Chip ocr automatic visual recognition training method and device

Abstract

The application discloses a kind of chip OCR automatic visual identification training method and device, it belongs to OCR identification training method and equipment, it solves the problems that a kind of automatic identification chip LOT character is lacked in prior art, LOT character is cut off or cut out also can effectively identify chip OCR automatic visual identification training method and device.It mainly includes the following steps: positioning component is waiting for material, feeding, positioning component homing, camera component photographing, image analysis and output structure;S1 positioning component is waiting for material: positioning component is waiting for material in X axis direction;S2 feeding: manually place wafer on positioning component;S3 positioning component homing: positioning component with wafer is moved to the camera component below in X axis direction;S4 camera component photographing: camera in camera component focuses in Z axis direction, camera is photographed in Y axis direction.The application is mainly used for chip LOT character identification.

Description

TECHNICAL FIELD

[0001] The present application relates to an OCR recognition training method and device, in particular, especially to a chip OCR automatic visual recognition training method and device. BACKGROUND

[0002] The LOT character on the wafer is printed by photolithography technology when the wafer chip is processed, so a special device is needed to identify the one-dimensional dot matrix code to obtain the LOT information of the wafer. The chip wafer LOT character is often small in font, and the character becomes blurred after scribing, so that the naked eye may miss it and cannot be quickly compared and analyzed with the batch wafer Lot number, which is prone to errors. If the wafer LOT recognition is wrong, the smart card chip packaged with the wafer will be abnormal, resulting in very serious loss.

[0003] Before the chip is shipped, it is a complete 4-inch, 6-inch, 8-inch, or 12-inch silicon circle. If each chip is to be used independently, it needs to be independently separated along the scribing line by a water jet, but since the position and size of the chip LOT character are uncertain for each chip manufacturer, the LOT character may be cut off or removed during water jet cutting, so the dot matrix code may be missing, thus increasing the difficulty of automatic chip recognition. SUMMARY

[0004] The purpose of the present application is to provide a chip OCR automatic visual recognition training method and device to solve the problem that there is no automatic chip LOT character recognition in the prior art, and the LOT character can be effectively recognized even if it is cut off or removed.

[0005] The present application is achieved by the following technical solutions:

[0006] A chip OCR automatic visual recognition training method, comprising the following steps: positioning component waiting for material, feeding, positioning component returning to the home position, camera component photographing, image analysis, and output structure;

[0007] S1 Positioning component waiting for material: the positioning component waits for feeding in the X-axis direction;

[0008] S2 Feeding: manually placing the wafer on the positioning component;

[0009] S3 Positioning component returning to the home position: the positioning component with the wafer moves to the position below the camera component in the X-axis direction;

[0010] S4 Camera component photographing: the camera in the camera component focuses in the Z-axis direction, and the camera photographs in the Y-axis direction;

[0011] S5 image analysis: the photographed image is analyzed by PLC to establish an OCR character training library; the image analysis includes complete dot matrix character set recognition analysis and fragment sample set character splicing recognition analysis; in the fragment sample set character splicing recognition analysis, a multi-angle and multi-dimensional shooting mode is adopted to simulate the cutting mode, the characters are cut into pieces, and various types of splicing are performed using the fragment characters to form a fragment sample set, complete character conversion, and the fragment sample set can well solve the problem of inaccurate recognition caused by insufficient or missing data;

[0012] S6 output structure: output the converted result, and then repeat the S1 step.

[0013] Further, in the complete dot matrix character set recognition analysis, the camera photographs a complete character picture, which is analyzed by PLC to form a complete sample set.

[0014] Further, in the S5 image analysis, the photographed image is identified once in the forward direction and once in the reverse direction to ensure the accuracy of the identification. After repeated identification, the characters are checked and compared in the whole wafer LOT, and if the LOT has been identified or not in the LOT batch, it means that the identification is abnormal and needs manual intervention for checking and supplementing the required printed label result to generate an accurate wafer LOT number.

[0015] Further, in the complete dot matrix character set recognition analysis, an electron microscope camera is used for shooting, and MATLAB is used for the OCR character training library. Through the changes in brightness, connection, and gray scale, a multi-dimensional OCR character training library is established, i.e., the establishment of the dot matrix character set.

