A wafer six-side detection automatic focusing method

Abstract

The present invention discloses an automatic focusing method for six-sided wafer inspection, comprising: correcting the position of a die using an XYR loading platform and acquiring an image of the die to be inspected; converting edge points in the edge profile of the die into geometric parameters using a Hough transform algorithm based on the image of the die to be inspected, and calculating the position compensation value of the die; determining the optimal focal position of the front / back inspection station of the die to be inspected based on the optical properties of the die, the light deflection angle, and the incident light intensity, and adjusting the focus position of a focusing module; controlling an electric slide carrying an optical camera to move into position and focus the image based on the position compensation value of the die; and evaluating and optimizing the accuracy of the model by comparing actual inspection data with the predicted results of a refractive error correction model and a die front / back focus adjustment model. The present invention effectively solves the problems of position error, optical property differences, and refractive error in traditional technologies, achieving efficient and accurate die inspection.

Description

Technical Field

[0001] The present invention relates to the field of new generation information technology, and in particular to an automatic focusing method for detecting six sides of a wafer. Background Art

[0002] With the rapid development of the semiconductor industry, wafer production and inspection have become crucial components of the production line. In the wafer manufacturing process, the die, as the fundamental unit, has a quality that directly impacts the performance and reliability of the entire chip. Therefore, efficient and accurate die quality inspection, particularly six-sided inspection and autofocus adjustment, has become a crucial issue in industrial automation. Wafer inspection systems, particularly six-sided inspection of the die, present challenges not only in image clarity but also in precisely adjusting the focus position on each surface to ensure accurate image capture at each inspection station. Due to variations in die shape, surface optical properties, and incident light angle, traditional autofocus systems often fail to maintain a consistent focus position across all inspection stations, resulting in image blur, increased detection errors, and even missed or false detections. Against this backdrop, die autofocus technology has emerged. Existing autofocus technologies primarily rely on simple focus adjustment or static position compensation. However, these methods are unable to address the effects of die position drift, optical property variations, and fluctuations in incident light intensity. Furthermore, the polycrystalline structure of the die leads to differences in surface and subsurface refractive properties, further complicating focus adjustment. Therefore, achieving efficient and accurate inspection requires precise dynamic focus adjustment technology to compensate for die position changes, optical property variations, and refractive errors in real time. Existing autofocus systems typically perform independent focus adjustments at each inspection station, but the accuracy and efficiency of this process are often limited by multiple factors. For example, light source stability, slight die position deviations, and varying refractive properties between the die surface and subsurface can all lead to focus shifts. While some machine learning-based models have been employed for focus optimization, existing methods still face the following technical challenges: First, the die position compensation process requires high-precision real-time calibration; second, the significant differences in optical properties between different dies prevents traditional focus adjustment methods from achieving global consistency; and third, the impact of refractive errors on focus is particularly pronounced in polycrystalline dies, and existing technologies are unable to accurately compensate for these deviations, resulting in insufficient inspection accuracy. Therefore, a new autofocus method is urgently needed that can dynamically adjust the focus position based on the die's optical properties, real-time position changes, and light deflection, while ensuring consistent focus accuracy across polycrystalline structures and different inspection stations. Summary of the Invention

[0003] The present invention addresses the problems existing in the above-mentioned prior art and provides a wafer six-side detection and auto-focusing method, which mainly includes:

[0004] Correct the position of the die through the loading XYR platform, and use the loading transfer module to move the die to the picking position of the second loading head to obtain the image of the die to be inspected;

[0005] According to the image of the grain to be inspected, the edge points in the edge contour of the grain are converted into geometric parameters using the Hough transform algorithm, and the position compensation value of the grain is calculated;

[0006] Determine the optimal focus position of the front / back inspection station of the die to be inspected based on the optical properties of the die, the light deflection angle, and the incident light intensity, and adjust the focus position of the focus module;

[0007] According to the position compensation value of the grain, the electric slide carrying the optical camera is controlled to move to the position and focus the image;

[0008] Evaluate and optimize model accuracy by comparing actual inspection data with predictions from the refractive error correction model and the die front / back focus adjustment model.

[0009] Furthermore, the method of correcting the position of the die by the loading XYR platform and moving the die to the picking position of the second loading head by the loading transfer module to obtain the image of the die to be inspected includes:

[0010] The grains are placed at the grain detection position by manual loading, and the optical camera at the grain detection position is used to obtain the grain image; if the grain position in the grain image is different from the preset material transfer point position, the grain position is corrected through the loading XYR platform, and the grain is lifted up by the ejector module; the grains are placed on the loading transfer module through the first loading head, and the loading transfer module is used to move the grains to the picking position of the second loading head; the loading correction camera at the picking position of the second loading head is used to obtain the image of the grain to be detected.

[0011] Furthermore, the method of converting edge points in the edge contour of the grain into geometric parameters using the Hough transform algorithm based on the grain image to be detected and calculating the position compensation value of the grain includes:

[0012] According to the image of the grain to be inspected, the Canny edge detection algorithm is used to obtain the edge contour of the grain; the edge points in the edge contour of the grain are converted into geometric parameters through the Hough transform algorithm to determine the position of the currently inspected grain; through the historical inspection records of the grain, the initial value of the correct imaging position obtained by the front / back inspection station when the grain of the same specification as the currently inspected grain is in the correct imaging position is obtained; according to the correct imaging position of the front / back inspection station of the grain, the initial value of the correct imaging position is set for the side / end inspection station, and the initial value of the correct imaging position of the front / back and the initial value of the correct imaging position of the side / end are recorded; by comparing the initial value of the correct imaging position of the front / back and the initial value of the correct imaging position of the side / end with the position of the currently inspected grain, the position compensation value of the grain is obtained.

[0013] Furthermore, the method of determining the optimal focus position of the front / back detection station of the grain to be detected based on the optical properties of the grain, the light deflection angle, and the incident light intensity, and adjusting the focus position of the focus module includes:

[0014] The second loading head sucks up the grains and places them on the inspection table. The optical properties of the grains to be inspected are obtained through real-time grain inspection records. The optical properties include surface refractive index and transparency. The optical sensor measures the incident angle and incident light intensity of the light source, obtains the relative angle between the incident light and the grain surface, and calculates the light deflection angle after the light is incident on the surface in combination with the surface refractive index of the grain, and obtains the propagation path information of the light after passing through the grain surface. Based on the optical properties of the grain, the light deflection angle and the incident light intensity, a recurrent neural network is used for model training to construct the front / back of the grain. Back focus adjustment model; based on the real-time optical properties, light deflection angle and light intensity of the grain to be inspected, the grain front / back focus adjustment model is used to determine the optimal focus position of the front / back inspection station of the grain to be inspected, and the focus position of the focus module is adjusted to obtain the front / back image of the grain; if the grain to be inspected is a polycrystalline structure, the refractive angle deviation of the grain under different incident angles and refractive indices is determined based on the surface and sub-surface optical properties of the polycrystalline structure grain, the optimal focus position of the front / back inspection station of the grain is determined, and the focus position of the focus module is adjusted.

[0015] It also includes determining the refractive angle deviation of the grain at different incident angles and refractive indices based on the optical properties of the surface and sub-surface of the polycrystalline structure grain, and determining the optimal focus position of the front / back detection station of the grain.

[0016] The method of determining the refractive angle deviation of the grains at different incident angles and refractive indices based on the optical properties of the surface and subsurface of the polycrystalline grains and determining the optimal focus position of the front / back inspection station of the grains specifically includes:

[0017] According to the optical properties of the surface and subsurface of the grain, the incident light intensity and the incident angle of the light source, Snell's law is used to obtain the refraction angle , then the refractive deviation is ,in, is the refractive index of the incident medium, is the angle of incidence, is the refractive index of the transmitting medium, is the refraction angle; based on the incident light intensity, the refraction deviation is used to correct the model , calculate the comprehensive refractive deviation of the grain , where k is a constant that is calibrated according to the optical characteristics of the system, and I is the incident light intensity; obtain the working distance and focal length of the optical camera at the current front / back inspection station, and use the focus drift formula , determine the focus offset , and according to the focus offset, control the front / back detection station focus module to adjust the focus position.

[0018] Furthermore, the method of controlling the electric slide carrying the optical camera to move to a certain position and focus the image according to the position compensation value of the grain includes:

[0019] If the die flows to the side / end face inspection station, the electric slide carrying the optical camera is controlled to move to the position and focus the image according to the position compensation value of the die to obtain the side / end face image of the die.

[0020] Furthermore, the accuracy of the model is evaluated and optimized by comparing the actual detection data with the prediction results of the refractive deviation correction model and the die front / back focus adjustment model, including:

[0021] Based on the die position information, optical properties, light deflection angle and incident light intensity recorded during the die inspection and focus adjustment process, the accuracy of the model is evaluated by comparing the actual inspection data with the prediction results of the refractive deviation correction model and the die front / back focus adjustment model. If the model's prediction result is lower than the preset accuracy threshold, the model is optimized until the prediction accuracy meets the preset requirements, and the optimized die front / back focus adjustment model and refractive deviation correction model are implemented. The optimization process includes adjusting the parameters of the die front / back focus adjustment model, improving the calculation method of optical property focus compensation, or redesigning the input features of the refractive deviation model, adding new variables or adjusting the weights of existing features.

[0022] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0023] The present invention provides an automatic focusing method for six-sided inspection of wafers. The present invention accurately corrects the position of the grain by combining a loading XYR platform, and uses a loading transfer module to ensure the precise movement of the grain in each inspection station, thereby reducing the displacement errors commonly seen in traditional manual or low-precision automated methods. After acquiring the grain image, the present invention uses the Hough transform algorithm to efficiently extract the edge contour and calculate the position compensation value, further optimizing the positioning of the grain at different inspection stations. The present invention combines the optical properties of the grain, the light deflection angle, and the intensity of the incident light to adjust the focus position in real time, ensuring clear images at the front and back inspection stations and avoiding image blur caused by focus offset. The present invention is particularly suitable for polycrystalline grains by dynamically adjusting the focus and compensating for the refractive deviation of the grain surface and subsurface, thereby improving the consistency of the focus position between different stations and ensuring the imaging accuracy of each inspection station. The present invention effectively solves the problems of position error, optical property difference, and refractive deviation in traditional technologies, and realizes efficient and accurate grain detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is a flow chart of a wafer six-side detection and auto-focusing method according to the present invention;

[0025] Figure 2 Schematic diagram of a wafer six-side detection auto-focusing method according to the present invention;

[0026] Figure 3 This is another schematic diagram of a wafer six-side detection and auto-focusing method according to the present invention. DETAILED DESCRIPTION

[0027] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0028] like Figure 1-3 In this embodiment, a wafer six-side detection and auto-focusing method may specifically include:

[0029] In step S101 , the position of the die is corrected by the loading XYR platform, and the loading transfer module is used to move the die to the picking position of the second loading head to obtain an image of the die to be inspected.

[0030] Dies are manually placed at the die inspection station, where an optical camera captures an image of the die. If the die's position in the image differs from the preset material transfer point, the XYR loading platform corrects the die's position and the ejector module lifts the die. The first loading loader places the die onto the loading transfer module, which then moves the die to the retrieving position of the second loading loader. The loading correction camera at the retrieving position of the second loading loader captures an image of the die to be inspected.

[0031] For example, during the die inspection process, a batch of die is manually loaded onto the loading platform for inspection. The die are then placed at the die inspection station, where an optical camera captures images of the die. The images reveal a slight deviation in the die's position, with the actual imaged die position differing by approximately 2 mm from the preset material transfer point. Therefore, to correct the die's position, the loading XYR platform is used to align the die precisely with the preset transfer point. The XYR platform then uses precise motion adjustments to ensure the horizontal and angular position of the die is corrected to submillimeter accuracy. Next, a pin module is used to lift the die, ensuring its stable position on the platform and preparing it for subsequent retrieval. Once lifted, the first loading pin accurately places the die onto the loading transfer module, ensuring that the die does not tilt or move improperly during transport. The loading transfer module then smoothly moves the die from its current position to the retrieval position of the second loading pin, ensuring that the die is not damaged during transport. When the die reaches the pick-up position of the second loading pin, the loading and deflection correction camera at the pick-up position takes another image of the die to confirm its correct positioning. If the image shows a slight deviation from the ideal position, say 1.5 mm, this means that despite the calibration by the XYR stage and ejector module, the die still needs slight repositioning. This further calibration ensures that the die is in the ideal position, ready for subsequent inspection and analysis.

[0032] Step S102 : Based on the image of the grain to be inspected, the edge points in the edge contour of the grain are converted into geometric parameters using the Hough transform algorithm, and the position compensation value of the grain is calculated.

