A system and method for microscopic character recognition and defect detection based on deep learning

Abstract

The present invention discloses a system and method for microscopic character recognition and defect detection based on deep learning, including a microscopic imaging visual device and a computer, wherein the computer has a microscopic character recognition defect detection system connected to the microscopic imaging visual device, the microscopic character recognition defect detection system includes a focus control module, a defect detection module and a character recognition module, the focus control module is connected to a lifting stepper motor, the defect detection module is used to perform defect detection on a microchip on a sample collection bar, the character recognition module is used to recognize the characters of the microchip on the sample collection bar, and the microscopic character recognition defect detection system outputs defect conditions and character recognition conditions according to defect detection and character recognition. The present invention can automatically realize microscopic imaging focusing operations, improve focusing efficiency, focusing quality and subsequent microscopic imaging quality, realize large-scale, standardized, and automated rapid detection of microchips, and comprehensively realize defect detection and character recognition.

Description

Technical Field

[0001] The present invention relates to the field of microchip detection, and in particular to a system and method for microscopic character recognition and defect detection based on deep learning. Background Art

[0002] With the development of modern industry, integrated circuits are not only getting bigger and bigger in scale, but also have higher and higher requirements in terms of precision. Among them, as an indispensable part of large-scale integrated circuits, chips are becoming increasingly powerful, and their tasks in the system are also becoming more diversified and complex. Microchips are a very important type of special chip in the information age. Unlike traditional large-scale and large-area chips such as CPUs, single-chip microcomputers and FPGAs, microchips are very small, and their cross-sectional area is often only tens of square millimeters or less, but in actual industrial production and scientific research, microchips play a vital role. In some small spaces or embedded structures that do not require too much computing power, due to the limitations of process, cost, heat dissipation and energy consumption, only microchips can meet the needs of the project. However, precisely because the area of ​​microchips is small, it is generally difficult to observe the information and characters on their surface with the naked eye. Therefore, it is very difficult to confirm the serial number engraved on such chips during their production process and the subsequent maintenance process that may occur. If the tiny characters are observed manually with the help of a microscope, this method is slow and consumes a lot of human resources. This brings technical difficulties to the quality inspection of microchips, and also brings great troubles to batch control, good product screening and subsequent troubleshooting in the production process. The difficulties of existing technologies are: 1. Difficulty in positioning and focusing due to the small size of the chip; 2. Low recognition accuracy. The highest recognition accuracy of traditional methods is not high (no more than 75%), and manual participation is required, which cannot achieve automatic focusing, defect detection and character recognition; 3. Defects generated by microchips during the production process, such as stains and damage, are not detected by traditional methods, and the production process cannot be traced and corrected. Summary of the invention

[0003] The following technical problems exist in the prior art: positioning and focusing are difficult due to the small size of the chip, the recognition accuracy of the traditional method is not high, and manual participation is required; the traditional method does not detect the defects (such as stains and damage, etc.) generated by the microchip during the production process, and is unable to trace and correct the production process; the purpose of the present invention is to provide a microscopic character recognition and defect detection system and method based on deep learning, which can automatically realize the microscopic imaging focusing operation, improve the focusing efficiency, focusing quality and subsequent microscopic imaging quality, realize large-scale, standardized and automated rapid detection of microchips and comprehensively realize defect detection and character recognition, quickly obtain microscopic character recognition defect data, easily identify and count alarm signals and locate the problems of the batch of microchips, which is of great significance to improving the yield rate of microchips.

