a normal candidate image of the photovoltaic panel and a defect candidate image of the photovoltaic panel;
and the training module 130 is configured to train the cyclically generated countermeasure network CycleGAN model according to the training samples and the candidate images to obtain a target CycleGAN model for generating a defect image of the photovoltaic panel.
According to an embodiment of the present application, the output module 120 is further configured to: generating a first defect candidate image and a third normal candidate image corresponding to the normal sample image, and generating a first normal candidate image corresponding to the first defect candidate image; and generating a corresponding second normal candidate image and a corresponding third defect candidate image according to the defect sample image, and generating a corresponding second defect candidate image according to the second normal candidate image.
According to an embodiment of the present application, the CycleGAN model includes a first generator and a second generator, and the output module 120 is further configured to: inputting the normal sample image into the first generator, outputting the first defect candidate image from the normal sample image by the first generator, and inputting the first defect candidate image into the second generator, outputting the first normal candidate image from the first defect candidate image by the second generator, and inputting the defect sample image into the first generator, outputting the third defect candidate image from the defect sample image by the first generator. Inputting the defect sample image into the second generator, outputting the second normal candidate image from the defect sample image by the second generator, inputting the second normal candidate image into the first generator, outputting the second defect candidate image from the second normal candidate image by the first generator, and inputting the normal sample image into the second generator, outputting the third normal candidate image from the normal sample image by the second generator.
According to an embodiment of the present application, the CycleGAN model includes a first discriminator and a second discriminator, and the apparatus 1000 is further configured to: and judging the normal sample image of the photovoltaic panel and the normal candidate image of the photovoltaic panel according to the first discriminator. And judging the defect sample image of the photovoltaic panel and the defect candidate image of the photovoltaic panel according to a second judging device.
In an embodiment of the present application, the training module 130 is further configured to: based on the discrimination results of the first and second discriminators, acquiring a total loss function of the cycleGAN model; and adjusting model parameters of the CycleGAN model based on the total loss function, and continuing to train the adjusted CycleGAN model for the next time until a training end condition is met to obtain the target CycleGAN model.
In an embodiment of the present application, the training module 130 is further configured to: acquiring a first loss function of the cycleGAN model according to the normal sample image, the second normal candidate image and the judgment result of the defect sample image and the first defect candidate image; acquiring a second loss function of the cycleGAN model according to the normal sample image, the first normal candidate image and the judgment result of the defect sample image and the second defect candidate image; acquiring a third loss function of the cycleGAN model according to the normal sample image, the third normal candidate image and the judgment result of the defect sample image and the third defect candidate image; and acquiring the total loss function of the cycleGAN model according to the first loss function, the second loss function and the third loss function.
In an embodiment of the present application, the apparatus 1000 is further configured to: obtaining a target optimization function; and adjusting the model parameters of the cycleGAN model according to the target optimization function and the total loss function.
According to the training device for generating the cycleGAN model of the defect image of the photovoltaic panel, the training sample is obtained and comprises the normal sample image and the defect sample image of the photovoltaic panel, the training sample is input into the to-be-trained cyclic generation countermeasure network cycleGAN model for generating the defect image of the photovoltaic panel, the candidate image of the photovoltaic panel output by the cycleGAN model is obtained, the candidate image comprises the normal candidate image of the photovoltaic panel and the defect candidate image of the photovoltaic panel, and the cycleGAN model is trained according to the training sample and the candidate image to obtain the target cycleGAN model for generating the defect image of the photovoltaic panel. Therefore, the target CycleGAN model for generating the defect images of the photovoltaic panel can convert the normal images of the photovoltaic panel into the defect images of the photovoltaic panel of different types, the problem that the defect images of the photovoltaic panel are few is solved, and the defect detection accuracy of the photovoltaic panel is improved.
In order to implement the foregoing embodiment, the present embodiment provides a photovoltaic panel defect image generating apparatus based on a CycleGAN model, and fig. 10 is a schematic structural diagram of the photovoltaic panel defect image generating apparatus based on the CycleGAN model provided in the embodiment of the present application.
As shown in fig. 10, the training apparatus 2000 for generating a CycleGAN model of a defect image of a photovoltaic panel includes: a second acquisition module 210 and an input module 220.
An obtaining module 210, configured to obtain a photovoltaic panel image to be processed;
and the input module 220 is used for inputting the photovoltaic panel defect image to be processed into a target CycleGAN model for generating a photovoltaic panel defect image so as to obtain the photovoltaic panel defect image.
According to the photovoltaic panel defect image generation device based on the cycleGAN model, the photovoltaic panel defect image to be processed is input into the target cycleGAN model for generating the photovoltaic panel defect image by acquiring the photovoltaic panel image to be processed, so that the photovoltaic panel defect image is obtained. Therefore, the target CycleGAN model based on generation of the photovoltaic panel defect images can convert normal images of the photovoltaic panel into photovoltaic panel defect images of different types, the problem that the number of the photovoltaic panel defect images is small is solved, and the accuracy of photovoltaic panel defect detection is improved.
In order to implement the above embodiments, the present application also proposes an electronic device 3000, as shown in fig. 11, including: a memory 310, a processor 320 and a computer program stored in the memory 310 and executable on the processor 320, wherein the processor implements the method for training a CycleGAN model for generating a defect image of a photovoltaic panel according to the first aspect or the method for generating a defect image of a photovoltaic panel based on the CycleGAN model according to the second aspect.
In order to achieve the above embodiments, the present application proposes a non-transitory computer readable storage medium storing computer instructions, wherein the computer instructions are configured to cause the computer to execute the training method of the CycleGAN model for generating a defect image of a photovoltaic panel according to the first aspect or the method of generating a defect image of a photovoltaic panel based on the CycleGAN model according to the second aspect.
In order to implement the foregoing embodiments, the present application further provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, the computer program implements the method for training a CycleGAN model for generating a defect image of a photovoltaic panel according to the first aspect or the method for generating a defect image of a photovoltaic panel based on a CycleGAN model according to the second aspect.
It should be understood that various forms of the flows shown above, reordering, adding or deleting steps, may be used. For example, the steps described in the present application may be executed in parallel, sequentially, or in different orders, and the present invention is not limited thereto as long as the desired results of the technical solutions disclosed in the present application can be achieved.
The above-described embodiments should not be construed as limiting the scope of the present application. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may be made, depending on design requirements and other factors. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present application, are intended to be included within the scope of this application.