[0016] Further, in the fragment sample set character splicing recognition analysis, an industrial intelligent camera is used, and MATLAB is used for the OCR character training library. Based on the dot matrix character set, a straight line algorithm is used to simulate the cutting line angle, the data collected by the camera is fragmented, a batch of fragment character set pictures are obtained, a multi-dimensional OCR character training library is established, i.e., the establishment of the fragment character set, and through manual annotation and forced intervention, the fragment character pictures can be continuously expanded, the establishment of the fragment character set is realized, and thus the two deep learning models of the dot matrix character set and the fragment character set are improved, and the accuracy of the identification is further ensured.

[0017] The utility model provides an automatic visual identification training method device for using chip OCR, including frame, be equipped with controller, positioning assembly and camera frame on the frame, positioning assembly includes X axle sliding table electric jar, be equipped with fixed seat on the sliding table I of X axle sliding table electric jar, be equipped with guide pillar and carrying plate on the fixed seat, be equipped with adjustment hole on the carrying plate, be equipped with adjustment structure on the guide pillar and slide, adjustment structure is located between adjustment hole, be equipped with Z axle sliding table electric jar on the camera frame, be equipped with Y axle sliding table electric jar on the sliding table II of Z axle sliding table electric jar, be equipped with camera fixed frame on the sliding table III of Y axle sliding table electric jar, be equipped with camera on the camera fixed frame.

[0018] Further, the adjustment structure includes a limiting block, the limiting block is slidably arranged on the guide pillar, and the limiting block is fastened to the guide pillar through an adjusting pin.

[0019] Further, the limiting block is an L-shaped limiting block, and a limiting plate of the L-shaped limiting block clamps a wafer on the carrying plate.

[0020] Further, the carrying plate is provided with a taking hole, the taking hole facilitates placing and taking the wafer; and the carrying plate is provided with a magnet, and the magnet is used for attracting a wafer ring.

[0021] Further, the frame is provided with door plates and indicator lights, one of the door plates is provided with an operation display screen, and the operation display screen and the indicator lights are connected with the controller respectively.

[0022] Compared with the prior art, the utility model has the beneficial effects that:

[0023] 1. The LOT character on the wafer is automatically identified, the problem that manual operation is prone to error is solved, the qualified rate is ensured, full-size wafers are adapted, and the efficiency is improved.

[0024] 2. The OCR identification library separately screens and establishes the case that the LOT character is missing, and the accuracy of identification is ensured. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 is a structural diagram of the utility model Figure I ;

[0026] Figure 2 is a structural diagram of the utility model Figure II ;

[0027] Figure 3 is a partial structural diagram of the utility model Figure I ;

[0028] Figure 4 is a partial structural diagram of the utility model Figure II ;

[0029] Figure 5 is a structural diagram of the carrying plate of the utility model

[0030] Figure 6 is a structural schematic diagram of the adjusting structure of the present application;

[0031] Figure 7 is a flow chart of the present application.

[0032] In the figure: 1, indicator light; 2, door panel; 3, operation display screen; 4, rack; 5, camera rack; 6, X-axis sliding table electric cylinder; 7, fixed seat; 8, guide column; 9, limit block; 10, adjusting pin; 11, adjusting hole; 12, carrier plate; 13, Z-axis sliding table electric cylinder; 14, taking hole; 15, Y-axis sliding table electric cylinder; 16, wafer; 17, camera fixing frame; 18, camera. DETAILED DESCRIPTION

[0033] The present application will be further described and illustrated below in conjunction with the accompanying drawings.

[0034] Example 1, as shown in a chip OCR automatic visual recognition training method, comprising the following steps: positioning component standby, feeding, positioning component homing, camera component photographing, image analysis and output structure; Figure 7 S1 Positioning component standby: the positioning component waits for feeding in the X-axis direction;

[0035] S2 Feeding: manually place the wafer on the positioning component;

[0036] S3 Positioning component homing: the positioning component with the wafer is driven by the X-axis sliding table electric cylinder 6 to the camera component below;

[0037] S4 Camera component photographing: the camera in the camera component focuses in the Z-axis direction, and the camera photographs in the Y-axis direction;

[0038] S5 Image analysis: the photographed image is analyzed by PLC for OCR recognition, and an OCR character training library is established. The establishment of the OCR character training library is a long-term training and improvement process. Image analysis includes complete dot matrix character set recognition analysis and fragment sample set character splicing recognition analysis. In the fragment sample set character splicing recognition analysis, a multi-angle and multi-dimensional shooting mode is adopted to simulate the cutting line cutting mode, the characters are cut into pieces, and the fragment characters are used for various splicing to form a fragment sample set, complete character conversion, and the fragment sample set can well solve the disadvantages of inaccurate recognition caused by insufficient or missing data;

[0039] S6 Output structure: output the converted result, and then repeat step S1.

[0040]

[0041] ​Embodiment 2, a chip OCR automatic visual recognition training method, in the complete dot matrix character set recognition analysis, after the camera takes the complete character picture, the PLC carries out the OCR recognition analysis, forms the complete sample set; in the S5 image analysis, in order to ensure the accuracy of identification, the system is identified once, and then it is identified once. After repeated identification, the characters are checked and compared in the whole wafer lot, if the lot has appeared in the identification or not in the lot batch, it is indicated that this identification is abnormal, and manual intervention is required to check and supplement the required printed label result, so as to generate the accurate wafer lot number; in the complete dot matrix character set recognition analysis, the electron microscope camera is used for shooting, the OCR character training library adopts MATLAB, through the light change, the change of the body, the change of the gray scale, the multi-dimensional OCR character training library is realized, that is, the establishment of the dot matrix character set; in the fragment sample set character splicing recognition analysis, the industrial intelligent camera is used, the OCR character training library adopts MATLAB, on the basis of the dot matrix character set, the straight line algorithm is used to simulate the cutting line angle, the data collected by the camera is segmented into fragments, a batch of fragment character set pictures are obtained, the multi-dimensional OCR character training library is established, that is, the establishment of the fragment character set, by using these fragment character sets, artificial marking can be used to continuously expand the fragment character picture, the establishment of the fragment character set is realized, so as to perfect the two deep learning models of the dot matrix character set and the fragment character set, and the accuracy of identification is more ensured, and the rest is the same as embodiment 1.

[0042] As shown in Figure 1 Embodiment 3, a device using the chip OCR automatic visual recognition training method, as shown in FIG. 4, comprising a rack 4, the rack 4 is provided with a PLC controller, a positioning assembly and a camera rack 5, the positioning assembly comprises an X-axis sliding table electric cylinder 6, a fixed seat 7 is arranged on the sliding table I of the X-axis sliding table electric cylinder 6, a guide column 8 and a carrier plate 12 are arranged on the fixed seat 7, an adjusting hole 11 is arranged on the carrier plate 12, an adjusting structure is slidably arranged on the guide column 8, and the adjusting structure is located between the adjusting holes 11, a Z-axis sliding table electric cylinder 13 is arranged on the camera rack 5, a Y-axis sliding table electric cylinder 15 is arranged on the sliding table II of the Z-axis sliding table electric cylinder 13, a camera fixing frame 17 is arranged on the sliding table III of the Y-axis sliding table electric cylinder 15, a camera 18 is arranged on the camera fixing frame 17, the PLC controller controls and coordinates the actions of each electric cylinder, and the camera 18 selects an electron microscope camera or an industrial intelligent camera; the X-axis sliding table electric cylinder 6, the Y-axis sliding table electric cylinder 15 and the Z-axis sliding table electric cylinder 13 are matched with three sensors respectively, so as to realize the control of the positions of the camera 18 and the carrier plate 12, after the carrier plate 12 moves the wafer 16 to the position, the Z-axis automatically focuses, the Y-axis step scans the characters on the wafer 16, and the scanning is completed; the sliding table cylinder is prior art, and its structure will not be described in detail.