[0004] The purpose of the present invention is achieved through the following technical solutions:

[0005] A microscopic character recognition and defect detection system based on deep learning includes a horizontal platform, a sample collection bar and a computer. The sample collection bar is provided with a plurality of microchips. A vertical rod is fixed vertically on the horizontal platform. A lifting stepping motor is provided on the top of the vertical rod. A lifting linkage rod is provided horizontally on the lifting stepping motor. The lifting stepping motor is used to drive the lifting linkage rod to move up and down. A light source lifting cross bar is installed in the middle of the vertical rod. A microscopic imaging visual device is provided at the end of the lifting linkage rod. A ring light source corresponding to the microscopic imaging visual device is provided at the end of the light source lifting cross bar. A microscopic character recognition defect detection system connected to the microscopic imaging visual device is provided inside the computer. The microscopic character recognition defect detection system includes a focus control module, a defect detection module and a character recognition module. The focus control module is connected to the lifting stepping motor. The defect detection module is used to perform defect detection on the microchip on the sample collection bar. The character recognition module is used to recognize the characters of the microchip on the sample collection bar. The microscopic character recognition defect detection system outputs defect conditions and character recognition conditions according to defect detection and character recognition.

[0006] In order to better realize the present invention, a micro-motion platform is provided on the horizontal platform seat, a sample collection bar positioning seat is provided on the micro-motion platform, the sample collection bar is installed on the sample collection bar positioning seat, the micro-motion platform is used to drive the micro-motion platform to move forward, backward, left and right on the horizontal plane, and the micro-motion platform is connected to the focus control module.

[0007] Preferably, the microscopic imaging visual device comprises a camera and a telecentric lens corresponding to the camera, the camera is mounted on a lifting linkage rod, and the camera of the microscopic imaging visual device is connected to a microscopic character recognition defect detection system in a computer via a data cable.

[0008] Preferably, the camera is an area array CMOS camera, and the data cable is a USB communication data cable.

[0009] A method for microscopic character recognition and defect detection based on deep learning, including a microscopic character recognition defect detection system, wherein the method is as follows:

[0010] S1. The sample collection bar is provided with N microchip attachment sites, and the N microchips are correspondingly installed on the N microchip attachment sites of the sample collection bar, and each microchip attachment site is correspondingly assigned an ID number; a three-dimensional focus movement system is built, and the sample collection bar is arranged on the XY axis direction movement system of the three-dimensional focus movement system, and the microscopic imaging visual device is arranged on the Z axis movement system of the three-dimensional focus movement system. The microscopic imaging visual device includes an area array CMOS camera and a telecentric lens corresponding to the area array CMOS camera, and the area array CMOS camera is connected to a microscopic character recognition defect detection system;

[0011] S2. The microscopic character recognition defect detection system includes a focus control module, a defect detection module and a character recognition module. The focus control module is connected to the XY axis direction movement system and the Z axis movement system respectively. The focus control module drives the sample collection bar to move in the X axis and Y axis directions and the microscopic imaging visual device to move up and down in the Z axis direction to perform automatic focusing operation;

[0012] S3. A neural network model is built in the character recognition module of the microscopic character recognition defect detection system. The neural network model includes a text detection network, a direction identifier, and a text content network. The neural network model is subjected to deep learning training for text region detection, direction recognition, and text content recognition.

[0013] S4, the microscopic imaging visual device performs focused microscopic imaging on each microchip of the sample collection strip and transmits the microchip image to the microscopic character recognition defect detection system, wherein the microchip image covers at least one microchip and the ID number of the microchip attachment site;

[0014] S5. The defect detection module of the microscopic character recognition defect detection system performs threshold-based defect judgment on the microchip image and obtains the defect result; the neural network model of the character recognition module performs character recognition on the microchip image and obtains the character recognition result; the microscopic character recognition defect detection system forms microscopic character recognition defect data by corresponding the defect result, character recognition result and ID number one by one and saves the output.

[0015] Preferably, the XY axis movement system of the three-dimensional focusing movement system in step S1 of the present invention includes a horizontal platform seat and a micro-motion platform installed on the horizontal platform seat, and the micro-motion platform is used to drive the sample collection bar to move in the horizontal plane in the X-axis or Y-axis direction; the Z axis movement system includes a lifting stepper motor, and the lifting stepper motor is used to drive the microscopic imaging vision device to move up and down in the Z axis direction.

