PCB Defect Detection Method, Device, Equipment, Medium

Abstract

The present invention discloses a PCB defect detection method, device, equipment, and medium that combines manual recognition with artificial intelligence. The detection method includes scanning a PCB circuit board to obtain a PCB image; using a detection model of an AOI device to perform AI recognition on the PCB image to obtain defect data including the corresponding positions of one or more recognized defects; intercepting a local image as a PCB sub-image according to the defect data and sending it to a manual inspection workstation; performing manual recognition on each PCB sub-image to obtain a first manual marking result; comparing the first manual marking result and the AI marking result of each PCB sub-image. If they are consistent, the first manual marking result or the AI marking result is used as the final detection result of the PCB sub-image, and the AI marking result is obtained by performing AI recognition on the PCB sub-image using an AI model. The present invention combines the defect results recognized by AI to judge whether the manual recognition is accurate, so as to ensure the accuracy rate of PCB defect detection and reduce or avoid quality problems caused by manual recognition errors.

Description

Technical Field

[0001] The present invention relates to the technical field of PCB defect detection, and particularly relates to a PCB defect detection method, device, equipment, and medium that combine manual identification with artificial intelligence. Background Art

[0002] During the process of PCB defect detection, an automatic optical inspection device (AOI device) is usually used to scan the circuit board to obtain defect information, and then this is used as the detection result and sent to the maintenance personnel to repair the circuit board. However, the recognition accuracy of the AOI device is limited. The detected defects and their types are directly sent to the maintenance personnel. Therefore, there are often cases where areas that are clearly defect-free are misjudged as having defects. Since the maintenance personnel only have basic skills for repairing circuit boards, such as soldering open circuits with missing solder or repairing short-circuit areas, once there is a misjudged defect, the maintenance personnel will be confused and at a loss, seriously affecting the defect detection and repair efficiency of the PCB.

[0003] If the manual method is used to identify PCB defects, there are always certain risks. For example, the inspector may mark a defect that should be classified as a broken wire in a duplicate category, or the inspector may not identify all the defects and miss the real defects and pass them on as non-defective states. This will directly result in defective products or components not being detected and thus being sent to the next production step or directly released as the final qualified products, unable to guarantee the production quality of the PCB.

[0004] The disclosure of the above background art content is only for assisting in understanding the inventive concept and technical solution of the present invention. It does not necessarily belong to the prior art of this patent application, nor will it necessarily provide technical guidance. Without clear evidence indicating that the above content was publicly available before the filing date of this patent application, the above background art should not be used to evaluate the novelty and inventiveness of this application. Summary of the Invention

[0005] The object of the present invention is to provide a PCB defect detection method, device, equipment, and medium that combine manual identification with artificial intelligence to ensure the accuracy rate of PCB defect detection.

[0006] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0007] A PCB defect detection method that combines manual identification with artificial intelligence, the detection method comprising:

[0008] Scanning the PCB circuit board to obtain a PCB image;

[0009] Use the detection model of the AOI device to identify the PCB image, and obtain defect data, where the defect data includes the corresponding positions of one or more identified defects;

[0010] According to the defect data, intercept the local image at the corresponding position as the PCB sub-image, and send it to the manual inspection workstation;

[0011] Perform manual identification on each PCB sub-image to obtain the first manual marking result;

[0012] Compare the first manual marking result and the AI marking result corresponding to each PCB sub-image:

[0013] If the first manual marking result and the AI marking result of the PCB sub-image are consistent, then use the first manual marking result or the AI marking result as the final inspection result of the PCB sub-image;

[0014] If the first manual marking result and the AI marking result of the PCB sub-image are inconsistent, then perform the manual marking on this PCB sub-image again to obtain the second manual marking result of this PCB sub-image;

[0015] If the second manual marking result is consistent with the AI marking result of this PCB sub-image, then use the second manual marking result or the AI marking result as the final inspection result of this PCB sub-image, and record the error rate of the manual marking;

[0016] Among them, the AI marking result is obtained by using the AI model to perform AI identification on the PCB sub-image.

[0017] Further, the detection method further includes:

[0018] If the first manual marking result and the AI marking result of the PCB sub-image are inconsistent, then perform the manual marking on this PCB sub-image again to obtain the second manual marking result of this PCB sub-image. If the second manual marking result is consistent with the AI marking result of this PCB sub-image, then use the second manual marking result or the AI marking result as the final inspection result of this PCB sub-image, and record the error rate of the manual marking.

