CN114022426A

CN114022426A - Serial analysis application method and system for AI (Artificial Intelligence) identification of solar cell EL (electro-luminescence) image

Info

Publication number: CN114022426A
Application number: CN202111247881.3A
Authority: CN
Inventors: 不公告发明人
Original assignee: Suzhou Sanxi Intelligent Technology Co ltd
Current assignee: Suzhou Sanxi Intelligent Technology Co ltd
Priority date: 2021-10-26
Filing date: 2021-10-26
Publication date: 2022-02-08

Abstract

The invention discloses a serial analysis application method and a serial analysis application system for AI identification of a solar cell EL image. When defect features are added to AI software, the device only needs to mark the defect features singly, so that a large amount of workload is reduced, iteration updating is more convenient, and meanwhile, the defect features with low tolerance cannot be missed, so that the serious missed defect in EL image detection is avoided, and a better data guidance effect is achieved.

Description

Serial analysis application method and system for AI (Artificial Intelligence) identification of solar cell EL (electro-luminescence) image

Technical Field

The invention relates to the field of solar cells, in particular to a serial analysis application method and system for AI identification of solar cell EL images.

Background

The AI artificial intelligence image recognition technology has been widely applied in the photovoltaic industry, especially in EL image recognition of solar cells and modules, more and more AI software is added in recent years, and the AI recognition process generally includes four parts, namely feature sample screening, feature labeling, training, prediction, and the like.

In actual operation, an EL image sample usually contains many different defect features, and conventionally, each different defect feature is completely marked in the sample, and then trained into a standard model, and the model is used to predict a new sample. However, from the perspective of the user, multiple feature labels are performed in one sample, the workload is large, the operation is complex, and once a defect is problematic or needs to be updated, the whole model needs to be modified;

in the invention of patent No. CN202010192560.7, an application of AI technology in ceramic wafer defect identification is proposed, and in the solar energy field, during the detection and use, the conventional way is to perform one-time identification prediction on the EL image to be identified by AI software, and identify the defect features of the EL image therefrom, while in the actual use, the tolerance for different types of defect features is different, whereas in the solar cell technology, there are twenty to thirty types of defect features altogether, wherein there are some defect features similar to them (such as black spots and over-etching), and only one-time identification prediction is performed, the AI software is likely to identify the defect features with high similarity as one of them, so that the defect with low tolerance is determined to be missed and even appears, the AI logically identifies the defect as a defect, a part of which is similar to the B defect, and the area is identified as not having a complete defect feature, and serious misjudgment results are caused.

Disclosure of Invention

The invention aims to solve the defects in the prior art, and provides a serial analysis application method for AI identification of a solar cell EL image.

In order to achieve the purpose, the invention adopts the following technical scheme: the method is characterized in that: the method comprises the following steps:

s1, marking defect features on the template EL image, where all the defect features existing on the template EL image need to be marked when marking the defect features;

s2, adding the marked template EL image to AI software for model learning, forming a standard model of the defect characteristics through the model learning, and memorizing;

s3, adding the EL image to be detected to AI software for identification and prediction, and marking the defect characteristics on the EL image to be detected;

and S4, outputting a defect characteristic prediction result graph of the EL image to be detected by the AI software, and correspondingly generating an information table of the defect characteristics.

Preferably, there are several defect features in S1, and the marking of the defect features is performed independently.

Preferably, the number of times of model learning is set to a plurality of batches, the same defect feature is added to the model learning in the same batch, and the model learning in different batches is independently performed.

Preferably, the number of times of model learning per batch is set by itself.

Preferably, the identification and prediction are performed n times, where n is the number of the defect feature categories learned and memorized by the AI software.

Preferably, the n recognition predictions are independent of each other.

Preferably, the information table generates m groups, where m is the number of defect feature categories existing on the EL image to be detected and predicted and identified by the AI software.

Preferably, the information table includes an area, an occurrence number, a confidence level and a gray level of the defect feature.

A serial analysis application method system for AI identification of solar cell EL image comprises a model learning module, a model memory module, an identification prediction module and a result output module;

the model learning module is used for training and learning template defect characteristics, forming a standard model through template defect characteristic memory, and transmitting the standard model to the model memory module;

the model memory module is used for storing the standard model;

the recognition and prediction module compares the standard model with the EL image to be detected, marks defect characteristics on the EL image to be detected, and finally transmits the marked EL image to the result output module;

and the result output module is used for calculating the marked defect characteristics on the marked EL image and generating a data table.