[0016] Preferably, the three-dimensional focus movement system in step S1 of the present invention includes a vertical rod, a lifting stepping motor is arranged on the upper part of the vertical rod, and a ring-shaped light source corresponding to the telecentric lens is installed in the middle part of the vertical rod through a light source lifting cross bar; the height of the ring-shaped light source is adjusted by lifting and lowering, and the brightness of the ring-shaped light source is adjusted until the contrast between the characters of the microchip on the sample collection strip and the background is prominent and the characters are clear, so as to facilitate the area array CMOS camera to shoot and later identify.

[0017] The present invention provides another preferred technical solution (in order to achieve point focusing and point microscopic imaging): in step S1 of the present invention, the sample collection strip is provided with 32 microchip attachment sites, and each microchip attachment site is recorded as (x1, y1), (x2, y2) ... (x32, y32) in sequence according to the site coordinates. After the focus control module in step S2 completes the focusing, the microscopic character recognition defect detection system establishes an XY axis coordinate system according to the microscopic imaging visual device when the focusing is completed and correspondingly identifies the site coordinates of the microchip attachment site; in step S4, the microscopic character recognition defect detection system controls the XY axis direction movement system to move according to the site coordinates to realize the microscopic imaging visual device to perform point focusing or / and point microscopic imaging on the microchip on each microchip attachment site.

[0018] The present invention provides another preferred technical solution: the neural network model method in step S3 includes the following:

[0019] S31. Build a text detection network in the neural network model. The text detection network is based on the segmentation DB text detection algorithm. First, the ResNet convolutional body is used to perform feature extraction and upsampling fusion operations to obtain a feature map. Then, the feature map is used to predict the probability map. Finally, an approximate binary map is calculated through the feature map and the probability map. The binary map represents the area where the text is located.

[0020] S32, the direction identifier in the neural network model performs 180-degree inversion of the microchip (the microchip is placed with the characters facing upward but inverted 180 degrees in the horizontal plane, then the microchip image will also be inverted 180 degrees, and this recognition process is performed to facilitate rapid standardization and accurate processing), and its output is two results of 0 degrees or 180 degrees. If the result is 180 degrees, it will be reversed and enter step S33;

[0021] S33. In the neural network model, the text content network is built using the Rosetta OCR system and the fully convolutional model. The text content network is trained using the CTC loss function.

[0022] In step S5, the defect detection module uses the binary image and threshold segmentation to make defect judgment. If a defect exists, the output defect result contains an alarm signal. The microscopic character recognition defect data method of the microscopic character recognition defect detection system includes:

[0023] S51, if the text area is not detected or the text confidence is lower than 0.92 or the number of character bits is incorrect, the character recognition result of the microscopic character recognition defect data contains an alarm signal;

[0024] S52: If the character can be recognized but defects and / or stains are detected, the defect result of the microscopic character recognition defect data contains an alarm signal.

[0025] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0026] (1) The present invention can automatically realize the microscopic imaging focusing operation, improve the focusing efficiency, focusing quality and subsequent microscopic imaging quality, realize large-scale, standardized and automated rapid detection of microchips and comprehensively realize defect detection and character recognition, quickly obtain microscopic character recognition defect data, easily identify and count alarm signals and locate the problem of the batch of microchips, which is of great significance to improving the yield rate of microchips.

[0027] (2) The present invention is equipped with a three-dimensional focus movement system and controls the three-dimensional focus movement system through a focus control module and automatically realizes the focus operation according to the clarity detection. It can detect defects including stains and damage on the microchip. The trained neural network model in the character recognition module can recognize the microchip and ID number with high efficiency and high precision. The recognition accuracy initially reaches more than 90%. If the trained neural network model is stronger, it can reach more than 98%.

[0028] (3) The present invention can realize point focusing and point microscopic imaging of each microchip on the sample collection strip in a batch and standardized manner, and then perform defect detection and character recognition in sequence to obtain microscopic character recognition defect data; it can also focus, microscopically image, and perform graphic detection processing on multiple microchips on the sample collection strip at one time, and correspond the data to the microchip through ID numbers or site coordinates, so as to efficiently complete the microchip detection operation.