[0033] Based on the image of the grain to be inspected, the Canny edge detection algorithm is used to obtain the edge contour of the grain. The Hough transform algorithm is used to convert the edge points in the edge contour of the grain into geometric parameters to determine the position of the currently inspected grain. The historical inspection records of the grain are used to obtain the initial value of the correct imaging position obtained by the front / back inspection station when the grain of the same specification as the currently inspected grain is in the correct imaging position. Based on the correct imaging position of the front / back inspection station of the grain, the initial value of the correct imaging position is set for the side / end inspection station, and the initial value of the correct imaging position of the front / back and the initial value of the correct imaging position of the side / end are recorded. By comparing the initial value of the correct imaging position of the front / back and the initial value of the correct imaging position of the side / end with the position of the currently inspected grain, the position compensation value of the grain is obtained.

[0034] For example, in the actual operation of grain inspection, the grain is placed at the grain inspection location by manual loading, and the optical camera at this location is used to obtain the image data of the grain. After the image is acquired, the Canny edge detection algorithm is used to extract the edge contour of the grain, and the four boundaries of the grain are successfully detected. The edge points in the grain edge contour are converted into geometric parameters through the Hough transform algorithm, and then the actual position of the grain in the image is determined. The four boundary position data of the grain at the current inspection station are obtained as follows: side a' = 310mm, side b' = 120mm, side c' = 320mm, and side d' = 130mm. By consulting the historical inspection records of the grains, the standard position of this type of grain in the past with the same specifications is obtained. It is found that under normal circumstances, the standard imaging position at the front inspection station is a = 300mm, b = 100mm, and the standard imaging position at the back inspection station is c = 310mm, d = 110mm. Based on this standard data, the system further sets the correct initial imaging positions for the side / end inspection stations. After analysis and calculation, the initial positions of the side inspection station are set to a = 400mm, b = 150mm, and the initial positions of the end inspection station are set to c = 420mm, d = 170mm. After obtaining the actual position of the current inspection die, these actual positions are compared with the initial standard positions to calculate the position compensation value for each inspection station. The calculated compensation values ​​are: a - a' = 300mm - 310mm = -10mm, b - b' = 100mm - 120mm = -20mm. Ultimately, these position compensation values ​​indicate that the current front / back inspection position of the die is offset by 10mm in the X axis and 20mm in the Y axis from the standard position, requiring corresponding adjustments at the side / end inspection station.

[0035] Step S103 , determining the optimal focus position of the front / back inspection station of the die to be inspected according to the optical properties of the die, the light deflection angle, and the incident light intensity, and adjusting the focus position of the focus module.

[0036] The grains are sucked up by the second loading head and placed on the inspection table. The optical properties of the grains to be inspected are obtained through real-time grain inspection records. The optical properties include surface refractive index and transparency. The incident angle and incident light intensity of the light source are measured by an optical sensor to obtain the relative angle between the incident light and the grain surface. Combined with the surface refractive index of the grain, the angle of deflection of the light after it is incident on the surface is calculated to obtain the propagation path information of the light after passing through the grain surface. Based on the optical properties of the grains, the angle of deflection of the light and the intensity of the incident light, a recurrent neural network is used for model training to construct a grain front / back focus adjustment model. Based on the optical properties, light deflection angle and light intensity of the grains to be inspected obtained in real time, the grain front / back focus adjustment model is used to determine the optimal focus position of the front / back inspection station of the grain to be inspected, and the focus position of the focus module is adjusted to obtain the front / back image of the grain. If the grain to be inspected is a polycrystalline structure, the refractive angle deviation of the grain under different incident angles and refractive indices is determined based on the surface and sub-surface optical properties of the polycrystalline grain, the optimal focus position of the front / back inspection station of the grain is determined, and the focus position of the focus module is adjusted.

[0037] For example, during a grain inspection process, a second loading head is used to pick up a grain from the loading platform and accurately place it on the inspection table. Through real-time grain inspection records, the optical properties of the grain are automatically obtained, including a surface refractive index of 1.5 and a transparency of 80%. The optical sensor installed in the system measures the incident angle of the light source to be 15 degrees, and the incident light intensity is 500 lux. According to the incident angle of the light source and the refractive index of the grain surface, Snell's law is used to calculate that the deflection angle of the incident light after entering the grain surface is 10 degrees, and the propagation path of the light after passing through the grain surface is determined by the image system. A recurrent neural network is used for model training, and a grain front / back focus adjustment model is constructed based on a large amount of historical data. The model can predict the optimal focus position based on the optical properties of the grain to be inspected, the light deflection angle and the incident light intensity obtained in real time. The current die is a standard silicon substrate. Real-time data is fed into the model, which determines the optimal focus position for the front inspection station at 200μm from the die surface and 180μm for the back inspection station. Focus modules are precisely adjusted to achieve the optimal focus for both the front and back inspection stations. Images of the front and back sides of the die are captured to ensure no defocus or blurring. If the die to be inspected is polycrystalline, meaning its surface and subsurface have different optical properties, their surface and subsurface structures, particularly for thin films, alloys, or composite materials, may exhibit polycrystalline structures. For example, a surface refractive index of 1.5, a subsurface refractive index of 1.4, and a surface-to-subsurface angle of 30 degrees are used. Refraction angle correction is then applied based on these different refractive indices and the angle of incidence. The front and back focus positions are then recalculated based on the polycrystalline structure of the die. The focus module is then adjusted to compensate for the refractive deviation caused by the subsurface, ultimately ensuring optimal focus at each inspection station.

[0038] Among them, according to the optical properties of the surface and subsurface of the polycrystalline structure grain, the refractive angle deviation of the grain under different incident angles and refractive indices is determined, and the optimal focus position of the front / back detection station of the grain is determined.

[0039] According to the optical properties of the surface and subsurface of the grain, the incident light intensity and the incident angle of the light source, Snell's law is used to obtain the refraction angle , then the refractive deviation is ,in, is the refractive index of the incident medium, is the angle of incidence, is the refractive index of the transmitting medium, is the refraction angle. Based on the incident light intensity, the refraction deviation is used to correct the model. , calculate the comprehensive refractive deviation of the grain , where k is a constant that is calibrated according to the optical characteristics of the system, and I is the incident light intensity. Get the working distance and focal length of the optical camera at the current front / back inspection station, and use the focus drift formula , determine the focus offset , and according to the focus offset, control the front / back detection station focus module to adjust the focus position.

[0040] For example, during a certain die inspection process, the incident angle of the light source incident on the die is The angle of incidence is 30° and the incident light intensity is 800 lux. The refractive index of the grain surface is is 1.5, the refractive index of the subsurface is 1.4, and Snell's law is used to calculate the refraction angle of the grain surface ,get , then the refractive deviation Based on the incident light intensity of 800 lux, the refractive error correction model is used. , calculate the comprehensive refractive deviation of the grain , where k is a constant that is calibrated according to the optical characteristics of the system, and I is the incident light intensity. If the focal length of the die is f = 10 mm, the focus drift formula is used , get the focus offset Based on the calculated focus offset Δf = 0.429mm, the autofocus system needs to adjust the focus position by approximately 0.429mm to ensure a clear image.

[0041] Step S104 , according to the position compensation value of the die, controlling the electric slide carrying the optical camera to move to the desired position and focus the image.

[0042] If the die flows to the side / end face inspection station, the electric slide carrying the optical camera is controlled to move to the position and focus the image according to the position compensation value of the die to obtain the side / end face image of the die.

[0043] For example, during a grain inspection process, when the grain flows to the side / end inspection station and obtains the initial position compensation value of 2 mm, the electric slide starts to move, carrying the optical camera to adjust to the correct position according to the compensation value. If the ideal position of the side inspection station is (100, 150), and the grain is adjusted to (102, 152) due to the need for initial compensation, the electric slide accurately controls the camera to move 2 mm to the left and 2 mm upward to ensure that the camera is accurately aligned with the side of the grain. After the adjustment is completed, the system further adjusts the focus. If the focus offset has been obtained in the initial calculation and the offset is 0.5 mm. At this time, the electric slide will further adjust the camera focal length according to the calculated focus offset to ensure that every detail of the grain image can be clearly presented when imaging the side / end surface. After the focus adjustment, the optical camera successfully obtains the side / end surface image of the grain.

[0044] Step S105 , evaluating and optimizing the accuracy of the model by comparing the actual detection data with the prediction results of the refractive deviation correction model and the die front / back focus adjustment model.

[0045] Based on the die position information, optical properties, light deflection angle, and incident light intensity recorded during the die inspection and focus adjustment process, the model accuracy is evaluated by comparing the actual inspection data with the prediction results of the refractive error correction model and the die front / back focus adjustment model. If the model's prediction results fall below the preset accuracy threshold, the model is optimized until the prediction accuracy meets the preset requirements, and the optimized die front / back focus adjustment model and refractive error correction model are implemented. The optimization process may include adjusting the parameters of the die front / back focus adjustment model, improving the calculation method for optical property focus compensation, or redesigning the input features of the refractive error model by adding new variables or adjusting the weights of existing features.

[0046] For example, during the front and back focus adjustment and refractive deviation correction process of the grain, the position, optical properties, light deflection angle and incident light intensity of the grain are recorded through real-time detection, and the surface refractive index of the grain to be detected is 1.5, the incident angle of the incident light is 25 degrees, and the incident light intensity is 600 lux. Based on these input data, the front / back focus adjustment model of the grain and the refractive deviation correction model are used for prediction, and it is calculated that the ideal focus position of the front detection station is 150μm, and the ideal focus position of the back detection station is 155μm. The actual grain image data is obtained, which shows that the focus position of the current grain at the front detection station is 160μm, and the focus position of the back detection station is 165μm. By comparing the actual detection data and the predicted results, it is found that the focus positions of the front and back sides have a deviation of 5μm. To ensure imaging accuracy, the model's accuracy is evaluated based on this deviation. If the current model's predictions exceed the preset accuracy threshold of 2μm, model optimization is required. This optimization process involves adjusting parameters in the die front / back focus adjustment model. For example, by changing the optical property weights used in the focus adjustment process or refining the calculation method for optical property focus compensation, the front and back focus positions of the die are more accurately determined. Furthermore, the refractive error model is reevaluated. Taking into account the influence of light deflection angle and incident light intensity, the optimization process determines whether the input features of the refractive error model should be adjusted, including a correction factor for refractive error based on incident light intensity, to improve prediction accuracy. Following these optimizations, the model is retrained and the optimized predictions are compared with the new actual inspection data. The final front focus position is 150.2μm, and the back focus position is 155.1μm, meeting the preset accuracy threshold. By implementing the optimized die front / back focus adjustment model and refractive error correction model, focus adjustment for each die inspection is more accurate and consistent, thereby improving the efficiency and accuracy of the entire inspection process.

[0047] The above description is merely a preferred embodiment of the present application and an illustration of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to the technical solutions formed by the specific combination of the above-mentioned technical features, but also encompasses other technical solutions formed by any combination of the above-mentioned technical features or their equivalents without departing from the concept of this application. For example, a technical solution formed by replacing the above-mentioned features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A wafer six-side detection autofocus method, characterized in that: The method comprises: Correct the position of the die through the loading XYR platform, and use the loading transfer module to move the die to the picking position of the second loading head to obtain the image of the die to be inspected; According to the image of the grain to be inspected, the edge points in the edge contour of the grain are converted into geometric parameters using the Hough transform algorithm, and the position compensation value of the grain is calculated; Determine the optimal focus position of the front / back inspection station of the die to be inspected based on the optical properties of the die, the light deflection angle, and the incident light intensity, and adjust the focus position of the focus module; According to the position compensation value of the grain, the electric slide carrying the optical camera is controlled to move to the position and focus the image; Evaluate and optimize model accuracy by comparing actual inspection data with the predictions of the refractive error correction model and the die front / back focus adjustment model.

2. The method according to claim 1, wherein The method comprises: correcting the position of the die by the loading XYR platform, and using the loading transfer module to move the die to the picking position of the second loading head to obtain the image of the die to be inspected, including: The grains are placed at the grain detection position by manual loading, and the optical camera at the grain detection position is used to obtain the grain image; if the grain position in the grain image is different from the preset material transfer point position, the grain position is corrected through the loading XYR platform, and the grain is lifted up by the ejector module; the grains are placed on the loading transfer module through the first loading head, and the loading transfer module is used to move the grains to the picking position of the second loading head; the loading correction camera at the picking position of the second loading head is used to obtain the image of the grain to be detected.

