CN114842275B - Circuit board defect judging method, training method, device, equipment and storage medium - Google Patents

Abstract

The application discloses a circuit board defect judging method, a training method, a device, equipment and a storage medium, and relates to the field of surface defect judgment; the circuit board defect judging method comprises the following steps: acquiring a target image of a circuit board to be distinguished; inputting the target image into a trained target discrimination model to obtain defect discrimination information of the target image; the target discrimination model is obtained by training a circuit board sample image set; the circuit board sample image set comprises a plurality of sample images and label information corresponding to the sample images, the label information is obtained based on semantic description information of the sample images, and the label information comprises: defect type information and defect level information. The defect judging method aims to solve the technical problem that the defect grade of the defects cannot be effectively judged by the existing defect judging method.

Description

Circuit board defect discrimination method, training method, device, equipment and storage medium

technical field

The present application relates to the field of surface defect discrimination, in particular to a circuit board defect discrimination method, training method, device, equipment and storage medium.

Background technique

The circuit board is an important electronic component, a support for electronic components, and a carrier for electrical connections of electronic components. Whether the circuit board itself has defects will directly affect the performance of the equipment using the circuit board, so it is particularly necessary to identify the defects of the circuit board. There are many types of circuit board defects, including more than a dozen defects such as exposed copper, board contamination, line scratches, thin solder mask, offset solder mask, and blurred text. Different defects have different shapes, sizes, colors, and positions. However, the existing defect identification methods cannot effectively determine the defect level of defects.

Contents of the invention

The main purpose of this application is to provide a circuit board defect discrimination method, training method, device, equipment and storage medium, aiming to solve the technical problem that the existing defect discrimination method cannot effectively determine the defect level of the defect.

In order to solve the above-mentioned technical problems, the present application proposes: a circuit board defect discrimination method, comprising the following steps:

Obtain the target image of the circuit board to be discriminated;

inputting the target image into a trained target discrimination model to obtain defect discrimination information of the target image;

Wherein, the target discrimination model is obtained by training a circuit board sample image set; the circuit board sample image set includes several sample images and label information corresponding to several sample images, and the label information is based on several sample images Semantic description information is obtained, and the label information includes: defect type information and defect level information.

In order to enable the target discriminant model to effectively and automatically identify circuit board defects in subsequent practical applications, as some optional implementation modes of the present application, the target image is input into the trained target discriminant model to obtain the target image Before the defect discrimination information, it also includes:

Obtain several sample images;

Semantically annotating the defect information of several sample images respectively to obtain the circuit board sample image set;

Based on the circuit board sample image set, an initial target discrimination model is trained to obtain the target discrimination model.

In order to improve the semantic annotation content and the actual information of the defect closer, and extract the general semantic annotation content from several sample images, thereby improving the efficiency of the discrimination level; as some optional implementation methods of this application, the described The defect information of the sample image is semantically annotated to obtain the circuit board sample image set, including:

Based on historical data, perform semantic discrimination on several sample images and corresponding semantic interpretations, and obtain semantic discrimination results;

Obtaining a target semantic interpretation corresponding to the semantic discrimination result based on the semantic discrimination result;

Based on the target semantic interpretation, the defect information of several sample images is respectively semantically annotated to obtain the circuit board sample image set.

In this application, aiming at the actual application scenario of circuit board defects, based on the historical data of the circuit board, semantic discrimination is performed on several sample images and corresponding semantic interpretations, and the target semantic interpretation is obtained based on the semantic discrimination results. And based on the semantic interpretation of the target, general defect type information is obtained, that is, as some optional implementation manners of the present application, the defect type information includes at least one of remaining copper, broken disk and pit.

In order to enable the target recognition model to automatically discriminate target defects, this application assigns different target semantic interpretations to the above defect types, that is, assigns quantitative standard rules to the target recognition model; as some optional implementation modes of this application, When the defect type information in the sample image is remaining copper information, the target semantic interpretation is the interpretation of line gap width.

In order to enable the target recognition model to automatically discriminate target defects, this application assigns different target semantic interpretations to the above defect types, that is, assigns quantitative standard rules to the target recognition model; as some optional implementation modes of this application, When the defect type information in the sample image is broken disk information, the target semantic interpretation is line width interpretation.

In order to enable the target recognition model to automatically discriminate target defects, this application assigns different target semantic interpretations to the above defect types, that is, assigns quantitative standard rules to the target recognition model; as some optional implementation modes of this application, When the defect type information in the sample image is pit information, the target semantic interpretation is an interpretation of line density or line width.

In order to improve the semantic annotation content and the actual information of the defect closer, and extract the general semantic annotation content from several sample images, thereby improving the efficiency of discrimination levels; as some optional implementation methods of this application, the label information also includes : Defect location information.