[0029] (4) The present invention effectively improves the positioning and focusing speed and quality, improves the defect detection and recognition accuracy, and significantly improves the efficiency of character detection of large batches of microchips, which is beneficial to improving the yield rate of microchips and completely replacing manual work in this field. The defect detection function allows manufacturers to trace the upstream chip production steps. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a schematic diagram of the process of the method for microscopic character recognition and defect detection based on deep learning in the embodiment;

[0031] Figure 2A software interface diagram for character recognition in the embodiment;

[0032] Figure 3 The three types of microscopic character recognition defect data in the embodiment are corresponding diagrams;

[0033] Figure 4 It is a structural schematic diagram of a deep learning-based microscopic character recognition and defect detection system in an embodiment;

[0034] Figure 5 It is a schematic diagram of the cooperation between the sample collection strip positioning seat and the sample collection strip in the embodiment.

[0035] The names corresponding to the reference numerals in the accompanying drawings are:

[0036] 1-horizontal platform seat, 2-vertical rod, 3-lifting stepping motor, 31-lifting linkage rod, 4-microscopic imaging visual device, 41-camera, 42-telecentric lens, 5-ring light source, 51-light source lifting cross bar, 6-micro-motion platform, 7-sample collection bar, 8-computer, 9-data cable, 10-sample collection bar positioning seat. DETAILED DESCRIPTION

[0037] The present invention is further described in detail below in conjunction with embodiments:

[0038] Example

[0039] like Figure 4 , Figure 5 As shown, a system for microscopic character recognition and defect detection based on deep learning includes a horizontal platform base 1, a sample collection bar 7 and a computer 8. The sample collection bar 7 is provided with a plurality of microchips. A vertical rod 2 is vertically fixed to the horizontal platform base 1. A lifting stepper motor 3 is provided on the top of the vertical rod 2. A lifting linkage rod 31 is horizontally provided on the lifting stepper motor 3. The lifting stepper motor 3 is used to drive the lifting linkage rod 31 to move up and down. A light source lifting cross bar 51 is installed in the middle of the vertical rod 2. A microscopic imaging visual device 4 is provided at the end of the lifting linkage rod 31. The end of the light source lifting cross bar 51 is provided with a microscopic imaging visual device The device 4 corresponds to a ring light source 5; the computer 8 has a microscopic character recognition defect detection system connected to the microscopic imaging visual device 4, the microscopic character recognition defect detection system includes a focus control module, a defect detection module and a character recognition module, the focus control module is connected to the lifting stepper motor 3, the defect detection module is used to perform defect detection on the microchip on the sample collection bar 7, the character recognition module is used to recognize the characters of the microchip on the sample collection bar 7, and the microscopic character recognition defect detection system outputs defect conditions and character recognition conditions according to defect detection and character recognition.

[0040] According to a preferred embodiment of the present invention, a micro-motion platform 6 is provided on the horizontal platform seat 1, a sample collection bar positioning seat 10 is provided on the micro-motion platform 6, the sample collection bar 7 is installed on the sample collection bar positioning seat 10, the micro-motion platform 6 is used to drive the micro-motion platform 6 to move forward, backward, left and right on the horizontal plane, and the micro-motion platform 6 is connected to the focus control module.

[0041] According to a preferred embodiment of the present invention, the microscopic imaging visual device 4 includes a camera 41 (preferably, the camera 41 is an area array CMOS camera) and a telecentric lens 42 corresponding to the camera 41. The camera 41 is installed on the lifting linkage rod 31. The camera 41 of the microscopic imaging visual device 4 is connected to the microscopic character recognition defect detection system in the computer 8 via a data cable 9 (preferably, the data cable 9 is a USB communication data cable, and this embodiment adopts a USB3.0 bus).