[0019] Further, the detection method further includes:

[0020] If the second manual marking result is inconsistent with the AI marking result of the PCB sub-image, the PCB sub-image is manually reviewed to obtain the manual review result of the PCB sub-image. If the manual review result is consistent with the AI marking result of the PCB sub-image, the manual review result or the AI marking result is used as the final detection result of the PCB sub-image; wherein, the operator of the manual review is different from the operator of the manual marking.

[0021] Further, if the manual review result is inconsistent with the AI marking result of the PCB sub-image, the manual review result is used as the final detection result of the sub-PCB image, and the model used for AI marking is optimized. The steps of the optimization include:

[0022] S101. Save the PCB sub-images with inconsistent manual review results and AI marking results as training sample images;

[0023] S102. Set defect labels for the training sample images;

[0024] S103. Add the training sample images and their corresponding defect labels to a preset training sample set;

[0025] S104. Train the model used for AI marking based on the new training sample set, thereby updating the parameters of the model to obtain a better model.

[0026] Further, if the manual review result is inconsistent with the first manual marking result of the PCB sub-image, record the error rate of the manual marking.

[0027] Further, the detection method further includes:

[0028] If the second manual marking result is consistent with the first manual marking result of the PCB image, the model used for AI marking is optimized; the steps of the optimization include:

[0029] S201. Save the PCB sub-images with inconsistent second manual marking results and AI marking results as training sample images;

[0030] S202. Set defect labels for the training sample images;

[0031] S203. Add the training sample images and their corresponding defect labels to a preset training sample set;

[0032] S204. Train the model used for AI marking based on the new training sample set, thereby updating the parameters of the model to obtain a better model.

[0033] Further, the first manual marking result, the AI marking result, and the manual review result each include one or more of the defect location information, defect type information, and defect quantity information in the PCB sub-image.

[0034] A PCB defect detection device combining manual recognition and artificial intelligence, the detection device comprising:

[0035] A scanning unit configured to scan a PCB circuit board to obtain a PCB image;

[0036] An identification unit including a manual marking unit and an AI marking unit, the manual marking unit being configured to perform manual defect marking on the PCB sub-image to obtain a first manual marking result, and the AI marking unit being configured to perform AI defect marking on the PCB sub-image to obtain an AI marking result;

[0037] A processing unit configured to compare the first manual marking result and the AI marking result corresponding to each PCB sub-image. If the first manual marking result and the AI marking result of the PCB sub-image are consistent, the processing unit uses the first manual marking result or the AI marking result as the final detection result of the PCB sub-image; if the first manual marking result and the AI marking result of the PCB sub-image are inconsistent, the processing unit performs the manual marking on the PCB sub-image again to obtain a second manual marking result of the PCB sub-image; if the second manual marking result is consistent with the AI marking result of the PCB sub-image, the processing unit uses the second manual marking result or the AI marking result as the final detection result of the PCB sub-image, sends a reminder message to the corresponding staff, and records the error rate of the manual marking.

[0038] An electronic device, the electronic device comprising:

[0039] One or more processors;

[0040] A storage device having stored thereon one or more programs;

[0041] When the one or more programs are executed by the one or more processors, the one or more processors implement the PCB defect detection method as described above.

[0042] A computer storage medium storing a computer program programmed or configured to execute the PCB defect detection method as described above.

[0043] Advantages of the present invention: An AI model is used to identify defects in a PCB and the identification result is used as an auxiliary tool. This auxiliary tool is then used to determine whether the manual identification result is accurate. When the AI identification result is inconsistent with the manual identification result, the staff is reminded to conduct a re-inspection. This not only enables the timely discovery of misjudgments and missed judgments in the manual identification result, but also reduces or avoids PCB quality problems caused by misjudgments and missed judgments through re-inspection, thereby ensuring the accuracy rate of PCB defect detection to a certain extent. Brief Description of the Drawings

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0045] Figure 1 It is the first flowchart of the PCB defect detection method provided by an exemplary embodiment of the embodiment of the present invention;