In order to enable the target discrimination model to have the ability to automatically identify defect levels, this application includes defect type information and defect location information when performing semantic labeling, and as some optional implementation modes of this application, the defect Level information includes: acceptance information and scrap information.

In order to improve the efficiency and accuracy of automatically discriminating defect levels by the target discrimination model, as some optional implementations of the present application, the target image is input into the trained target discrimination model to obtain defect discrimination information of the target image ,include:

inputting the target image into a trained target discrimination model to obtain defect information of the target image;

Based on the defect information of the target image, it is compared with a preset discrimination standard to obtain discrimination information of the defect.

In order to solve the above-mentioned technical problems, the present application also proposes: a circuit board defect discrimination model training method, comprising the following steps:

Obtain several sample images;

Semantically annotating defect information of several sample images respectively to obtain the circuit board sample image set; wherein, the circuit board sample image set includes several sample images and label information corresponding to several sample images, The label information includes: defect type information and defect level information;

Based on the circuit board sample image set, an initial target discrimination model is trained to obtain the target discrimination model.

The circuit board defect discrimination model obtained by the above training method is learned by rules, that is, the defect information of several sample images is semantically labeled to obtain the circuit board sample image set; wherein, the circuit board sample image set Including several sample images and label information corresponding to several sample images, the label information includes: defect type information and defect level information; based on the circuit board sample image set, the initial target discrimination model is trained to obtain the Therefore, in practical applications, it can automatically perform defect level discrimination on the circuit board defects.

In order to improve the semantic annotation content and the actual information of the defect closer, and extract the general semantic annotation content from several sample images, thereby improving the efficiency of the discrimination level; as some optional implementation methods of this application, the described The defect information of the sample image is semantically annotated to obtain the circuit board sample image set, including:

Based on historical data, perform semantic discrimination on several sample images and corresponding semantic interpretations, and obtain semantic discrimination results;

Obtaining a target semantic interpretation corresponding to the semantic discrimination result based on the semantic discrimination result;

Based on the target semantic interpretation, the defect information of several sample images is respectively semantically annotated to obtain the circuit board sample image set.

As some optional implementations of the present application, the circuit board surface defect discrimination model is a Faster RCNN algorithm model. In this embodiment, the performance of the Faster RCNN algorithm model is superior, and a high-precision object discrimination performance is realized. Faster RCNN uses a two-stage network and RPN to achieve discrimination. Compared with other first-order networks, the two-stage network is more accurate, especially for high-precision, multi-scale, and small object problems. The advantages of the two-stage network are more obvious. Faster RCNN works well on multiple data sets and object tasks. For personal data sets, it can often achieve better results after Fine-tune. There are many points that can be optimized in the entire algorithm framework of Faster RCNN, providing a broad space for algorithm optimization. Therefore, adopting the Faster RCNN algorithm model as the circuit board surface defect discrimination model can improve the discrimination speed and accuracy of the discrimination model for target defects.

In order to solve the above technical problems, this application also proposes: a circuit board defect discrimination device, comprising:

The first acquisition module is used to acquire the target image of the circuit board to be identified;

The second acquisition module is used to input the target image into the trained target discrimination model to obtain the defect discrimination information of the target image; wherein, the target discrimination model is obtained by training the circuit board sample image set; the circuit board The sample image set includes several sample images and label information corresponding to the sample images, the label information is obtained based on the semantic description information of the several sample images, and the label information includes: defect type information and defect level information.

In order to solve the above technical problems, this application also proposes: an electronic device, the electronic device includes a memory and a processor, the memory stores a computer program, and the processor executes the computer program to realize the above-mentioned method.

In order to solve the above technical problems, the present application also proposes: a computer-readable storage medium, on which a computer program is stored, and a processor executes the computer program to implement the above-mentioned method.

There are many types of circuit board defects, including more than a dozen defects such as exposed copper, board contamination, line scratches, thin solder mask, offset solder mask, and blurred text. Different defects have different shapes, sizes, colors, and positions. However, the existing defect identification methods cannot effectively determine the defect level of defects, and can only manually determine the defect level based on factors such as defect area or defect number. This manual method of determining defect level is relatively subjective due to the lack of quantitative standards. Large, and the efficiency of discriminating the defect level is low. Compared with the prior art, this application adopts circuit board sample image training to obtain a target discrimination model, and the circuit board sample image includes several sample images and label information corresponding to several sample images, and the label information is based on several The semantic description information of the sample image is obtained, and the label information includes: defect type information and defect level information; since different defect type label information and defect level label information are added to the sample image, so that the target discrimination model During the training process, different defects can be identified and the level of the defect can be automatically identified; therefore, when the level of defects of the circuit board to be identified is actually determined, the target image of the circuit board to be identified is input into the The automatic discrimination grade result information can be obtained in the target discrimination model, thereby improving the efficiency of the discrimination grade of the defect position of the circuit board.