Preferably, a model adding module is arranged in the model learning module.

The invention has the following beneficial effects:

1. in the invention, the marks of a plurality of defect characteristics are independent, namely when a certain defect characteristic is marked, only the defect characteristic is marked, and other defect characteristics are not marked, so that when the defect characteristic is added to AI software, only the defect characteristic is required to be updated independently, and compared with the mode of updating once, all the defect characteristics are required to be marked, the time of more than several times is saved on the training time, thereby greatly increasing the iterative updating speed of the model, and when a manufacturer finds a problem in production and use, the problem response speed can be greatly increased, thereby reducing the loss as much as possible;

2. the prediction identification of different defect characteristics is carried out independently, and the AI software memorizes the defect characteristics for a plurality of times of identification prediction, so that when the defect characteristics with extremely high similarity are identified and predicted, only two groups of defect characteristics are judged to exist, and the defect characteristics with low tolerance cannot be missed, thereby avoiding the missed judgment of serious defects in the EL image detection and having better data guidance effect;

3. when the AI software is trained and learned, only one defect feature is trained and learned in the same batch, different defect features are mutually performed without mutual interference, so that the memory effect of the AI software is better, the accuracy of recognition and prediction is improved, and the learning times can be finely adjusted for the defect features with extremely high similarity, so that the AI learning effect is improved.

Drawings

FIG. 1 is a flow chart of an application method of the present invention;

FIG. 2 is a block diagram of the system of the present invention;

FIG. 3 is a system interface diagram of the present invention;

FIG. 4 is a comparison graph of black spots and over-etching under an EL image of a cell (black spots on the left side and over-etching on the right side);

fig. 5 is a comparison of scratch and belt mark under EL image of the cell (scratch on left and belt mark on right).

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicate orientations or positional relationships based on the orientations or positional relationships shown in the drawings, and are only for convenience of description and simplicity of description, but do not indicate or imply that the device or element being referred to must have a particular orientation, be constructed and operated in a particular orientation, and thus, should not be construed as limiting the present invention; the terms "first," "second," and "third" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance, and furthermore, unless otherwise explicitly stated or limited, the terms "mounted," "connected," and "connected" are to be construed broadly and may be, for example, fixedly connected, detachably connected, or integrally connected; can be mechanically or electrically connected; they may be connected directly or indirectly through intervening media, or they may be interconnected between two elements. The specific meanings of the above terms in the present invention can be understood in specific cases to those skilled in the art.

Referring to fig. 1, the present invention provides an embodiment: a serial analysis application method for AI identification of solar cell EL images comprises the following steps:

s1, marking the defect characteristics on the template EL image, wherein when the defect characteristics are marked, all the defect characteristics existing on the template EL image need to be marked;

s2, adding the marked template EL image into AI software for model learning, forming a standard model of defect characteristics through the model learning, and memorizing;

Further, there are several defect features in S1, and the marking of the several defect features is performed independently.

In the invention, the marks of a plurality of defect characteristics are independent, namely, when a certain defect characteristic is marked, only the defect characteristic is marked, and other defect characteristics are not marked, so that when the defect characteristic is added to the AI software, only a single pair of defect characteristics is needed to be marked, and a large amount of workload is reduced.

Furthermore, the number of times of model learning is set to be a plurality of batches, the same defect characteristics are added to the model learning in the same batch, and the model learning in different batches is independently carried out.

In the traditional model learning process, the learning of different defect characteristics of AI software is synchronous, although the method enables AI to memorize a large amount of defect characteristics more quickly, different defect characteristics have influence and easily influence the training effect of the AI software.

Furthermore, the model learning times of each batch are set by self.

For a certain defect feature, the higher the learning frequency, the better the learning effect, so for the defect feature with high similarity, the more learning frequency can be correspondingly set.

Furthermore, the recognition and prediction are carried out for n times, wherein n is the number of defect feature categories after the AI software learns and memorizes.

Further, the n recognition predictions are independent of each other.