Claims (13)

1. A training method for a CycleGAN model for generating images of defects of a photovoltaic panel, characterized in that it comprises:
acquiring a training sample, wherein the training sample comprises a normal sample image and a defect sample image of a photovoltaic panel;
inputting the training sample into a cyclic generation countermeasure network CycleGAN model to be trained for generating a defect image of the photovoltaic panel, and acquiring a candidate image of the photovoltaic panel output by the CycleGAN model, wherein the candidate image comprises a normal candidate image of the photovoltaic panel and a defect candidate image of the photovoltaic panel;
and training the cycleGAN model according to the training sample and the candidate image to obtain a target cycleGAN model for generating a defect image of the photovoltaic panel.
2. The training method according to claim 1, wherein the inputting the training sample into a CycleGAN model to be trained for generating a defect image of a photovoltaic panel, and obtaining a candidate image of the photovoltaic panel output by the CycleGAN model, further comprises:
generating a first defect candidate image and a third defect candidate image corresponding to the normal sample image, and generating a first normal candidate image corresponding to the first defect candidate image;
and generating a corresponding second normal candidate image and a third normal candidate image according to the defect sample image, and generating a corresponding second defect candidate image according to the second normal candidate image.
3. The training method of claim 2, wherein the CycleGAN model comprises a first generator and a second generator, the method further comprising:
inputting the normal sample image into the first generator, outputting the first defect candidate image from the normal sample image by the first generator, and inputting the first defect candidate image into the second generator, outputting the first normal candidate image from the first defect candidate image by the second generator, and inputting the normal sample image into the second generator, outputting the third defect candidate image from the normal sample image by the second generator;
inputting the defect sample image into the second generator, outputting the second normal candidate image by the second generator based on the defect sample image, inputting the second normal candidate image into the first generator, outputting the second defect candidate image by the first generator based on the second normal candidate image, inputting the defect sample image into the first generator, and outputting the third normal candidate image by the first generator based on the defect sample image.
4. A training method according to claim 3, wherein the CycleGAN model comprises a first arbiter and a second arbiter, the method further comprising:
according to the first discriminator, discriminating the normal sample image of the photovoltaic panel and the normal candidate image of the photovoltaic panel;
and judging the defect sample image of the photovoltaic panel and the defect candidate image of the photovoltaic panel according to the second discriminator.
5. The training method of claim 4, wherein the training of the cyclic generation countermeasure network CycleGAN model from the training samples and the candidate images to obtain a target CycleGAN model for generating a defect image of a photovoltaic panel further comprises:
acquiring a total loss function of the cycleGAN model based on the discrimination results of the first discriminator and the second discriminator;
and adjusting model parameters of the CycleGAN model based on the total loss function, and continuing to train the adjusted CycleGAN model for the next time until a training end condition is met to obtain the target CycleGAN model.
6. The training method according to claim 5, wherein the obtaining a loss function of the CycleGAN model based on the discrimination results of the first and second discriminators comprises:
acquiring a first loss function of the cycleGAN model according to the normal sample image, the second normal candidate image and the judgment result of the defect sample image and the first defect candidate image;