[0046] Figure 2 It is the second flowchart of the PCB defect detection method provided by an exemplary embodiment of the embodiment of the present invention. Detailed Embodiments

[0047] In order to enable those skilled in the art to better understand the solution of the present invention, and to more clearly understand the purpose, technical solution and its advantages of the present invention, the following describes the technical solutions in the embodiments of the present invention clearly and completely in combination with specific embodiments and with reference to the drawings. It should be noted that the implementation manners not depicted or described in the drawings are forms known to those of ordinary skill in the art. Additionally, although this document may provide examples containing parameters with specific values, it should be understood that the parameters do not necessarily exactly equal the corresponding values, but may approximate the corresponding values within an acceptable error tolerance or design constraint. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of the present invention. In addition, the terms "including" and "having" in the description and claims of the present invention, and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, device, product or equipment that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or equipment.

[0048] In an embodiment of the present invention, a PCB defect detection method combining manual recognition and artificial intelligence is provided. As Figure 1 and 2 shown, this detection method includes the following steps:

[0049] In this embodiment, taking a PCB image with three defect quantities as an example, the defect detection process of this PCB image is specifically described:

[0050] First, scan the PCB circuit board to be detected to obtain the PCB image, that is, the complete image of the PCB, and use the detection model of the AOI device to identify the PCB image to obtain defect data. The defect data of each PCB image includes the corresponding positions of one or more identified defects. In this embodiment, the PCB image includes the corresponding positions Loc1, Loc_2, and Loc3 of the three identified defects.

[0051] Secondly, according to the defect data, intercept the local images at the corresponding positions as PCB sub-images, and send the PCB sub-images to the manual detection workstation, so as to perform manual recognition on each PCB sub-image to obtain the first manual marking result. Specifically, first intercept the corresponding local images as PCB sub-images according to the preset interception parameters (such as the size of the intercepted area) and the corresponding positions of the three defects in this PCB image, so as to obtain three PCB sub-images Img_1, Img_2, and Img_3. Among them, Img_1 corresponds to Loc1, Img_2 corresponds to Loc_2, and Img_3 corresponds to Loc3. Then send these three PCB sub-images to the staff A at the manual detection workstation, so that the staff A manually marks these three PCB sub-images respectively to obtain the corresponding first manual marking results R1, R2, and R3. At the same time, these three PCB sub-images are also sent to the AI marking module to perform AI recognition on these three PCB sub-images. Specifically, input these three PCB sub-images into the trained model for recognition to obtain the AI marking results r1, r2, and r3. It should be emphasized that in this embodiment, manual recognition and AI recognition are carried out simultaneously. However, in an embodiment of the present invention, AI recognition can also be performed first and then manual recognition, or manual recognition can be performed first and then AI recognition, which does not limit the protection scope of the present invention.

[0052] It should be noted that in this embodiment, the first manual marking results R1, R2, R3, and the second manual marking results and manual review results that appear below only include the defect category information marked by the staff (including but not limited to open circuit or short circuit). For example, R1 indicates that the defect at Loc1 is an open circuit, R2 indicates that the defect at Loc2 is an open circuit, and R3 indicates that the defect at Loc3 is a short circuit. The AI marking results r1, r2, r3 only include the defect category information output by the model. For example, r1 indicates that the defect at Loc1 is a short circuit, r2 indicates that the defect at Loc2 is an open circuit, and r3 indicates that the defect at Loc3 is a short circuit.

[0053] Finally, compare the first manual marking results corresponding to each PCB sub-image with the AI marking results obtained by AI recognition, so as to output the final detection result of the PCB sub-image.

[0054] Specifically, first compare whether the first manual marking results and the AI marking results of these three PCB sub-images are consistent:

[0055] If R1 = r1, R2 = r2, and R3 = r3, that is, the two are consistent, then use the first manual marking result or the AI marking result as the final detection result of the PCB sub-image.

[0056] If the first manual marking result and the AI marking result of one of these three PCB sub-images (such as Img_3) are inconsistent, and the remaining two are consistent, that is, R1 = r1, R2 = r2, and R3 ≠ r3, then use the first manual marking result or the AI marking result of Img_1 and Img_2 as the final detection results of their respective PCB sub-images. And send a warning message to the manual detection workstation to manually mark the PCB sub-image Img_3 again (at this time, the staff performing the manual marking work is still staff A), so as to obtain the second manual marking result R3' of Img_3. Then compare whether the second manual marking result R3' of Img_3 is consistent with the AI marking result r3 and the first manual marking result R3 of this PCB sub-image Img_3:

[0057] If the second manual marking result R3' of the PCB sub-image Img_3 is consistent with the AI marking result r3 of Img_3, then use the second manual marking result R3' or the AI marking result r3 as the final detection result of this PCB sub-image Img_3. In one embodiment, a reminder message can be sent to the corresponding staff and the error rate of the first manual marking is recorded, that is, the error rate of staff A is recorded. When the error rate of the manual marking reaches the preset error rate threshold, the staff is guided to improve their marking accuracy.

[0058] If the second manual marking result R3' of the PCB sub-image Img_3 is inconsistent with the AI marking result r3 of Img_3, then the PCB sub-image Img_3 is manually rechecked (at this time, the operator of the manual recheck is not staff member A, that is, the operator of the manual recheck is different from the operator of the manual marking), so as to obtain the manual recheck result R3'' of the PCB sub-image, and then compare whether the manual recheck result R3'' of Img_3 is consistent with the AI marking result r3 of the PCB sub-image Img_3:

[0059] If the manual recheck result R3'' of the PCB sub-image Img_3 is consistent with the AI marking result r3 of Img_3, then either the manual recheck result R3'' or the AI marking result r3 is used as the final detection result of the PCB sub-image.

[0060] If the manual recheck result R3'' of the PCB sub-image Img_3 is inconsistent with the AI marking result r3 of Img_3, then the manual recheck result R3'' is used as the final detection result of the PCB sub-image Img_3, and the steps for optimizing the model used for AI marking include:

[0061] S101: Save the PCB sub-images with inconsistent manual recheck results and AI marking results as training sample images. Specifically, in this embodiment, the PCB sub-image Img_3 is the training sample image.

[0062] S102: Set defect labels for the training sample images. In this embodiment, the defect labels only include defect type information, and the manual recheck result R3'' of Img_3 can be directly used as its defect label.

[0063] S103: Add the training sample image and its corresponding defect label to a preset training sample set. Specifically, add the PCB sub-image Img_3 and the defect label R3'' to the training sample set.

[0064] S104: Train the model used for AI marking based on the new training sample set, so as to update the parameters of the model, and thus obtain a better model.

[0065] In this embodiment, it is also possible to compare whether the manual recheck result R3'' of the PCB sub-image Img_3 is consistent with the first manual marking result R3 of Img_3. If the manual recheck result R3'' of the PCB sub-image Img_3 is inconsistent with the first manual marking result R3 of Img_3, it means that the result recognized by staff member A for the first time is incorrect, then record the error rate of the manual marking, that is, record the error rate of staff member A. When A reaches the preset error rate threshold, provide guidance to staff member A to improve his / her marking accuracy.

[0066] In addition, in this embodiment, when the second manual marking result R3' of the PCB sub-image Img_3 is inconsistent with the AI marking result r3 of Img_3, but the second manual marking result R3' of the PCB sub-image Img_3 is consistent with the first manual marking result R3 of Img_3, the above manual review process may not be performed, and the model used for AI marking may be directly optimized based on the consistency between the second manual marking result R3' and the first manual marking result R3. The optimization steps include:

[0067] S201. Save the PCB sub-image with inconsistent second manual marking result and AI marking result as a training sample image. Specifically, in this embodiment, the PCB sub-image Img_3 is the training sample image.

[0068] S202. Set a defect label for the training sample image. In this embodiment, the defect label only includes defect type information, and the first manual marking result R3 or the second manual marking result R3' of Img_3 can be directly used as the defect label of the training sample image Img_3.

[0069] S203. Add the training sample image and its corresponding defect label to a preset training sample set. Specifically, add the PCB sub-image Img_3 and the defect label R3 (or R3') to the training sample set.

[0070] S204. Train the model used for AI marking based on the new training sample set, thereby updating the parameters of the model, and obtaining a better model.

[0071] The present invention proposes to use an AI system with an AI model formed after a certain number of training times and a database with a large number of identical or similar images and their respective label information to compare a previously reviewed and marked defective PCB image with a PCB image being detected to obtain an AI marking result, and monitor the accuracy of the defect classification by the staff based on the comparison result between the manual marking result and the AI marking result. Whenever a mismatch between the manual marking result and the AI marking result is detected, the staff will be reminded to conduct a re-review. This allows the staff to have the opportunity to view the image again and, in the case where it is confirmed that the AI marking result is incorrect after one or two manual re-checks, be able to optimize the AI model to adjust or correct the defect classification.

[0072] In an embodiment of the present invention, a PCB defect detection device combining manual recognition and artificial intelligence is provided. The detection device includes a scanning unit, a recognition unit, and a processing unit. Among them, the scanning unit is configured to scan a PCB circuit board to obtain a PCB image; the recognition unit includes a manual marking unit and an AI marking unit. The manual marking unit is configured to perform manual defect marking on a PCB sub-image to obtain a first manual marking result, and the AI marking unit is configured to perform AI defect marking on the PCB sub-image to obtain an AI marking result; the processing unit is configured to compare the first manual marking result and the AI marking result corresponding to each PCB sub-image. If the first manual marking result and the AI marking result of the PCB sub-image are consistent, the processing unit uses the first manual marking result or the AI marking result as the final detection result of the PCB sub-image; if the first manual marking result and the AI marking result of the PCB sub-image are inconsistent, the processing unit performs manual marking on the PCB sub-image again to obtain a second manual marking result of the PCB sub-image; if the second manual marking result is consistent with the AI marking result of the PCB sub-image, the processing unit uses the second manual marking result or the AI marking result as the final detection result of the PCB sub-image, sends a reminder message to the corresponding staff, and records the error rate of the manual marking.

[0073] The idea of the embodiment of this PCB defect detection device belongs to the same idea as the working process of the PCB defect detection method in the above embodiment. The entire content of the above embodiment of the PCB defect detection method is incorporated into the embodiment of this PCB defect detection device by way of full reference and will not be elaborated here.

[0074] In an embodiment of the present invention, an electronic device is provided. The electronic device includes one or more processors and a storage device, and one or more programs are stored on the storage device. When the one or more programs are executed by the one or more processors, the one or more processors are caused to implement the PCB defect detection method as described above.

[0075] The idea of the embodiment of this electronic device belongs to the same idea as the working process of the PCB defect detection method in the above embodiment. The entire content of the above embodiment of the PCB defect detection method is incorporated into the embodiment of this electronic device by way of full reference and will not be elaborated here.

[0076] In an embodiment of the present invention, a computer storage medium is provided. A computer program programmed or configured to execute the PCB defect detection method as described above is stored in the computer storage medium.

[0077] The idea of the embodiment of this computer storage medium belongs to the same idea as the working process of the PCB defect detection method in the above embodiment. The entire content of the embodiment of the PCB defect detection method is incorporated into the embodiment of this computer storage medium by way of full reference and will not be elaborated herein.

[0078] The above are only the preferred embodiments of the present invention, and do not limit the scope of its patents accordingly. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present invention, directly or indirectly applied in other related technical fields, shall be equally included in the scope of patent protection of the present invention.

Claims

1. A PCB defect detection method combining manual recognition and artificial intelligence, characterized in that, the detection method includes: Scanning the PCB circuit board to obtain a PCB image; Using the detection model of the AOI device to identify the PCB image to obtain defect data, where the defect data includes the corresponding positions of one or more identified defects; According to the defect data, intercept the local image at the corresponding position as a PCB sub-image and send the PCB sub-image to the manual inspection workstation; Manually identify each PCB sub-image to obtain a first manual marking result, which includes defect category information marked by the staff; And send each PCB sub-image to the AI marking module, and use the AI model to perform AI recognition on the PCB sub-image to obtain the AI marking result corresponding to each PCB sub-image, which includes defect category information output by the AI model; During the process of identifying the first manual marking result, the staff cannot obtain the AI marking result; Compare the first manual marking result and the AI marking result corresponding to each PCB sub-image: If the first manual marking result and the AI marking result of the PCB sub-image are consistent, then use the first manual marking result or the AI marking result as the final detection result of the PCB sub-image; If the first manual marking result and the AI marking result of the PCB sub-image are inconsistent, then perform the manual marking on the PCB sub-image again to obtain the second manual marking result of the PCB sub-image, and the same staff performs the manual marking again; If the second manual marking result is consistent with the AI marking result of the PCB sub-image, then use the second manual marking result or the AI marking result as the final detection result of the PCB sub-image, send a reminder message to the corresponding staff, and record the error rate of the manual marking; If the second manual marking result is inconsistent with the AI marking result of the PCB sub-image, then perform a manual review on the PCB sub-image to obtain the manual review result of the PCB sub-image. If the manual review result is consistent with the AI marking result of the PCB sub-image, then use the manual review result or the AI marking result as the final detection result of the PCB sub-image, and record the error rate of the manual marking; wherein, the operator of the manual review is different from the operator of the manual marking.

2. The PCB defect detection method combining manual recognition and artificial intelligence according to claim 1, characterized in that, If the manual review result is inconsistent with the AI marking result of the PCB sub-image, then use the manual review result as the final detection result of the PCB sub-image, and optimize the model used for the AI marking. The optimization steps include: S101. Save the PCB sub-image with inconsistent manual review result and AI marking result as a training sample image; S102. Set defect labels for the training sample image; S103. Add the training sample image and its corresponding defect label to a preset training sample set; S104. Train the model used for AI marking based on the new training sample set, thereby updating the parameters of the model to obtain a better model.

3. The PCB defect detection method combining manual recognition and artificial intelligence according to claim 1, wherein, the detection method further includes: if the second manual marking result is consistent with the first manual marking result of the PCB sub-image, optimize the model used for AI marking; the steps of the optimization include: S201. Save the PCB sub-image with inconsistent second manual marking result and AI marking result as a training sample image; S202. Set a defect label for the training sample image; S203. Add the training sample image and its corresponding defect label to a preset training sample set; S204. Train the model used for AI marking based on the new training sample set, thereby updating the parameters of the model to obtain a better model.

4. The PCB defect detection method combining manual recognition and artificial intelligence according to claim 1, wherein, the first manual marking result, the AI marking result and the manual review result all include one or more of defect position information, defect type information and defect quantity information in the PCB sub-image.

5. A PCB defect detection device combining manual recognition and artificial intelligence, wherein, the detection device includes: a scanning unit configured to scan a PCB circuit board to obtain a PCB image; a recognition unit including a manual marking unit and an AI marking unit, the manual marking unit is configured to perform manual defect marking on the PCB sub-image to obtain a first manual marking result, and the first manual marking result includes defect category information marked by a staff member; the AI marking unit is configured to perform AI defect marking on the PCB sub-image to obtain an AI marking result, and the AI marking result includes defect category information output by the AI marking unit; during the process of recognizing the first manual marking result, the staff member cannot obtain the AI marking result; A processing unit, which is configured to compare the first manual marking result and the AI marking result corresponding to each PCB sub-image. If the first manual marking result and the AI marking result of the PCB sub-image are consistent, the processing unit uses the first manual marking result or the AI marking result as the final detection result of the PCB sub-image; if the first manual marking result and the AI marking result of the PCB sub-image are inconsistent, the processing unit performs the manual marking on the PCB sub-image again to obtain the second manual marking result of the PCB sub-image, and the same staff member performs the manual marking again; if the second manual marking result is consistent with the AI marking result of the PCB sub-image, the processing unit uses the second manual marking result or the AI marking result as the final detection result of the PCB sub-image, sends a reminder message to the corresponding staff member, and records the error rate of the manual marking; if the second manual marking result is inconsistent with the AI marking result of the PCB sub-image, a manual review is performed on the PCB sub-image to obtain the manual review result of the PCB sub-image. If the manual review result is consistent with the AI marking result of the PCB sub-image, the manual review result or the AI marking result is used as the final detection result of the PCB sub-image, and the error rate of the manual marking is recorded; wherein, the operator of the manual review is different from the operator of the manual marking.

6. An electronic device, characterized in that, the electronic device includes: one or more processors; a storage device on which one or more programs are stored; when the one or more programs are executed by the one or more processors, the one or more processors are caused to implement the PCB defect detection method according to any one of claims 1 to 4.

7. A computer storage medium, characterized in that, the computer storage medium stores a computer program that is programmed or configured to execute the PCB defect detection method according to any one of claims 1 to 4.

Images