In the traditional mode, the identification and prediction are only carried out once, the AI software probably determines the defect characteristics with high similarity as one of the defect characteristics, and the reference is made to fig. 4 and fig. 5, so that the defect with low tolerance is determined to be missed, and in the production and use process, the missed determination can cause partial serious defects to be hidden, so that firstly, the produced finished product has defects, the yield of subsequent products is influenced, secondly, an enterprise can not find process defects in time, raw materials are wasted, and the harmfulness is extremely high.

Further, m groups are generated in the information table, wherein m is the number of defect feature categories on the EL image to be detected, which are predicted and identified by the AI software.

The AI software identifies and predicts one defect characteristic, and then generates an information table correspondingly, and a user can view the information table of each defect characteristic on the use interface.

Further, the information table includes the area, the occurrence number, the confidence level and the gray level of the defect feature.

The information such as the area, the occurrence frequency and the like of the defect features can indicate the severity of the defect features, and the method has practical guiding significance.

Referring to fig. 2, a system for a serial analysis application method for AI recognition of a solar cell EL image includes a model learning module, a model memorizing module, a recognition predicting module, and a result outputting module.

The model learning module is used for training and learning the defect characteristics of the template, forming a standard model through template defect characteristic memory and transmitting the standard model to the model memory module;

the model memory module is used for storing the standard model;

the identification and prediction module is used for comparing the standard model with the EL image to be detected, marking defect characteristics on the EL image to be detected and finally transmitting the marked EL image to the result output module;

Furthermore, a model adding module is arranged in the model learning module.

Through the model adding module, new defect characteristics can be conveniently added to the AI software, and the use is more convenient.

The working principle is as follows: in the invention, the marks of a plurality of defect characteristics are independent, namely when a certain defect characteristic is marked, only the defect characteristic is marked, and other defect characteristics are not marked, when the AI software is learned, the marking is carried out in multiple batches, the training learning of only one defect characteristic is carried out in the same batch, different defect characteristics are carried out mutually, and mutual interference cannot be formed, so that the memory effect of the AI software is better, and the learning effect of the AI software is improved; when the AI software is actually used, the prediction and identification of different defect characteristics are independently carried out, and the AI software memorizes the defect characteristics for a plurality of times to carry out the recognition and prediction, so that when the defect characteristics with extremely high similarity are recognized and predicted, only two groups of defect characteristics are judged to exist, the defect characteristics with low tolerance are not missed, and the recognition and prediction effect of the AI software is improved.

Finally, it should be noted that: although the present invention has been described in detail with reference to the foregoing embodiments, it will be apparent to those skilled in the art that modifications may be made to the embodiments or portions thereof without departing from the spirit and scope of the invention.

Claims

1. A serial analysis application method for AI identification of solar cell EL images is characterized in that: the method comprises the following steps:

2. The method for serially analyzing and applying AI recognition to an EL image of a solar cell according to claim 1, wherein: there are several defect features in the S1, and the marking of the several defect features is performed independently.

3. The method for serially analyzing and applying AI recognition to an EL image of a solar cell according to claim 1, wherein: the number of times of model learning is set to be a plurality of batches, the same defect feature is added to the model learning in the same batch, and the model learning in different batches is independently carried out.

4. The method for serially analyzing and applying AI identification to EL images of solar cells as claimed in claim 3, wherein: the model learning times of each batch are set by self.

5. The method for serially analyzing and applying AI recognition to an EL image of a solar cell according to claim 1, wherein: and the recognition prediction is carried out for n times, wherein n is the number of the defect characteristic categories after the AI software learns and memorizes.

6. The method for serially analyzing and applying AI identification to an EL image of a solar cell as claimed in claim 5, wherein: the n times of the identification prediction are independent.

7. The method for serially analyzing and applying AI recognition to an EL image of a solar cell according to claim 1, wherein: and m groups are generated in the information table, wherein m is the number of the defect characteristic categories on the EL image to be detected, which is predicted and identified by AI software.

8. The method for serially analyzing and applying AI recognition to an EL image of a solar cell according to claim 1, wherein: the information table includes the area, the number of occurrences, the confidence level and the gray level of the defect feature.

9. A system of a serial analysis application method for AI identification of solar cell EL images is characterized in that: the system comprises a model learning module, a model memory module, an identification prediction module and a result output module;

the model memory module is used for storing the standard model;

10. The method for serially analyzing and applying AI recognition to an EL image of a solar cell according to claim 9, wherein: and a model adding module is arranged in the model learning module.