acquiring a second loss function of the cycleGAN model according to the normal sample image, the first normal candidate image and the discrimination result of the defect sample image and the second defect candidate image;
acquiring a third loss function of the cycleGAN model according to the normal sample image, the third normal candidate image and the judgment result of the defect sample image and the third defect candidate image;
and acquiring the total loss function of the cycleGAN model according to the first loss function, the second loss function and the third loss function.
7. The training method of claim 6, further comprising: :
obtaining a target optimization function;
and adjusting the model parameters of the cycleGAN model according to the target optimization function and the total loss function.
8. A photovoltaic panel defect image generation method based on a CycleGAN model comprises the following steps:
acquiring a photovoltaic panel image to be processed;
and inputting the photovoltaic panel image to be processed into a target cycleGAN model for generating a photovoltaic panel defect image so as to obtain the photovoltaic panel defect image.
9. Training device for generating a CycleGAN model of a defect image of a photovoltaic panel, characterized in that it comprises:
the device comprises a first acquisition module, a second acquisition module and a third acquisition module, wherein the first acquisition module is used for acquiring training samples, and the training samples comprise normal sample images and defect sample images of a photovoltaic panel;
the output module is used for inputting the training sample into a to-be-trained cyclic generation countermeasure network (CycleGAN) model for generating a photovoltaic panel defect image, and acquiring a candidate image of the photovoltaic panel output by the CycleGAN model, wherein the candidate image comprises a normal candidate image of the photovoltaic panel and a defect candidate image of the photovoltaic panel;
and the training module is used for training the cycleGAN model according to the training sample and the candidate image so as to obtain a target cycleGAN model for generating a defect image of the photovoltaic panel.
10. A photovoltaic panel defect image generation device based on a CycleGAN model comprises:
the acquisition module acquires a photovoltaic panel image to be processed;
and the input module is used for inputting the photovoltaic panel defect image to be processed into a target cycleGAN model for generating the photovoltaic panel defect image so as to obtain the photovoltaic panel defect image.
11. An electronic device comprising a processor and a memory;
wherein the processor runs a program corresponding to the executable program code by reading the executable program code stored in the memory for implementing the method according to any one of claims 1-7 or claim 8.
12. A computer-readable storage medium, on which a computer program is stored, which program, when being executed by a processor, carries out the method according to any one of claims 1-7 or claim 8.
13. A computer program product comprising a computer program which, when executed by a processor, implements the method according to any one of claims 1-7 or claim 8.
CN202210790763.5A 2022-07-06 2022-07-06 Training method and device for CycleGAN model for generating defect images of photovoltaic panel Pending CN115147377A (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN202210790763.5A CN115147377A (en) 2022-07-06 2022-07-06 Training method and device for CycleGAN model for generating defect images of photovoltaic panel
PCT/CN2023/079045 WO2024007602A1 (en) 2022-07-06 2023-03-01 Method and apparatus for training cyclegan model for generating defect images of photovoltaic panel

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210790763.5A CN115147377A (en) 2022-07-06 2022-07-06 Training method and device for CycleGAN model for generating defect images of photovoltaic panel

Publications (1)

Publication Number Publication Date
CN115147377A true CN115147377A (en) 2022-10-04

Family

ID=83412835

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210790763.5A Pending CN115147377A (en) 2022-07-06 2022-07-06 Training method and device for CycleGAN model for generating defect images of photovoltaic panel

Country Status (2)