[0043] Embodiment 4, a device using the chip OCR automatic visual recognition training method, as shown inFigure 6 As shown, the adjusting structure includes a limiting block 9, which is slidingly arranged on the guide column 8, and is fastened to the guide column 8 by an adjusting pin 10. The position of the limiting block 9 on the guide column 8 can be adjusted by rotating the adjusting pin 10, so as to realize the positioning of wafers 16 of different sizes. The limiting block 9 is an L-shaped limiting block, and the limiting plate of the L-shaped limiting block clamps the wafer 16 on the carrier plate 12. Figure 5 As shown, the carrier plate 12 is provided with a taking hole 14, which facilitates the placing and taking of the wafer 16. The carrier plate 12 is provided with a magnet, which is used to attract the wafer ring for positioning. The rack 4 is provided with door plates 2 and an indicator lamp 1. One of the door plates 2 is provided with an operation display screen 3. The operation display screen 3 and the indicator lamp 1 are respectively connected with a controller, and the others are the same as in Embodiment 3.

[0044] Dot character recognition requires a large amount of data source data, but the same chip character data also has only part of the data, so in the establishment of the OCR character recognition process, the character training library is a long-term accumulation process, and the larger the data set base is, the higher the recognition accuracy is.

[0045] In the establishment of the OCR character training library, a group of multi-angle and multi-dimensional shooting is adopted, and an algorithm is used to simulate the cutting method of the cutting line, so as to cut the characters into pieces. The fragmented characters are used for multi-type splicing to form a fragmented sample set. The special training of the fragmented sample set can well solve the disadvantages of inaccurate recognition caused by insufficient or missing data. The double verification of the independent training of the complete sample set and the fragmented sample set guarantees that the accuracy of the OCR recognition of the equipment is as high as 99% or more.

[0046] Algorithm training:

[0047] In the first step, the algorithm test picture is shot by an electron microscope camera. The electron microscope can better capture the detailed picture of the character. Each complete character can collect a relatively high-definition picture, which is used for the OCR character training set of the complete character. The OCR character training library adopts MATLAB. On this basis, we increase the light change, the change of the connected body, the change of the gray scale, and other ways to realize the training library under the multi-dimension of the character. A large number of data sets are added in the subsequent period to continuously improve our dot character set.

[0048] Second step uses the industrial intelligent camera, the industrial intelligent camera field of vision is big, can realize the whole string character photograph, but each character detail part is slightly blurred, plus the influence of the cutting line and the position change and other space influence, often the recognition rate is not high, on the basis of the first step dot matrix character set, using straight line algorithm simulates the cutting line angle to divide the data collected by the industrial camera into fragments, gets a batch of fragment character set pictures, using these fragment character sets, supervised learning, manual annotation, forced intervention to continuously expand the fragment character pictures, realizes the establishment of the fragment character set, so as to perfect the dot matrix character set and the fragment character set two kinds of deep learning model, and more ensures the accuracy of recognition.

[0049] The operation flow of the device is: the first identification of the wafer 16 needs to manually put the wafer into the platform, select two characters manually for the first character, and the selected area will be automatically identified in the selection process. If the identification is correct, click the confirmation button, the system will automatically move the platform to identify one by one, and stop after all characters are identified. In order to ensure the accuracy of identification, the system will identify once, and then identify once in the opposite direction.

[0050] After repeated identification, the characters will be checked and compared in the whole batch of wafers LOT. If the LOT has been identified or not in the LOT batch, it means that the identification is abnormal and needs to be intervened and checked to generate accurate wafer LOT number.

[0051] After manual setting, if the first wafer identification is correct, it can enter the automatic identification mode of the same type. Only need to put the wafer into the carrier and fix it, the system will automatically identify. After the identification of all wafers of the same type is completed, all identification data and imported data are matched one by one. If there is an error in the matching process, the printed label will not be generated. Only need to manually complete the required printed label result to print. This greatly improves the operation efficiency of employees. The device can meet the identification of 4-inch, 6-inch, 8-inch and 12-inch wafers.