Description of drawings

FIG. 1 is a schematic structural diagram of an electronic device in a hardware operating environment involved in an embodiment of the present application;

Fig. 2 is a schematic flow chart of the circuit board defect discrimination method described in the embodiment of the present application;

FIG. 3 is a schematic flow diagram of obtaining defect discrimination information according to an embodiment of the present application;

FIG. 4 is a schematic flow diagram of obtaining a target discrimination model according to an embodiment of the present application;

Fig. 5 is a schematic diagram of the discrimination result of residual copper as the circuit board defect described in the embodiment of the present application;

Fig. 6 is a schematic diagram of the discrimination result of the circuit board defect described in the embodiment of the present application as a broken disk;

Fig. 7 is a schematic diagram of the discriminant result of the circuit board defect described in the embodiment of the present application as a pit;

8 is a schematic flow diagram of the circuit board defect discrimination model training method described in the embodiment of the present application;

9 is a schematic flow diagram of obtaining a circuit board sample image set according to an embodiment of the present application;

FIG. 10 is a schematic diagram of functional modules of the circuit board defect discrimination device according to the embodiment of the present application.

Detailed ways

It should be understood that the specific embodiments described here are only used to explain the present application, and are not intended to limit the present application.

In the prior art, automatic optical inspection (Automated Optical Inspection, referred to as AOI) machine is usually used to identify circuit board defects. AOI detects common defects encountered in welding production based on optical principles. In the production process, AOI uses high-speed and high-precision visual processing technology to automatically detect various mounting errors and welding defects on the PCB. The range of PCB boards can range from fine-pitch high-density boards to low-density large-size boards, and online inspection solutions can be provided to improve production efficiency and welding quality. Find and eliminate errors early in the assembly process for good process control by using AOI as a defect reduction tool. Early detection of defects will avoid sending defective boards to subsequent assembly stages, and AOI will reduce repair costs and will avoid scrapping non-repairable boards.

When using AOI automatic detection, the machine automatically scans the PCB circuit board through the camera, collects images, compares the tested solder joints with the qualified parameters in the database, and checks out the defects on the PCB circuit board through image processing, and passes the display or The automatic mark displays/marks the defects, and the inspectors visually locate the defects and classify them. However, due to the instability of manual visual positioning of defects, it is difficult to unify the inspection standards of different inspectors, and the inspectors have been working for a long time. The fatigue state of work may also lead to inaccurate visual classification results, which eventually lead to detection results that do not meet the requirements. And the circuit board is an important electronic component, a support for electronic components, and a carrier for electrical connections of electronic components. Whether there is a defect in the circuit board itself will directly affect the performance of the equipment using the circuit board, so it is particularly necessary to identify the defect of the circuit board. There are many types of circuit board defects, including more than a dozen defects such as exposed copper, board contamination, line scratches, thin solder mask, offset solder mask, and blurred text. Different defects have different shapes, sizes, colors, and positions. However, the existing defect identification methods cannot effectively determine the defect level of defects.

Referring to FIG. 1 , FIG. 1 is a schematic structural diagram of an electronic device of a hardware operating environment involved in an embodiment of the present application.

As shown in FIG. 1 , the electronic device may include: a processor 1001, such as a central processing unit (Central Processing Unit, CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Wherein, the communication bus 1002 is used to realize connection and communication between these components. The user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a wireless fidelity (WIreless-FIdelity, WI-FI) interface). The memory 1005 may be a high-speed random access memory (Random Access Memory, RAM) memory, or a stable non-volatile memory (Non-Volatile Memory, NVM), such as a disk memory. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001 .

Those skilled in the art can understand that the structure shown in FIG. 1 does not constitute a limitation on the electronic device, and may include more or less components than shown in the figure, or combine some components, or arrange different components.

As shown in FIG. 1 , memory 1005 as a storage medium may include an operating system, a data storage module, a network communication module, a user interface module, and an electronic program.

In the electronic device shown in Figure 1, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the electronic device of the present application can be set in In the electronic device, the electronic device invokes the circuit board defect identification device stored in the memory 1005 through the processor 1001, and executes the circuit board defect identification method provided in the embodiment of the present application.

As shown in Figure 2, the embodiment of the present application provides: a circuit board defect discrimination method, comprising the following steps:

S10. Acquiring the target image of the circuit board to be identified;

In a specific application, the circuit board to be judged refers to the circuit board that needs to be judged whether it has defects, and the target image refers to the image of the circuit board to be judged, which can be obtained by manual photography or AOI. In practical applications, the circuit board to be judged The target image of the board can also be obtained after being processed by a computer vision algorithm. The computer vision algorithm uses cameras and computers instead of human eyes to perform machine vision such as recognition, tracking and measurement of targets, and further performs graphics processing to make computer processing images that are more suitable for human eyes to observe or to be sent to instruments for identification.