Country Link
CN (1) CN115147377A (en)
WO (1) WO2024007602A1 (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115311526A (en) * 2022-10-11 2022-11-08 江苏智云天工科技有限公司 Defect sample generation method and system based on improved Cycle GAN network
WO2024007602A1 (en) * 2022-07-06 2024-01-11 中国华能集团清洁能源技术研究院有限公司 Method and apparatus for training cyclegan model for generating defect images of photovoltaic panel

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117934470A (en) * 2024-03-22 2024-04-26 宁德时代新能源科技股份有限公司 Model training method, defect detection device, model training equipment and storage medium

Family Cites Families (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11501438B2 (en) * 2018-04-26 2022-11-15 Elekta, Inc. Cone-beam CT image enhancement using generative adversarial networks
JP2021168078A (en) * 2020-04-13 2021-10-21 株式会社小糸製作所 Image generation system and machine learning method for image generation system
CN112598611A (en) * 2020-12-16 2021-04-02 阳光保险集团股份有限公司 Method and device for synthesizing and identifying embossed bank card number image
CN112668623B (en) * 2020-12-22 2024-04-16 中国铁道科学研究院集团有限公司 Method and device for generating binaural pin defect sample based on generation countermeasure network
CN112906769A (en) * 2021-02-04 2021-06-04 国网河南省电力公司电力科学研究院 Power transmission and transformation equipment image defect sample amplification method based on cycleGAN
CN112950561B (en) * 2021-02-22 2022-07-26 中国地质大学（武汉） Optical fiber end face defect detection method, device and storage medium
CN113554630A (en) * 2021-07-30 2021-10-26 西安电子科技大学 Chip surface defect detection method, system, computer device and storage medium
CN113570603A (en) * 2021-09-26 2021-10-29 国网浙江省电力有限公司电力科学研究院 Contact network part defect data generation method and system based on cycleGAN
CN115147377A (en) * 2022-07-06 2022-10-04 中国华能集团清洁能源技术研究院有限公司 Training method and device for CycleGAN model for generating defect images of photovoltaic panel

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2024007602A1 (en) * 2022-07-06 2024-01-11 中国华能集团清洁能源技术研究院有限公司 Method and apparatus for training cyclegan model for generating defect images of photovoltaic panel
CN115311526A (en) * 2022-10-11 2022-11-08 江苏智云天工科技有限公司 Defect sample generation method and system based on improved Cycle GAN network

Also Published As

Publication number Publication date
WO2024007602A1 (en) 2024-01-11

Similar Documents

Publication Publication Date Title
CN115147377A (en) Training method and device for CycleGAN model for generating defect images of photovoltaic panel
Malof et al. Automatic solar photovoltaic panel detection in satellite imagery
CN107749058A (en) A kind of machine vision detection method and system of boiler tubing surface defect
CN103752533A (en) Method and device for sorting coal and gangue online through image method
JP2020053073A (en) Learning method, learning system, and learning program
CN106897998B (en) Method and system for predicting information of direct solar radiation intensity
CN116797533B (en) Appearance defect detection method and system for power adapter
CN110113327A (en) A kind of method and device detecting DGA domain name
CN117351659B (en) Hydrogeological disaster monitoring device and monitoring method
CN114548268A (en) Small sample garbage image classification method based on prototype network
CN112508278A (en) Multi-connected system load prediction method based on evidence regression multi-model
CN105760896A (en) Corrosion source joint de-noising method for multi-source heterogeneous big data
CN117349657A (en) Distributed data acquisition module and monitoring system for hydraulic engineering environment monitoring
CN110222814B (en) Ethylene cracking furnace tube heavy tube identification method based on embedded DCNN
CN108763289B (en) Massive heterogeneous sensor format data analysis method
CN106599910A (en) Printing file discriminating method based on texture recombination
CN116794553A (en) Intelligent fault diagnosis method and system for high-frequency power supply
CN115471845A (en) Converter station digital instrument identification method based on deep learning and OpenCV
CN113313178B (en) Cross-domain image example level active labeling method
Mazlan et al. Convolution Neural Network (CNN) Architectures Analysis for Photovoltaic (PV) Module Defect Images Classification
Heng et al. Anti-vibration hammer detection in UAV image
CN116206169A (en) Intelligent gangue target detection method
Hu et al. A detecting algorithm for occlusion on the surface of photovoltaic modules based on improved YOLOv5
CN110853010B (en) High-speed railway cable detection method based on FWA and SM
CN111488785A (en) Rotary kiln working condition detection method based on in-kiln image

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination