Artificial Intelligence Defective Image Classification Method and System

Abstract

The present invention provides an artificial intelligence defective image classification method and its system. The method includes: using a transfer unit to transfer a microelectromechanical microphone product to a specific position; using a positioning unit to position the microelectromechanical microphone product; using a first image acquisition unit to scan the microelectromechanical microphone product to acquire a to-be-tested image; comparing the to-be-tested image with a plurality of reference images, and judging the defect classification of the to-be-tested image according to the comparison result; marking the defective area on the to-be-tested image according to the comparison result; and classifying and storing the microelectromechanical microphone product according to the defect classification.

Description

Technical Field

[0001] The present invention relates to an artificial intelligence defective image classification method and an artificial intelligence defective image classification system. Background Art

[0002] Since the beginning of the 19th century, the mode of industrial production has gradually changed from manual to mechanized. After the advent of the microcomputer in the mid-19th century, people began to use it for some simple automatic control, giving rise to a new revolution in the industrial world. With the demands of the industrial production to increase output, speed, and accuracy, humans have started to invest heavily in the development of automated production processes. In addition, with the rise of the concept of artificial intelligence (AI), people have begun to think about how to enable the machines in automated systems to learn to "perceive" the world as seen by humans, and thus be able to learn human behavior patterns to make judgments about things.

[0003] To this day, machine vision has become an important part of industrial automation systems. A machine vision system is an image recognition analyzer that can automatically output control signals. A machine vision system is like the eyes and brain of a human. It uses the powerful computing power of a microcomputer to appropriately analyze and recognize the acquired images, and outputs control signals as one of the parameters for controlling the machine.

[0004] In recent years, the mechanical industry has not only moved towards "automation", but also gradually towards "intelligent" production. Therefore, the mechanical industry has introduced an AI image recognition system as the first step towards intelligence. In industrial processes, one of the important steps is "quality control inspection". In the past, quality control inspections were mostly done manually. However, there may be the following disadvantages in manual quality control inspections: Workers are prone to reduce the yield rate of quality control inspections due to being overloaded during long-term work; it consumes a large amount of human resources and inspection time costs, thus reducing production efficiency. Summary of the Invention

[0005] The present invention provides an artificial intelligence defective image classification method and its system, which provide automated image detection and improve the accuracy of defective image detection.

[0006] The present invention provides an artificial intelligence defective image classification system for detecting microelectromechanical microphone products. The system includes: a conveying unit, a positioning unit, a first image acquisition unit, a storage unit, and a processing unit. The conveying unit conveys the microelectromechanical microphone product to a specific position. The positioning unit positions the microelectromechanical microphone product. The first image acquisition unit scans the microelectromechanical microphone product to acquire a to-be-tested image. The storage unit stores a plurality of modules and a plurality of reference images. The processing unit is coupled to the conveying unit, the positioning unit, the first image acquisition unit, and the storage unit, and the processor executes the said modules. The said modules include: an image recognition module, which compares the to-be-tested image with the reference images and judges the defect classification of the to-be-tested image according to the comparison result; a defect marking module, which marks the defect area on the to-be-tested image according to the comparison result; and a product classification module, which classifies and stores the microelectromechanical microphone products according to the defect classification.

[0007] The present invention provides an artificial intelligence defective image classification method for detecting microelectromechanical microphone products. The method includes: using the conveying unit to convey the microelectromechanical microphone product to a specific position; using the positioning unit to position the microelectromechanical microphone product; using the first image acquisition unit to scan the microelectromechanical microphone product to acquire a to-be-tested image; comparing the to-be-tested image with a plurality of reference images and judging the defect classification of the to-be-tested image according to the comparison result; marking the defect area on the to-be-tested image according to the comparison result; and classifying and storing the microelectromechanical microphone products according to the defect classification.

[0008] Based on the above, the artificial intelligence defective image classification method and its system provided by the embodiments of the present invention judge the defect classification to which the to-be-tested image belongs by comparing the to-be-tested image with the reference images, and then classify and store the to-be-tested object corresponding to the to-be-tested image according to the defect classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 is a schematic diagram of an artificial intelligence defective image classification system according to an exemplary embodiment of the present invention.

[0010] Figure 2 is a flowchart of an artificial intelligence defective image classification method according to an exemplary embodiment of the present invention.

[0011] Figure 3 is a schematic diagram of an artificial intelligence defective image classification system according to an exemplary embodiment of the present invention.

[0012] Figure 4 is a schematic diagram of a preliminary screening operation according to an exemplary embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0013] Reference will now be made in detail to exemplary embodiments of the present invention, examples of which are illustrated in the accompanying drawings. Wherever possible, the same reference numerals are used in the drawings and the description to refer to the same or like parts.

[0014] Figure 1 FIG. 1 is a schematic diagram of an artificial intelligence defective image classification system according to an exemplary embodiment of the present invention.

[0015] Please refer to Figure 1 , the artificial intelligence defective image classification system 100 includes a transfer unit 110, a positioning unit 120, an image acquisition unit 130, a storage unit 140, and a processing unit 150. The processing unit 150 is coupled to the transfer unit 110, the positioning unit 120, the image acquisition unit 130, and the storage unit 140.

[0016] The transfer unit 110 is configured to transfer the object to be tested to a specific position. The object to be tested is, for example, a microphone product, and the present invention is not limited thereto. In an exemplary embodiment, the transfer unit 110 can be implemented by a conveying table. For example, the processing unit 150 controls the conveying table to transfer the object to be tested to a specific position. In particular, the specific position can be below the lens of the image acquisition unit 130 to facilitate subsequent image scanning of the object to be tested by the image acquisition unit 130. Alternatively, the specific position can be a position convenient for the detector to pick up the object to be tested. The detector can pick up the object to be tested from the specific position and place the object to be tested below the lens of the image acquisition unit 130 for image scanning.

[0017] The positioning unit 120 is configured to position the object to be tested. In an exemplary embodiment, the positioning unit 120 includes a fixing element to position and fix the object to be tested. The fixing element can have a groove with sufficient space for placing the object to be tested. For example, the positioning unit 120 is disposed below the lens of the image acquisition unit 130. The processing unit 150 can control the transfer unit 110 to transfer the object to be tested to a specific position, and the positioning unit 120 automatically positions the object to be tested at the specific position. Alternatively, the detector can pick up the object to be tested from the specific position and place the object to be tested below the lens of the image acquisition unit 130, and use the positioning unit 120 to position the object to be tested.

[0018] The image acquisition unit 130 (also referred to as the first image acquisition unit) is used to scan the object to be measured to acquire an image of the object to be measured. In an exemplary embodiment, the image acquisition unit 130 may employ a Scanning Electron Microscope (SEM) to scan the surface image of the object to be measured as the image to be measured. The image to be measured scanned by the scanning electron microscope has high resolution, which is higher than the resolution of the images acquired by general optical image acquisition devices. For example, the resolution of the image to be measured may be less than or equal to 1 nanometer (nm), and the magnification of the image relative to the object to be measured will be between 100,000 and 200,000 times. However, the resolution or magnification of the image may have higher or lower resolution depending on the scanning electron microscope used, and the present invention is not limited thereto.

[0019] In another exemplary embodiment, the image acquisition unit 130 may also employ a Scanning Acoustic Tomography (SAT) or an X-ray detection system, etc., which can penetrate a certain thickness of solid and liquid substances and obtain an internal image of the object, to scan the internal image of the object to be measured as the image to be measured.

[0020] The storage unit 140 is used to store various modules and various necessary data and codes when the artificial intelligence defect image classification system 100 is running. The storage unit 140 may be one or a combination of a fixed or removable random access memory (RAM), a read-only memory (ROM), a flash memory, a hard disk, or other similar devices. In particular, in this exemplary embodiment, the storage unit 140 stores an image recognition module 141, a defect marking module 142, and a product classification module 143. In addition, the storage unit 140 also stores a plurality of reference images 144.

[0021] The image recognition module 141 will compare the image to be measured with the reference images 144 and determine the defect classification of the image to be measured according to the comparison result. The defect marking module 142 will mark the defect area on the image to be measured according to the comparison result. The product classification module 143 will classify and store the object to be measured according to the defect classification. How to complete the operation of the artificial intelligence defect image classification system 100 through the image recognition module 141, the defect marking module 142, and the product classification module 143 will be described later.

[0022] The processing unit 150 is coupled to the storage unit 140 for accessing the data in the storage unit 140 to perform various operations of the artificial intelligence defective image classification system 100. In particular, the processing unit 150 executes the aforementioned image recognition module 141, defect marking module 142, and product classification module 143 to perform the artificial intelligence defective image classification method provided by the present invention. The processing unit 150 may be a central processing unit (CPU), or other programmable general-purpose or special-purpose microprocessors, digital signal processors (DSPs), programmable controllers, application specific integrated circuits (ASICs), or other similar components, or a combination of the above components. The present invention is not limited thereto.

[0023] Figure 2 It is a flowchart of the artificial intelligence defective image classification method illustrated according to an exemplary embodiment of the present invention.

[0024] Please refer to Figure 1 and Figure 2 , the method of this exemplary embodiment is applicable to the artificial intelligence defective image classification system 100 in the above embodiment. The following will describe the detailed steps of this artificial intelligence defective image classification method in conjunction with the various components in the artificial intelligence defective image classification system 100.

[0025] It should be noted first that this embodiment details the artificial intelligence defective image classification method with a microphone product as the object to be tested. In particular, the microphone product may be a micro-electro-mechanical systems (MEMS) microphone product. The present invention is not limited thereto.

[0026] In step S202, the processing unit 150 transmits the MEMS microphone product to a specific position through the transmission unit 110 and locates the MEMS microphone product through the positioning unit 120. In addition, if the MEMS microphone product has a cover, the cover can be removed by the heating unit before step S202 to expose the chip package structure inside the MEMS microphone product.

[0027] In step S204, the processing unit 150 scans the MEMS microphone product through the image acquisition unit 130 to acquire the image to be tested. Specifically, the image acquisition unit 130 can use a scanning electron microscope to scan the MEMS microphone product and generate the image to be tested corresponding to the scanned area.

[0028] Based on the image to be measured and the reference image 144 stored in the storage unit 140, the processing unit 150 will run the image recognition module 141, the defect marking module 142, and the product classification module 143, and then determine the classification of the detected MEMS microphone product.

[0029] Specifically, in step S206, the image recognition module 141 compares the image to be measured with multiple reference images 144, and determines the defect classification of the image to be measured according to the comparison result. The defect classification may include, for example, ballbond left products, wedge bond left products, chip crack products, contamination products, and / or OK products. The present invention is not limited thereto.

[0030] For example, the image recognition module 141 can use an image recognition model to determine whether there are defects similar to each reference image 144 in the image to be measured, so as to detect the defect classification of the object to be measured. In an exemplary embodiment, the image recognition model includes a convolutional neural network model (Convolutional Neural Network, CNN) and a classifier. When the image recognition module 141 inputs the image to be measured into the convolutional neural network model, the image to be measured will pass through a convolutional layer to extract features, a rectified linear units layer (ReLU layer) to enhance the function characteristics, a pooling layer to reduce the spatial size of the data, etc., so as to obtain the defect features of the defect (for example, the contour of the defect). After the image to be measured passes through the convolutional neural network model and obtains the defect features, the image recognition module 141 will further input the defect features into the classifier. The classifier determines which defect classification corresponding to which reference image 144 this defect belongs to based on the defect features, and generates a comparison result accordingly. The image recognition module 141 will determine the defect classification of the image to be measured according to the comparison result.

[0031] In this exemplary embodiment, the classifier included in the image recognition model is trained using multiple reference images 144, the target bounding boxes corresponding to each reference image 144, and the defect classification corresponding to each bounding box.

[0032] First, the processing unit 150 obtains an image data set. The image data set includes the above-mentioned multiple reference images 144, the target bounding boxes corresponding to each reference image 144, and the defect classifications corresponding to each bounding box. The target bounding boxes corresponding to the reference image 144 and the defect classifications corresponding to the target bounding boxes can be manually marked. Specifically, the processing unit 150 includes a defect labeling module (not shown). The defect labeling module receives, through an input unit (not shown), the target bounding boxes marked by the detector on the reference image 144 and the defect classifications corresponding to the input target bounding boxes, so as to label and classify the target bounding boxes of the reference image 144. For example, the defect labeling module can be a program such as LabelImg. In an exemplary embodiment, the reference image 144 is in a first image format, and the marked target bounding boxes are stored in a second image format, and the first image format is different from the second image format. For example, the first image format is the JPEG format, and the second image format is the XML (PASCAL VOC format).

[0033] Next, the processing unit 150 inputs the image data set into the image recognition model to train the image recognition model, and stores the trained image recognition model in the storage unit 140. However, in other embodiments of the present invention, the artificial intelligence defect image classification system 100 stores the image recognition model in a remote storage unit, and the image recognition module 141 obtains the corresponding image recognition model from the remote storage unit according to requirements.

[0034] Back to Figure 2 , in step S208, the defect labeling module 142 marks the defect area on the image to be tested according to the comparison result. For example, the defect labeling module 142 judges the defect area of the image to be tested according to the comparison result output by the image recognition model, and marks the defect area on the image to be tested.

[0035] In step S210, the product classification module 143 classifies and stores the MEMS microphone products according to the defect classification. Specifically, the product classification module 143 classifies and stores the MEMS microphone products corresponding to the image to be tested according to the defect classification judged by the image recognition module 141 for the image to be tested.

[0036] It is worth mentioning that in an exemplary embodiment of the present invention, the reference image 144 used to train the image recognition model is generated by scanning the same scanning electron microscope as the image to be tested. With the high-resolution reference image 144 generated by scanning the MEMS microphone product with a scanning electron microscope, the trained image recognition model can more accurately identify the defect classification of the image to be tested. Thereby, the accuracy of image recognition can be improved.

[0037] Since a scanning electron microscope requires a long image scanning time. Therefore, in another exemplary embodiment of the present invention, before the artificial intelligence defect image classification system 100 performs the above-mentioned defect classification, the objects to be tested with more obvious defects can be preliminarily screened first.

[0038] Figure 3 FIG. 4 is a schematic diagram of an artificial intelligence defect image classification system according to an exemplary embodiment of the present invention.

[0039] Please refer to Figure 3 , in addition to the same components as Figure 1 , the artificial intelligence defect image classification system 100 may further include a preliminary screening unit 160. The preliminary screening unit 160 is coupled to the processing unit 150 and includes, but is not limited to, an X-ray detection system 161, a heating unit 162, and an image acquisition unit 163 (also referred to as the second image acquisition unit).

[0040] The X-ray detection system 161 is used to pre-check the internal structure of the object to be tested. Since the completed object to be tested may have a cover, this cover will obscure the internal structure of the object to be tested. Therefore, the X-ray detection system 161 can be used to check whether there are obvious component deficiencies in the internal structure of the object to be tested first.

[0041] The heating unit 162 is used to heat the object to be tested to remove the cover of the object to be tested. The heating unit 162 is, for example, a hot plate, and the present invention is not limited thereto.

[0042] The image acquisition unit 163 is used to acquire the preliminary screening image of the object to be tested. The resolution of the image acquired by the image acquisition unit 163 is lower than that of the image acquisition unit 130. For example, the image acquisition unit 163 can be an optical image acquisition device. In the process of industrial manufacturing, the object to be tested will be sent to an automated optical inspection (AOI) device for preliminary defect screening. The image acquisition unit 163 is, for example, an optical image acquisition device provided in the automated optical inspection device. By scanning the object to be tested through the image acquisition unit 163 to acquire the image of the object to be tested, the automated optical inspection device will judge the structural integrity of the object to be tested by the acquired image, and then find out the objects to be tested suspected of having defects.

[0043] Figure 4 FIG. 24 is a schematic diagram of a preliminary screening operation according to an exemplary embodiment of the present invention. In this exemplary embodiment, the MEMS microphone 5 is taken as an example for illustration. However, the present invention does not limit the types of objects to be tested.

[0044] Please refer to Figure 4, the MEMS microphone 5 is a microphone product with a cover. In step S402, the processing unit 150 uses the X-ray detection system 161 to check whether the MEMS microphone 5 includes a chip. If the MEMS microphone 5 includes a chip, then in step S404, the processing unit 150 uses the heating unit 162 to remove the cover of the MEMS microphone 5 to obtain the chip package structure of the MEMS microphone 5. On the other hand, if the MEMS microphone 5 does not include a chip, the processing unit 150 can directly classify the MEMS microphone 5 (for example, classify the defect as "chip missing"), and store the MEMS microphone 5 separately according to the classification.