S20. Input the target image into the trained target discrimination model to obtain defect discrimination information of the target image; wherein, the target discrimination model is obtained by training the circuit board sample image set; the circuit board sample image set includes several A sample image and tag information corresponding to several sample images, the tag information is obtained based on the semantic description information of several sample images, and the tag information includes: defect type information and defect level information. In a specific application, the sample image is also scanned by an AOI machine and is specially used for training the initial model, and the sample image may contain many common defects.

In order to improve the efficiency and accuracy of the automatic discrimination of the defect level by the target discrimination model, as shown in FIG. Defect discrimination information of the target image, including:

S201. Input the target image into a trained target discrimination model to obtain defect information of the target image.

Wherein, the target discrimination model is obtained by training a circuit board sample image set; the circuit board sample image set includes several sample images and label information corresponding to several sample images, and the label information is based on several sample images Semantic description information is obtained, and the label information includes: defect type information and defect level information. In a specific application, the sample image is also scanned by an AOI machine and is specially used for training the initial model, and the sample image may contain many common defects. Wherein, the defect information of the target image includes: defect type information, defect position information and defect level information.

S202. Based on the defect information of the target image, compare it with a preset discrimination standard to obtain discrimination information of the defect. The preset discrimination standard refers to the discrimination standard set separately for several common defects of the circuit board, such as: when the defect type information in the sample image is the remaining copper information, if the remaining copper causes the remaining gap of the circuit If the width is less than 1/2, it will be scrapped, and if the remaining gap width exceeds 1/2, it will be accepted. When the defect type information in the sample image is broken disk information, if the broken disk causes the drilling position on the circuit board to exceed the copper surface area, then it will be rejected, and if it does not exceed the copper surface area, then it will be accepted. When the defect type information in the sample image is pit information, if the pits cause a large number of aggregated pits in the copper surface area, or the pits in the line area cause the line width to be less than 1/2, then it will be scrapped. If the above conditions are not met Conditions are accepted.

In order to enable the target discriminant model to effectively and automatically identify circuit board defects in subsequent practical applications, as shown in Figure 4, as some optional implementation modes of the present application, inputting the target image into the trained target discriminant model , before obtaining the defect discrimination information of the target image, further includes:

S01. Acquire several sample images.

Wherein, the sample image is an image obtained by scanning with an AOI machine and specially used for training the initial model, and the sample image may contain various common defects.

S02. Semantically label defect information of several sample images respectively to obtain the circuit board sample image set.

In the above steps, the circuit board sample image set is mainly: based on historical data, carry out semantic discrimination on several sample images and corresponding semantic interpretations to obtain semantic discrimination results; based on the semantic discrimination results, obtain all The target semantic interpretation corresponding to the semantic discrimination result; based on the target semantic interpretation, semantically annotate the defect information of several sample images to obtain the circuit board sample image set. The above steps are mainly to improve the closeness between the semantic annotation content and the actual defect information, and to extract the general semantic annotation content from several sample images, thereby improving the efficiency of the discrimination level. Wherein, the circuit board sample image set includes several sample images and label information corresponding to the several sample images, and the label information includes: defect type information and defect level information. In a specific application, the defect information is marked on the sample image manually or by a machine.

S03. Based on the circuit board sample image set, train an initial target discrimination model to obtain the target discrimination model.

The initial target discrimination model refers to the Faster RCNN algorithm model. In this embodiment, the performance of the FasterRCNN algorithm model is superior, and a high-precision object discrimination performance is realized. Faster RCNN uses a two-stage network and RPN to achieve discrimination. Compared with other first-order networks, the two-stage network is more accurate, especially for high-precision, multi-scale, and small object problems. The advantages of the two-stage network are more obvious. Faster RCNN works well on multiple data sets and object tasks. For personal data sets, it can often achieve better results after Fine-tune. There are many points that can be optimized in the entire algorithm framework of Faster RCNN, providing a broad space for algorithm optimization. Therefore, adopting the Faster RCNN algorithm model as the circuit board surface defect discrimination model can improve the discrimination speed and accuracy of the discrimination model for target defects.