3. The method according to claim 1, wherein The method of converting edge points in the edge contour of the grain into geometric parameters using the Hough transform algorithm based on the grain image to be detected and calculating the position compensation value of the grain includes: According to the image of the grain to be inspected, the edge contour of the grain is obtained using the Canny edge detection algorithm; the edge points in the edge contour of the grain are converted into geometric parameters through the Hough transform algorithm to determine the position of the currently inspected grain; through the historical inspection records of the grain, the initial value of the correct imaging position obtained by the front / back inspection station when the grain of the same specification as the currently inspected grain is in the correct imaging position is obtained; according to the correct imaging position of the front / back inspection station of the grain, the initial value of the correct imaging position is set for the side / end inspection station, and the initial value of the correct imaging position of the front / back and the initial value of the correct imaging position of the side / end are recorded; by comparing the initial value of the correct imaging position of the front / back and the initial value of the correct imaging position of the side / end with the position of the currently inspected grain, the position compensation value of the grain is obtained.

4. The method according to claim 1, wherein The method of determining the optimal focus position of the front / back detection station of the to-be-detected grain according to the optical properties of the grain, the light deflection angle, and the incident light intensity, and adjusting the focus position of the focus module, includes: The second loading head sucks up the grains and places them on the inspection table. The optical properties of the grains to be inspected are obtained through real-time grain inspection and recording. The optical properties include surface refractive index and transparency. The optical sensor measures the incident angle and incident light intensity of the light source, obtains the relative angle between the incident light and the grain surface, and calculates the light deflection angle after the light is incident on the surface in combination with the surface refractive index of the grain, and obtains the propagation path information of the light after passing through the grain surface. Based on the optical properties of the grain, the light deflection angle and the incident light intensity, a recurrent neural network is used for model training to construct the front / back view of the grain. Back focus adjustment model; based on the real-time optical properties, light deflection angle and light intensity of the grain to be inspected, the grain front / back focus adjustment model is used to determine the optimal focus position of the front / back inspection station of the grain to be inspected, and the focus position of the focus module is adjusted to obtain the front / back image of the grain; if the grain to be inspected is a polycrystalline structure, the refractive angle deviation of the grain under different incident angles and refractive indices is determined based on the surface and sub-surface optical properties of the polycrystalline structure grain, the optimal focus position of the front / back inspection station of the grain is determined, and the focus position of the focus module is adjusted.

5. The method according to claim 4, wherein The method of determining the refractive angle deviation of the grains at different incident angles and refractive indices based on the optical properties of the surface and subsurface of the polycrystalline grains and determining the optimal focus position of the front / back inspection station of the grains includes: According to the optical properties of the surface and subsurface of the grain, the incident light intensity and the incident angle of the light source, Snell's law is used to obtain the refraction angle , then the refractive error is ,in, is the refractive index of the incident medium, is the angle of incidence, is the refractive index of the transmitting medium, is the refraction angle; based on the incident light intensity, the refraction deviation is used to correct the model , calculate the comprehensive refractive deviation of the grain , where k is a constant that is calibrated according to the optical characteristics of the system, and I is the incident light intensity; obtain the working distance and focal length of the optical camera at the current front / back inspection station, and use the focus drift formula , determine the focus offset , and according to the focus offset, control the front / back detection station focus module to adjust the focus position, where f is the focal length of the grain.

6. The method according to claim 1, wherein The method of controlling the electric slide carrying the optical camera to move to a desired position and focus the image according to the position compensation value of the grain includes: If the die flows to the side / end face inspection station, the electric slide carrying the optical camera is controlled to move to the position and focus the image according to the position compensation value of the die to obtain the side / end face image of the die.

7. The method according to claim 1, wherein The accuracy of the model is evaluated and optimized by comparing the actual test data with the prediction results of the refractive deviation correction model and the die front / back focus adjustment model, including: Based on the die position information, optical properties, light deflection angle and incident light intensity recorded during the die inspection and focus adjustment process, the accuracy of the model is evaluated by comparing the actual inspection data with the prediction results of the refractive deviation correction model and the die front / back focus adjustment model. If the model's prediction result is lower than the preset accuracy threshold, the model is optimized until the prediction accuracy meets the preset requirements, and the optimized die front / back focus adjustment model and refractive deviation correction model are implemented. The optimization process includes adjusting the parameters of the die front / back focus adjustment model, improving the calculation method of optical property focus compensation, or redesigning the input features of the refractive deviation model, adding new variables or adjusting the weights of existing features.

Images