[0042] like Figure 1 to Figure 5 As shown, a method for microscopic character recognition and defect detection based on deep learning includes a microscopic character recognition defect detection system. The microscopic character recognition defect detection system of this embodiment is set in a computer 8, and the method is as follows:

[0043] S1. The sample collection bar 7 is provided with N microchip attachment sites, and the N microchips are correspondingly installed on the N microchip attachment sites of the sample collection bar 7, and each microchip attachment site is correspondingly assigned an ID number; a three-dimensional focusing movement system is built, and the sample collection bar 7 is arranged on the XY axis direction movement system of the three-dimensional focusing movement system, and the microscopic imaging visual device 4 is arranged on the Z axis movement system of the three-dimensional focusing movement system, and the microscopic imaging visual device 4 includes an area array CMOS camera and a telecentric lens 42 corresponding to the area array CMOS camera, and the area array CMOS camera is connected to the microscopic character recognition defect detection system;

[0044] S2, the microscopic character recognition defect detection system includes a focus control module, a defect detection module and a character recognition module. The focus control module is connected to the XY axis direction movement system and the Z axis movement system respectively. The focus control module drives the sample collection bar 7 to move in the X axis and Y axis directions and the microscopic imaging visual device 4 to move up and down in the Z axis direction to perform automatic focusing operation;

[0045] S3. A neural network model is built in the character recognition module of the microscopic character recognition defect detection system. The neural network model includes a text detection network, a direction identifier, and a text content network. The neural network model is subjected to deep learning training for text region detection, direction recognition, and text content recognition.

[0046] S4, the microscopic imaging visual device 4 performs focused microscopic imaging on each microchip of the sample collection strip 7 and transmits the microchip image to the microscopic character recognition defect detection system, wherein the microchip image covers at least one microchip and the ID number of the microchip attachment site;

[0047] S5. The defect detection module of the microscopic character recognition defect detection system performs threshold-based defect judgment on the microchip image and obtains the defect result; the neural network model of the character recognition module performs character recognition on the microchip image and obtains the character recognition result; the microscopic character recognition defect detection system forms microscopic character recognition defect data by corresponding the defect result, character recognition result and ID number one by one and saves the output.

[0048] According to one embodiment of the present invention, the XY axis direction moving system of the three-dimensional focusing moving system in step S1 includes a horizontal platform seat 1 and a micro-motion platform 6 installed on the horizontal platform seat 1, the micro-motion platform 6 is used to drive the sample collection bar 7 to move in the horizontal plane in the X-axis or Y-axis direction (the micro-motion platform 6 is used to achieve the focusing of the microchip horizontal plane on the sample collection bar 7 in the X-axis and Y-axis directions); the Z axis moving system includes a lifting stepper motor 3, and the lifting stepper motor 3 is used to drive the microscopic imaging visual device 4 to move up and down in the Z axis direction (the lifting stepper motor 3 is used to achieve vertical focusing).

[0049] According to one embodiment of the present invention, the three-dimensional focus movement system in step S1 includes a vertical rod 2, a lifting stepping motor 3 is arranged on the upper part of the vertical rod 2, and a ring light source 5 corresponding to the telecentric lens 42 is installed in the middle part of the vertical rod 2 through a light source lifting cross bar 51; the height of the ring light source 5 is lifted and adjusted, and the brightness of the ring light source 5 is adjusted until the contrast between the characters of the microchip on the sample collection strip 7 and the background is prominent and the characters are clear, so as to facilitate the area array CMOS camera to shoot and later identify (observe the clarity and brightness of the microchip characters, adjust the height and brightness of the ring light source until the contrast between the characters and the background is prominent and the characters are clearly visible).

[0050] According to an embodiment of the present invention (this embodiment can realize point focusing and point microscopic imaging), in step S1, the sample collection strip 7 is provided with 32 microchip attachment sites, and each microchip attachment site is recorded as (x1, y1), (x2, y2) ... (x32, y32) in sequence according to the site coordinates. After the focus control module completes the focusing in step S2, the microscopic character recognition defect detection system establishes an XY axis coordinate system according to the microscopic imaging visual device 4 when the focusing is completed and correspondingly identifies the site coordinates of the microchip attachment site; in step S4, the microscopic character recognition defect detection system controls the XY axis direction movement system to move according to the site coordinates to realize the microscopic imaging visual device 4 to perform point focusing or / and point microscopic imaging on the microchip on each microchip attachment site.