In this application, aiming at the actual application scenario of circuit board defects, based on the historical data of the circuit board, semantic discrimination is performed on several sample images and corresponding semantic interpretations, and the target semantic interpretation is obtained based on the semantic discrimination results. And based on the semantic interpretation of the target, general defect type information is obtained, that is, as some optional implementation manners of the present application, the defect type information includes at least one of remaining copper, broken disk and pit. In order to improve the semantic annotation content and the actual information of the defect closer, and extract the general semantic annotation content from several sample images, thereby improving the efficiency of discrimination levels; as some optional implementation methods of this application, the label information also includes : Defect location information. In order to enable the target discrimination model to have the ability to automatically identify defect levels, this application includes defect type information and defect location information when performing semantic labeling, and as some optional implementation modes of this application, the defect Level information includes: acceptance information and scrap information.

In order to enable the target recognition model to automatically identify target defects, this application has given different target semantic interpretations for the above-mentioned defect types, that is, given the target recognition model quantitative standard rules, and sorted out the relationship between defect level judgment and background semantics. The business logic relationship; such as:

1) As shown in Figure 5, when the defect type information in the sample image is residual copper information, the target semantic interpretation is the interpretation of the line gap width, that is, the residual copper causes the residual gap width of the line to be less than 1/2, then Scrap, if the remaining gap width exceeds 1/2, it will be accepted. Figure 5 is an image of the identification result of a circuit board whose defect is residual copper. In the figure, the black area is the base material area, and the yellow area is the copper surface area. Since the remaining copper needs to be clamped to control the width of the gap, compared with the blue frame of the existing identification results Area, the red frame area of the discrimination results of this application highlights the comparison between the defect and the gap width, and the label corresponding to this figure is Cu-NG.

2) As shown in Figure 6, when the defect type information in the sample image is broken disk information, the target semantic interpretation is the interpretation of line width, that is, the broken disk causes the drilling position on the circuit board to exceed the copper surface area, If it does not exceed the area of the copper surface, it will be accepted. Figure 6 is an image of the identification result of a circuit board whose defect is a broken disk. Since the broken disk needs to be clamped to control whether the drilling position exceeds the pad area, the red box marked in this application is more important than the traditionally identified green box marked area. Highlight the edge features of the hole, and the corresponding label in this figure is Broken Disk-OK.

3) As shown in Figure 7, when the defect type information in the sample image is pit information, the target semantic interpretation is the interpretation of line density or line width, that is, the pits cause a large amount of aggregation in the copper surface area If the pits, or pits in the line area cause the width of the line to be less than 1/2, it will be scrapped, and if the above conditions are not met, it will be accepted. Figure 7 is an image of the discrimination result of a circuit board whose defects are pits. Since the pits need to be marked for aggregation, each independent pit is marked in the discrimination result diagram described in this application, and the line width and The aggregation was judged. Compared with the red frame, the green frame shown in this application highlights the actual defect area/marked frame area, which improves the higher signal-to-noise ratio, and the samples with higher signal-to-noise ratio can It is more conducive to the training convergence of the model.

There are many types of circuit board defects, including more than a dozen defects such as exposed copper, board contamination, line scratches, thin solder mask, offset solder mask, and blurred text. Different defects have different shapes, sizes, colors, and positions. However, the existing defect identification methods cannot effectively determine the defect level of defects, and can only manually determine the defect level based on factors such as defect area or defect number. This manual method of determining defect level is relatively subjective due to the lack of quantitative standards. Large, and the efficiency of discriminating the defect level is low. Compared with the prior art, this application adopts circuit board sample image training to obtain a target discrimination model, and the circuit board sample image includes several sample images and label information corresponding to several sample images, and the label information is based on several The semantic description information of the sample image is obtained, and the label information includes: defect type information and defect level information; since different defect type label information and defect level label information are added to the sample image, so that the target discrimination model During the training process, different defects can be identified and the level of the defect can be automatically identified; therefore, when the level of defects of the circuit board to be identified is actually determined, the target image of the circuit board to be identified is input into the The automatic discrimination grade result information can be obtained in the target discrimination model, thereby improving the efficiency of the discrimination grade of the defect position of the circuit board.

In order to solve the above technical problems, this application also proposes: as shown in Figure 8, a circuit board defect discrimination model training method, comprising the following steps:

SS1. Acquiring several sample images.

In a specific application, the sample image is also scanned by an AOI machine and is specially used for training the initial model, and the sample image may contain many common defects.

SS2. Semantically label defect information of several sample images respectively, so as to obtain the circuit board sample image set.

In the above steps, the circuit board sample image set is mainly: based on historical data, carry out semantic discrimination on several sample images and corresponding semantic interpretations to obtain semantic discrimination results; based on the semantic discrimination results, obtain all The target semantic interpretation corresponding to the semantic discrimination result; based on the target semantic interpretation, semantically annotate the defect information of several sample images to obtain the circuit board sample image set. The above steps are mainly to improve the closeness between the semantic annotation content and the actual defect information, and to extract the general semantic annotation content from several sample images, thereby improving the efficiency of the discrimination level. Wherein, the circuit board sample image set includes several sample images and label information corresponding to the several sample images, and the label information includes: defect type information and defect level information. In a specific application, the defect information is marked on the sample image manually or by a machine.

In order to improve the semantic annotation content and the actual information of the defect closer, and extract the general semantic annotation content from several sample images, thereby improving the efficiency of the discrimination level; as shown in Figure 9, as some optional implementation methods of this application, The semantic labeling of defect information of several sample images respectively to obtain the circuit board sample image set includes:

SS21. Based on historical data, perform semantic discrimination on several sample images and corresponding semantic interpretations, and obtain semantic discrimination results.

Wherein, the historical data refers to the semantic discrimination result information data obtained based on the defect information in several sample images; the purpose of this step is to obtain the sample images and their corresponding semantic interpretation and The mapping relationship between the semantic discrimination results, so as to obtain a semantic interpretation with a higher discrimination rate, and mark the circuit board sample image set, so that in the subsequent training process of the target discrimination model, it is more conducive for the target discrimination model to learn the sample image And the mapping relationship between the corresponding semantic interpretation and the semantic discrimination result, so as to improve the discrimination rate of the target discrimination model for circuit board defects in practical applications.

SS22. Based on the semantic discrimination result, obtain a target semantic interpretation corresponding to the semantic discrimination result.

As mentioned above, based on the correct semantic discrimination result, the target semantic interpretation corresponding to the correct semantic discrimination result is obtained, that is, the semantic interpretation with a higher recognition rate is obtained as the target semantic interpretation, and the circuit board sample image set is annotated for subsequent training. In the target discrimination model process, it is more beneficial for the target discrimination model to learn the mapping relationship between the sample image and its corresponding semantic interpretation and the semantic discrimination result, thereby improving the effectiveness of the target discrimination model for circuit board defects in practical applications. Higher discrimination rate.

SS23. Based on the target semantic interpretation, perform semantic annotation on defect information of several sample images respectively, so as to obtain the circuit board sample image set.

As mentioned above, based on the target semantic interpretation, that is, the semantic interpretation with a higher recognition rate is used as the target semantic interpretation, and the circuit board sample image set is marked to obtain the circuit board sample image set; in order to train the target discriminant model in the subsequent process , which is more conducive to the target discrimination model learning the mapping relationship between the sample image and its corresponding semantic interpretation and semantic discrimination results, thereby improving the target discrimination model to have a higher discrimination of circuit board defects in practical applications Rate.

SS3. Based on the circuit board sample image set, train an initial target discrimination model to obtain the target discrimination model.

In a specific application, after obtaining the marked sample image set, it is necessary to learn the marked position, category and level through the initial model to obtain the trained discriminant model. The discriminant models used include RCNN algorithm model, FastRCNN algorithm model and One of the Faster RCNN algorithm model algorithms.

As some optional implementations of the present application, the circuit board surface defect discrimination model is a Faster RCNN algorithm model. In this embodiment, the performance of the Faster RCNN algorithm model is superior, and a high-precision object discrimination performance is realized. Faster RCNN uses a two-stage network and RPN to achieve discrimination. Compared with other first-order networks, the two-stage network is more accurate, especially for high-precision, multi-scale, and small object problems. The advantages of the two-stage network are more obvious. Faster RCNN works well on multiple data sets and object tasks. For personal data sets, it can often achieve better results after Fine-tune. There are many points that can be optimized in the entire algorithm framework of Faster RCNN, providing a broad space for algorithm optimization. Therefore, adopting the Faster RCNN algorithm model as the circuit board surface defect discrimination model can improve the discrimination speed and accuracy of the discrimination model for target defects.

The circuit board defect discrimination model obtained by the above training method is learned by rules, that is, the defect information of several sample images is semantically labeled to obtain the circuit board sample image set; wherein, the circuit board sample image set Including several sample images and label information corresponding to several sample images, the label information includes: defect type information and defect level information; based on the circuit board sample image set, the initial target discrimination model is trained to obtain the Therefore, in practical applications, it can automatically perform defect level discrimination on the circuit board defects.

Based on the same inventive concept, as shown in Figure 10, the present application also provides: a circuit board defect discrimination device, comprising:

The first acquisition module is used to acquire the target image of the circuit board to be identified;

The second acquisition module is used to input the target image into the trained target discrimination model to obtain the defect discrimination information of the target image; wherein, the target discrimination model is obtained by training the circuit board sample image set; the circuit board The sample image set includes several sample images and label information corresponding to the sample images, the label information is obtained based on the semantic description information of the several sample images, and the label information includes: defect type information and defect level information.

It should be noted that each module in the circuit board defect identification device in this embodiment corresponds to each step in the circuit board defect identification method in the foregoing embodiment, therefore, the specific implementation of this embodiment can refer to the foregoing circuit The implementation of the board defect discrimination method will not be repeated here.

In addition, in an embodiment, the embodiments of the present application also provide a computer-readable storage medium, on which a computer program is stored, and a processor executes the computer program to implement the above-mentioned Methods.

In some embodiments, the computer-readable storage medium can be memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface memory, optical disk, or CD-ROM; Various equipment. Computers can be various computing devices including smart terminals and servers.

In some embodiments, executable instructions may take the form of programs, software, software modules, scripts, or code written in any form of programming language, including compiled or interpreted languages, or declarative or procedural languages, and its Can be deployed in any form, including as a stand-alone program or as a module, component, subroutine or other unit suitable for use in a computing environment.

As an example, executable instructions may, but do not necessarily correspond to files in a file system, may be stored as part of files that hold other programs or data, for example, in a Hyper Text Markup Language (HTML) document in one or more scripts of the program in question, in a single file dedicated to the program in question, or in multiple cooperating files (for example, files that store one or more modules, subroutines, or sections of code).

As an example, executable instructions may be deployed to be executed on one computing device, or on multiple computing devices located at one site, or alternatively, on multiple computing devices distributed across multiple sites and interconnected by a communication network. to execute.

It should be noted that, in this document, the terms "comprising", "comprising" or any other variant are intended to cover a non-exclusive inclusion such that a process, method, article or system comprising a set of elements includes not only those elements, It also includes other elements not expressly listed, or elements inherent in the process, method, article, or system. Without further limitations, an element defined by the phrase "comprising a..." does not preclude the presence of additional identical elements in the process, method, article or system comprising that element.

The serial numbers of the above embodiments of the present application are for description only, and do not represent the advantages and disadvantages of the embodiments.

Through the description of the above embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus a necessary general-purpose hardware platform, and of course also by hardware, but in many cases the former is better implementation. Based on this understanding, the essence of the technical solution of this application or the part that contributes to the prior art can be embodied in the form of software products, and the computer software products are stored in a storage medium (such as read-only memory/random access Memory, magnetic disk, optical disk), including several instructions to enable a multimedia terminal device (which may be a mobile phone, computer, television receiver, or network device, etc.) to execute the methods described in various embodiments of the present application.

The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structures or equivalent process transformations made by using the description of the application and the accompanying drawings are directly or indirectly used in other related technical fields. , are all included in the patent protection scope of the present application in the same way.

Claims (8)