[0045] In step S406, the processing unit 150 uses the image acquisition unit 163 to scan the chip package structure to acquire a preliminary screening image, and performs a preliminary defect detection operation on the preliminary screening image. Specifically, the image acquisition unit 163 uses an optical image acquisition device to scan the chip package structure of the MEMS microphone 5, and generates a preliminary screening image corresponding to the scanned area. Then, the processing unit 150 performs a preliminary defect detection operation on the preliminary screening image to determine whether the preliminary screening image has a preset defect. If the preliminary screening image does not have a preset defect, then in step S408, the processing unit 150 uses the image acquisition unit 130 to acquire a to-be-tested image and perform Figure 2 the defect classification operation shown. The to-be-tested image reflects the chip package structure of the MEMS microphone 5. On the other hand, if the preliminary screening image has a preset defect, the processing unit 150 can directly classify the MEMS microphone 5 (for example, classify the defect as "chip defect"), and store the MEMS microphone 5 separately according to the classification.

[0046] It is worth mentioning that the detector can add, delete, or adjust the number of the X-ray detection system 161, the image acquisition unit 163, and the image acquisition unit 130 and the sequence of each defect detection according to requirements, and the present invention is not limited thereto.

[0047] In another exemplary embodiment of the present invention, the image acquisition unit 130 can also use an ultrasonic scanning microscope or an X-ray detection system to scan the MEMS microphone product and generate a to-be-tested image and a reference image 144. By scanning the to-be-tested object with an ultrasonic scanning microscope or an X-ray detection system, it is possible to penetrate the cover of the to-be-tested object and generate an image reflecting the internal structure of the chip package structure without going through the cover removal process. The processor 150 can use the generated image to train an image recognition model and use the image to perform defect classification. The method of recognizing images and training models can refer to the foregoing description and will not be elaborated here.

[0048] In summary, the artificial intelligence defective image classification method and its system provided by the embodiments of the present invention can collect high-resolution images to be tested and reference images, and determine the defective classification to which the images to be tested belong by comparing the images to be tested with the reference images. Furthermore, the objects to be tested corresponding to the images to be tested can be classified and stored according to the defective classification. Accordingly, the image recognition model trained by the high-resolution reference images can more accurately identify the defective classification of the images to be tested. Thereby, the accuracy of image recognition can be improved.

[0049] In this embodiment, the image recognition model is trained according to high-resolution reference images, and the trained image recognition model is used to recognize images to generate recognition results. In addition, regression analysis can be used to identify the relationship between process parameters and yield, and applied to establish automatic yield improvement to improve the production yield of the production line.

[0050] Although the present invention has been disclosed above with embodiments, it is not intended to limit the present invention. Any person with ordinary knowledge in the technical field can make some changes and modifications without departing from the spirit and scope of the present invention. Therefore, the protection scope of the present invention shall be subject to that defined by the appended claims.

[0051] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An artificial intelligence defective image classification system for detecting microelectromechanical microphone products, the system comprising: A transfer unit for transferring the microelectromechanical microphone product to a specific position; A positioning unit for positioning the microelectromechanical microphone product; A first image acquisition unit for scanning the microelectromechanical microphone product to acquire an image to be tested; A storage unit for storing a plurality of modules and a plurality of reference images; And A processing unit coupled to the transfer unit, the positioning unit, the first image acquisition unit and the storage unit, the processing unit executing the modules, the modules including: An image recognition module for comparing the image to be tested with the reference images and judging the defect classification of the image to be tested according to the comparison result; A defect marking module for marking the defect area on the image to be tested according to the comparison result; and A product classification module for classifying and storing the microelectromechanical microphone products according to the defect classification, wherein before using the transfer unit to transfer the microelectromechanical microphone product to a specific position, the processing unit is configured to: Use an X-ray detection system to check whether the microelectromechanical microphone product includes a chip; and If the microelectromechanical microphone product includes the chip, including: Use a heating unit to remove the cover of the microelectromechanical microphone product to expose the chip packaging structure inside the microelectromechanical microphone product; Use a second image acquisition unit to scan the chip packaging structure to acquire a preliminary screening image; Judge whether the preliminary screening image has a preset defect; If the preliminary screening image does not have the preset defect, use the first image acquisition unit to acquire the image to be tested to judge the defect classification of the image to be tested; and If the preliminary screening image has the preset defect, directly classify the microelectromechanical microphone into the defect classification of chip defect.

2. The artificial intelligence defective image classification system according to claim 1, wherein the first image acquisition unit includes at least one of a scanning electron microscope, an ultrasonic scanning microscope and an X-ray detection system.

3. The artificial intelligence defective image classification system according to claim 1, wherein the resolution of the image to be tested is less than or equal to 1 nanometer.

4. The artificial intelligence defective image classification system according to claim 1, wherein the image magnification of the image to be tested relative to the microelectromechanical microphone product is between 100,000 and 200,000 times.

5. The artificial intelligence defective image classification system according to claim 1, wherein the defect classification includes at least one of spherical wire missing items, wedge-shaped wire missing items, chip cracked items, contaminated items and qualified items.

6. The artificial intelligence defective image classification system according to claim 1, wherein the reference images are in a first image format, wherein after the processing unit marks and classifies the target bounding boxes of the reference images through the defect marking module, the target bounding boxes are stored in a second image format, wherein the first image format is different from the second image format.

7. The artificial intelligence defective image classification system according to claim 1, wherein the image to be tested reflects the chip packaging structure of the microelectromechanical microphone product.

8. The artificial intelligence defective image classification system according to claim 1, wherein the system further includes: The X-ray detection system, the heating unit, and the second image acquisition unit, the X-ray detection system, the heating unit, and the second image acquisition unit are coupled to the processing unit.

9. An artificial intelligence defect image classification method for detecting microelectromechanical microphone products, the method comprising: Using a transfer unit to transfer the microelectromechanical microphone product to a specific position; Using a positioning unit to position the microelectromechanical microphone product; Using a first image acquisition unit to scan the microelectromechanical microphone product to acquire a to-be-tested image; Comparing the to-be-tested image with a plurality of reference images, and judging the defect classification of the to-be-tested image according to the comparison result; Marking the defect area on the to-be-tested image according to the comparison result; And Classifying and storing the microelectromechanical microphone products according to the defect classification, wherein before using the transfer unit to transfer the microelectromechanical microphone product to a specific position, it further includes: Using an X-ray detection system to check whether the microelectromechanical microphone product includes a chip; and If the microelectromechanical microphone product includes the chip, it includes: Using a heating unit to remove the cover of the microelectromechanical microphone product to expose the chip packaging structure inside the microelectromechanical microphone product; Using a second image acquisition unit to scan the chip packaging structure to acquire a preliminary screening image; Judging whether the preliminary screening image has a preset defect; If the preliminary screening image does not have the preset defect, using the first image acquisition unit to acquire the to-be-tested image to judge the defect classification of the to-be-tested image; and If the preliminary screening image has the preset defect, directly classifying the microelectromechanical microphone into the defect classification of chip defect.

10. The artificial intelligence defect image classification method according to claim 9, wherein the first image acquisition unit includes at least one of a scanning electron microscope, an ultrasonic scanning microscope, and an X-ray detection system.

11. The artificial intelligence defect image classification method according to claim 9, wherein the resolution of the to-be-tested image is less than or equal to 1 nanometer.

12. The artificial intelligence defect image classification method according to claim 9, wherein the image magnification of the to-be-tested image relative to the microelectromechanical microphone product is between 100,000 and 200,000 times.

13. The artificial intelligence defect image classification method according to claim 9, wherein the defect classification includes at least one of spherical wire missing products, wedge-shaped wire missing products, chip cracked products, contaminated products, and qualified products.

14. The artificial intelligence defect image classification method according to claim 9, wherein the reference image is in a first image format, and the method further includes: After marking and classifying the target bounding box of the reference image through a defect marking module, storing the target bounding box in a second image format, wherein the first image format is different from the second image format.

15. The artificial intelligence defect image classification method according to claim 9, wherein the to-be-tested image reflects the chip packaging structure of the microelectromechanical microphone product.

Images