[0051] According to one embodiment of the present invention, there are 32 microchip attachment sites on a sample strip, and the coordinates of the 32 sites are recorded as (x1, y1), (x2, y2) ... (x32, y32) and written into the program of the focus control module; an automatic focus algorithm in the z-axis direction is designed, and the automatic focus algorithm is to first move the microscopic imaging visual device 4 down to a lower position z0, and then continuously increase it, and use the image processing method to calculate the clarity of the character area, and stop increasing when the clarity reaches a maximum value and begins to decrease (that is, the height is located after the clarity reaches the peak value), and the clarity at this time is used in step S4 to perform focusing microscopic imaging processing of the microscopic imaging visual device 4. The focus control module controls the micro-motion platform 6 to move to (x1, y1), and performs the microchip automatic focusing operation and microscopic imaging processing of the 32 site coordinates in turn according to the above method. The internal program of the focus control module sets the micro-motion platform 6 to be linked with the lifting stepper motor 3 for control.

[0052] According to one embodiment of the present invention, the neural network model method in step S3 includes the following:

[0053] S31. Build a text detection network in the neural network model. The text detection network is based on the segmentation DB text detection algorithm. First, the ResNet convolutional body is used to perform feature extraction and upsampling fusion operations to obtain a feature map. Then, the feature map is used to predict the probability map. Finally, an approximate binary map is calculated through the feature map and the probability map. The binary map represents the area where the text is located.

[0054] S32, the direction identifier in the neural network model performs a 180-degree inversion of the microchip. The microchip is placed with the characters facing upward but inverted 180 degrees in the horizontal plane, so the image of the microchip will also be inverted 180 degrees. This recognition process is performed to facilitate rapid standardization and accurate processing and recognition. The output is two results of 0 degrees or 180 degrees. If the result is 180 degrees, it will be reversed and enter step S33;

[0055] S33. In the neural network model, the text content network is built using the Rosetta OCR system and the fully convolutional model. The text content network is trained using the CTC loss function.

[0056] According to one embodiment of the present invention, in step S5, the defect detection module uses a binary image and performs defect judgment based on a threshold segmentation judgment. If a defect exists, the output defect result contains an alarm signal; the microscopic character recognition defect data method of the microscopic character recognition defect detection system includes (see Figure 3 ):

[0057] S51. If the text area is not detected or the text confidence is lower than 0.92 or the number of character bits is incorrect, the character recognition result of the microscopic character recognition defect data contains an alarm signal; for example, the alarm signal is "false, NG".

[0058] S52. If the character can be recognized but defects and / or stains are detected, the defect result of the microscopic character recognition defect data contains an alarm signal; for example, the alarm signal is "false, FD, XXXX", where "XXXX" is the recognition result.

[0059] S53. If the character can be recognized and has no defects or stains, the normal recognition signal included in the microscopic character recognition defect data is "true, XXXX".