1. A circuit board defect discrimination method is characterized in that, comprising the following steps: Obtain several sample images; Perform semantic annotation on defect information of several sample images respectively to obtain a circuit board sample image set, wherein the circuit board sample image set includes several sample images and label information corresponding to several sample images, and the label information Obtained based on the semantic description information of several sample images, the label information includes: defect type information and defect level information, and the defect level information includes: acceptance information and scrap information; The step of semantically annotating the defect information of several sample images to obtain a circuit board sample image set includes: performing semantic discrimination on several sample images and their corresponding semantic interpretations based on historical data, and obtaining Semantic discrimination results; based on the semantic discrimination results, obtaining target semantic interpretations corresponding to the semantic discrimination results; based on the target semantic interpretations, semantically annotating defect information of several sample images to obtain the circuit board A sample image set; wherein, based on the semantic discrimination result, the step of obtaining the target semantic interpretation corresponding to the semantic discrimination result includes: the defect type information includes at least one of residual copper, broken disk and pit , when the defect type information in the sample image is residual copper information, the semantic interpretation of the target is the interpretation of the line gap width, and the remaining copper causes the remaining gap width of the line to be less than 1/2, then it will be scrapped; if the remaining gap width exceeds 1 /2, it is accepted; when the defect type information in the sample image is broken information, the target semantic interpretation is the interpretation of the line width, that is, the broken plate causes the drilling position on the circuit board to exceed the copper surface area, then Scrap, if it does not exceed the copper surface area, it is acceptable; when the defect type information in the sample image is pit information, the target semantic interpretation is the interpretation of line density or the interpretation of line width, that is, pits cause copper If there are a large number of accumulated pits in the surface area, or the pits in the line area cause the line width to be less than 1/2, it will be scrapped, and if the above conditions are not met, it will be accepted; Based on the circuit board sample image set, an initial target discrimination model is trained to obtain the target discrimination model; Obtain the target image of the circuit board to be discriminated; Inputting the target image into a trained target discrimination model to obtain defect discrimination information of the target image; wherein, the target discrimination model is obtained by training a circuit board sample image set. 2. The circuit board defect discrimination method according to claim 1, wherein the label information further comprises: defect location information. 3. The circuit board defect discrimination method according to claim 1, wherein said inputting said target image into a trained target discrimination model to obtain defect discrimination information of said target image comprises: inputting the target image into a trained target discrimination model to obtain defect information of the target image; Based on the defect information of the target image, it is compared with a preset discrimination standard to obtain discrimination information of the defect. 4. A circuit board defect discrimination model training method, is characterized in that, comprises the following steps: Obtain several sample images; Semantically annotating defect information of several sample images to obtain a circuit board sample image set; wherein, semantically annotating defect information of several sample images respectively to obtain the circuit board sample image set, Including: based on historical data, performing semantic discrimination on several sample images and corresponding semantic interpretations to obtain semantic discrimination results; based on the semantic discrimination results, obtaining target semantic interpretations corresponding to the semantic discrimination results; based on the Target semantic interpretation, respectively semantically annotating the defect information of several sample images to obtain the circuit board sample image set; the circuit board sample image set includes several sample images and labels corresponding to several sample images Information, the label information includes: defect type information and defect level information, the defect level information includes: acceptance information and scrap information; wherein, based on the semantic discrimination result, the target corresponding to the semantic discrimination result is obtained The step of semantic interpretation includes: the defect type information includes at least one of residual copper, broken plate and pit, and when the defect type information in the sample image is residual copper information, the target semantic interpretation is line Interpretation of the gap width. If the remaining copper causes the remaining gap width of the line to be less than 1/2, it will be scrapped. If the remaining gap width exceeds 1/2, it will be accepted; when the defect type information in the sample image is broken information, the The semantic interpretation of the above target is the interpretation of the line width, that is, the position of the drilled hole on the circuit board exceeds the copper surface area due to a broken disk, then it will be scrapped, and if it does not exceed the copper surface area, it will be accepted; when the defect type information in the sample image is For pit information, the semantic interpretation of the target is the interpretation of the line density or the line width, that is, the pits cause a large number of aggregated pits in the copper surface area, or the pits in the line area cause the line width to be less than 1/2, then it is scrapped , if the above conditions are not met, it will be accepted; Based on the circuit board sample image set, an initial target discrimination model is trained to obtain the target discrimination model. 5. circuit board defect discrimination model training method according to claim 4, is characterized in that, described target discrimination model is Faster RCNN algorithm model. 6. A circuit board defect discrimination device, characterized in that, comprising: The sample set acquisition module is used to acquire several sample images; semantically annotate the defect information of several sample images respectively to obtain a circuit board sample image set, wherein the circuit board sample image set includes several sample images and several The label information corresponding to the sample image, the label information is obtained based on the semantic description information of several sample images, the label information includes: defect type information and defect level information, and the defect level information includes: acceptance information and Scrapping information; the step of semantically annotating the defect information of several sample images respectively to obtain a circuit board sample image set includes: performing semantic annotation on several sample images and corresponding semantic interpretations based on historical data Discriminate to obtain the semantic discrimination result; based on the semantic discrimination result, obtain the target semantic interpretation corresponding to the semantic discrimination result; based on the target semantic interpretation, perform semantic annotation on the defect information of several sample images respectively, so as to obtain the The circuit board sample image set; wherein, based on the semantic discrimination result, the step of obtaining the target semantic interpretation corresponding to the semantic discrimination result includes: the defect type information includes residual copper, broken discs and pits At least one, when the defect type information in the sample image is residual copper information, the target semantic interpretation is the interpretation of the line gap width, and the remaining copper causes the residual gap width of the line to be less than 1/2, then it will be scrapped, if the remaining gap If the width exceeds 1/2, it is acceptable; when the defect type information in the sample image is broken information, the target semantic interpretation is the interpretation of the line width, that is, the position of the drill hole on the circuit board exceeds the copper surface caused by the broken plate area, it will be rejected, if it does not exceed the copper surface area, it will be accepted; when the defect type information in the sample image is pit information, the target semantic interpretation is the interpretation of the line density or the line width, that is, the pit If the pits cause a large number of accumulated pits in the copper surface area, or if the pits in the line area cause the line width to be less than 1/2, it will be scrapped. If the above conditions are not met, it will be accepted; A model training module, configured to train an initial target discrimination model based on the circuit board sample image set to obtain the target discrimination model; The first acquisition module is used to acquire the target image of the circuit board to be identified; The second acquisition module is configured to input the target image into a trained target discrimination model to obtain defect discrimination information of the target image; wherein, the target discrimination model is obtained by training a circuit board sample image set. 7. An electronic device, characterized in that the electronic device comprises a memory and a processor, wherein a computer program is stored in the memory, and the processor executes the computer program to realize any one of claims 1-5 the method described. 8. A computer-readable storage medium, wherein a computer program is stored on the computer-readable storage medium, and a processor executes the computer program to implement the method according to any one of claims 1-5 .

Images