Claims

1. A chip OCR automatic visual recognition training method, characterized in that: The process includes the following steps: waiting for materials for the positioning components, loading materials, positioning components returning to their positions, taking pictures with the camera components, image analysis, and outputting the structure. S1 Positioning Component Waiting for Material: The positioning component is waiting to be loaded in the X-axis direction; S2 loading: The wafer is manually placed on the positioning assembly; S3 Positioning Component Return: The positioning component containing the wafer moves to below the camera component in the X-axis direction; S4 camera component takes pictures: The camera in the camera component focuses in the Z-axis direction and takes pictures in the Y-axis direction; S5 Image Analysis: The captured images are subjected to OCR recognition and analysis by PLC to establish an OCR character training library; Image analysis includes complete dot matrix character set recognition and analysis and fragmented sample set character splicing recognition and analysis. In the character splicing recognition and analysis of fragmented sample sets, a multi-angle and multi-dimensional shooting method is adopted to simulate the cutting method of slashing lines to fragment the characters. The first step uses an electron microscope to capture detailed images of the characters, which are used as the OCR character training set for the complete characters. By changing the brightness, the concatenation, and the grayscale, a multi-dimensional training library for the characters is realized, and the dataset is continuously improved to refine the dot matrix character set. The second step uses an industrial intelligent camera to capture the entire string. Based on the dot matrix character set in the first step, a straight line algorithm is used to simulate the cutting line angle to fragment the data collected by the industrial camera. By using fragmented characters to concatenate different categories, a fragmented sample set is formed, and character conversion is completed. S6 Output Structure: Output the converted result, and then repeat step S1.

2. The chip OCR automatic visual recognition training method according to claim 1, characterized in that: In the complete dot matrix character set recognition and analysis, after the camera captures a complete character image, the PLC performs OCR recognition and analysis to form a complete sample set.

3. The chip OCR automatic visual recognition training method according to claim 1, characterized in that: In S5 image analysis, to ensure the accuracy of recognition, the system performs a forward recognition once and a reverse recognition once on the image captured.

4. The chip OCR automatic visual recognition training method according to claim 1, characterized in that: The complete dot matrix character set recognition and analysis uses electron microscope camera images. The OCR character training library is built using MATLAB, and multi-dimensional OCR character training library is established by changing the brightness, connecting characters, and changing the grayscale.

5. The chip OCR automatic visual recognition training method according to claim 1, characterized in that: Industrial intelligent cameras are used in the character splicing recognition analysis of fragmented sample sets. MATLAB is used for the OCR character training library. The cutting line is simulated using a straight line algorithm to simulate the cutting line angle. The data collected by the camera is fragmented to obtain a batch of fragmented character set images, thus realizing the establishment of a multi-dimensional OCR character training library.

6. An apparatus for using the chip OCR automatic visual recognition training method according to any one of claims 1-5, characterized in that: The device includes a frame (4), on which a controller, a positioning component and a camera mount (5) are provided. The positioning component includes an X-axis slide cylinder (6), on which a fixed seat (7) is provided on slide I of the X-axis slide cylinder (6), on which a guide post (8) and a carrier plate (12) are provided. On the carrier plate (12) are adjustment holes (11), and on which an adjustment structure is slidably provided on the guide post (8) and located between the adjustment holes (11). On the camera mount (5) is a Z-axis slide cylinder (13), on which a Y-axis slide cylinder (15) is provided on slide II of the Z-axis slide cylinder (13), and on which a camera mount (17) is provided on slide III of the Y-axis slide cylinder (15), and on which a camera (18) is provided.

7. The chip OCR automatic visual recognition training method according to claim 6, characterized in that: The adjustment structure includes a limiting block (9), which is slidably disposed on the guide post (8) and is fastened to the guide post (8) by an adjusting pin (10).

8. The chip OCR automatic visual recognition training method according to claim 7, characterized in that: The limiting block (9) is an L-shaped limiting block.

9. The chip OCR automatic visual recognition training method according to claim 6, characterized in that: The carrier plate (12) is provided with a pick-up hole (14), which facilitates the placement and pick-up of the wafer (16); the carrier plate (12) is provided with a magnet, which is used to hold the wafer ring for positioning.

10. The chip OCR automatic visual recognition training method according to claim 6, characterized in that: The frame (4) is provided with a door panel (2) and an indicator light (1). The door panel (2) is provided with an operation display screen (3). The operation display screen (3) and the indicator light (1) are respectively connected to the controller.

Images