[0060] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for microscopic character recognition and defect detection based on deep learning, characterized in that: A microscopic character recognition defect detection system is included, and the method is as follows: S1, the sample collection bar (7) is provided with N microchip attachment sites, and the N microchips are correspondingly installed on the N microchip attachment sites of the sample collection bar (7), and each microchip attachment site is correspondingly assigned an ID number; a three-dimensional focus movement system is constructed, the sample collection bar (7) is arranged on the XY axis direction movement system of the three-dimensional focus movement system, and the microscopic imaging visual device (4) is arranged on the Z axis movement system of the three-dimensional focus movement system, the microscopic imaging visual device (4) includes an area array CMOS camera and a telecentric lens (42) corresponding to the area array CMOS camera, and the area array CMOS camera is connected to the microscopic character recognition defect detection system; S2, the microscopic character recognition defect detection system comprises a focus control module, a defect detection module and a character recognition module, wherein the focus control module is connected to the XY axis direction moving system and the Z axis moving system respectively, and the focus control module drives the sample collection bar (7) to move in the X axis and Y axis directions and the microscopic imaging visual device (4) to move up and down in the Z axis direction to perform automatic focusing operation; S3. A neural network model is built in the character recognition module of the microscopic character recognition defect detection system. The neural network model includes a text detection network, a direction identifier, and a text content network. The neural network model is subjected to deep learning training for text region detection, direction recognition, and text content recognition. The neural network model method in step S3 includes the following: S31. Build a text detection network in the neural network model. The text detection network is based on the segmentation DB text detection algorithm. First, the ResNet convolutional body is used to perform feature extraction and upsampling fusion operations to obtain a feature map. Then, the feature map is used to predict the probability map. Finally, an approximate binary map is calculated through the feature map and the probability map. The binary map represents the area where the text is located. S32, the direction identifier in the neural network model performs 180-degree inversion identification of the microchip, and the output is two results of 0 degree or 180 degree. If the result is 180 degree, it will be reversed and enter step S33; S33. In the neural network model, the text content network is built using the Rosetta OCR system and the fully convolutional model. The text content network is trained using the CTC loss function. S4, the microscopic imaging visual device (4) performs focused microscopic imaging on each microchip of the sample collection strip (7) and transmits the microchip image to the microscopic character recognition defect detection system, wherein the microchip image covers at least one microchip and the ID number of the microchip attachment site; S5, the defect detection module of the microscopic character recognition defect detection system performs threshold-based defect judgment on the microchip image and obtains a defect result; the neural network model of the character recognition module performs character recognition on the microchip image and obtains a character recognition result; the microscopic character recognition defect detection system forms microscopic character recognition defect data by corresponding the defect result, the character recognition result and the ID number one by one and saves and outputs it; In step S5, the defect detection module uses the binary image and threshold segmentation to make a defect judgment. If a defect exists, the output defect result contains an alarm signal; The microscopic character recognition defect data method of the microscopic character recognition defect detection system includes: S51, if the text area is not detected or the text confidence is lower than 0.92 or the number of character bits is incorrect, the character recognition result of the microscopic character recognition defect data contains an alarm signal; S52: If the character can be recognized but defects and / or stains are detected, the defect result of the microscopic character recognition defect data contains an alarm signal.

2. A method for microscopic character recognition and defect detection based on deep learning according to claim 1, characterized in that: In step S1, the XY axis direction moving system of the three-dimensional focusing moving system comprises a horizontal platform seat (1) and a micro-motion platform (6) mounted on the horizontal platform seat (1), wherein the micro-motion platform (6) is used to drive the sample collection bar (7) to move in the horizontal plane in the X-axis or Y-axis direction; the Z axis moving system comprises a lifting stepping motor (3), wherein the lifting stepping motor (3) is used to drive the microscopic imaging visual device (4) to move in the Z axis direction.

3. A method for microscopic character recognition and defect detection based on deep learning according to claim 2, characterized in that: In step S1, the three-dimensional focus movement system comprises a vertical rod (2), a lifting stepping motor (3) is arranged on the upper part of the vertical rod (2), and a ring light source (5) corresponding to the telecentric lens (42) is installed in the middle part of the vertical rod (2) through a light source lifting cross bar (51); the height of the ring light source (5) is adjusted by lifting and lowering, and the brightness of the ring light source (5) is adjusted until the contrast between the characters of the microchip on the sample collection strip (7) and the background is prominent and the characters are clear.

4. A method for microscopic character recognition and defect detection based on deep learning according to claim 1, characterized in that: In step S1, the sample collection strip (7) is provided with 32 microchip attachment sites, and each microchip attachment site is recorded as (x1, y1), (x2, y2) ... (x32, y32) in sequence according to the site coordinates. After the focus control module in step S2 completes the focusing, the microscopic character recognition defect detection system establishes an XY axis coordinate system according to the microscopic imaging visual device (4) when the focusing is completed and corresponds to the site coordinates of the microchip attachment site; in step S4, the microscopic character recognition defect detection system controls the XY axis direction movement system to move according to the site coordinates to realize the microscopic imaging visual device (4) to perform point focusing or / and point microscopic imaging shooting on the microchip at each microchip attachment site.

Images