Visual Inspection Method for Solar Silicon Wafers

Abstract

The present invention discloses a visual inspection method for solar wafers, specifically relating to the technical field of wafer inspection; by acquiring optical images at different incident angles and synchronously collecting infrared transmission images, performing normalization processing, edge enhancement filtering and adaptive threshold segmentation on the collected multi-angle images, extracting suspected crack regions, respectively calculating the gray gradient change rate and the stress anomaly degree, comprehensively calculating the gray gradient change rate and the stress anomaly degree through a weighted scoring method, calculating the crack credibility score, and comparing it with a preset crack detection threshold to distinguish true hidden cracks from stress pseudo-defects; if the credibility score is lower than the threshold, secondary sampling is performed by adjusting the illumination angle to further optimize the determination result; the present invention effectively reduces the misjudgment rate caused by misjudgment of lattice stress pseudo-defects, and at the same time avoids the missed detection of true hidden cracks, improves the accuracy, stability and production yield of wafer defect detection, thereby enhancing the long-term reliability of photovoltaic modules.

Description

Technical Field

[0001] The present invention relates to the technical field of silicon wafer detection, and particularly to a visual detection method for solar silicon wafers. Background Art

[0002] With the rapid development of the photovoltaic industry, as the core material of solar cells, the quality of solar silicon wafers directly affects the photoelectric conversion efficiency and service life of the cells. During the manufacturing process of silicon wafers, various defects may occur, such as cracks, hidden cracks, chamfer defects, surface contamination, uneven etching, etc. Therefore, efficient and accurate visual inspection of silicon wafers has become a key link to improve product quality and reduce production costs.

[0003] Traditional silicon wafer detection methods mainly include manual visual inspection, laser scattering detection, ultrasonic detection, and machine vision detection. Among them, manual visual inspection has low efficiency and is easily affected by subjective factors, while although laser scattering detection and ultrasonic detection have high precision, their detection speed is slow, making it difficult to meet the high-speed detection requirements of modern production lines. Therefore, automated detection methods based on machine vision have gradually become the mainstream.

[0004] The existing technology has the following deficiencies:

[0005] During the visual inspection process of solar silicon wafers, there may be confusion between hidden cracks and lattice stress-induced pseudo-defects. A hidden crack refers to a tiny crack inside the silicon wafer, which may not immediately affect the cell performance, but may expand into a serious crack during subsequent cell encapsulation, welding, or long-term use, resulting in cell failure. And the pseudo-defects induced by lattice stress are due to the silicon wafer being subjected to thermal stress or mechanical stress during the manufacturing process, causing local lattice deformation or refractive index change, and showing a crack-like structure under specific lighting conditions.

[0006] Current visual inspection systems mainly rely on deep learning or image processing techniques to detect cracks. However, due to the high similarity of visual features between hidden cracks and lattice stress pseudo-defects (such as morphology, edge features), the detection system is prone to misjudgment. If the system misidentifies lattice stress pseudo-defects as hidden cracks, it may lead to an increase in the mis-rejection rate and affect the production yield; while if the system fails to accurately identify real hidden cracks, it may release potential defective products, ultimately affecting the long-term reliability of the components. Summary of the Invention

[0007] The purpose of the present invention is to provide a visual detection method for solar silicon wafers to solve the deficiencies in the background art.

[0008] To achieve the above purpose, the present invention provides the following technical solution: A visual detection method for solar silicon wafers, including:

[0009] Illuminate a solar silicon wafer with a multi - angle lighting device to generate optical images with multiple different incident angles, obtain the original detection images of the silicon wafer at multiple lighting angles, and simultaneously collect infrared transmission images;

[0010] Normalize the collected multi - angle images, enhance the image features through edge - enhancement filtering, and perform binarization processing using an adaptive threshold method to extract suspected crack regions existing on the silicon wafer surface;

[0011] Calculate the gray - level gradient change rate of the suspected crack regions at multiple lighting angles respectively, and calculate the stress anomaly degree of the suspected crack regions based on the stress distribution inside the silicon wafer through infrared transmission data;

[0012] Combining the calculation results of the gray - level gradient change rate and the stress anomaly degree, generate a crack credibility score using a weighted scoring method and compare it with the crack detection threshold. If the crack credibility score is higher than the crack detection threshold, it is determined as a hidden crack defect, and the defect location and characteristics are recorded; if the crack credibility score is lower than the crack detection threshold, adjust the lighting angle for secondary sampling to further confirm the defect type.

[0013] Preferably, the multi - angle lighting device includes multiple adjustable - angle LED light - source arrays, and the incident - angle range of the LED light - source arrays is , where: low - angle incident light is used to enhance the visibility of surface micro - scratches and shallow cracks; medium - angle incident light is used to provide uniform illumination and is suitable for overall defect detection; high - angle incident light is used to enhance the optical contrast of deep cracks and hidden - crack regions.

[0014] Preferably, calculate the gray - level gradient change rate of the suspected crack regions at multiple lighting angles respectively. The method for obtaining the gray - level gradient change rate is as follows:

[0015] The input original image is an RGB color image, which is converted into a gray - level image I(x, y). The Sobel operator is used to calculate the gray - level gradients of the image in the horizontal and vertical directions, using two 3×3 convolution kernels: horizontal - direction gradient: ; vertical - direction gradient: ; perform convolution calculation on the image, and calculate and respectively: ; ; In the formula, represents the convolution operation; is the gradient in the X direction; is the gradient in the Y direction. Combine and , calculate the gradient magnitude of each pixel: ; where: G(x,y) represents the total gradient magnitude of the pixel. Under multiple illumination angles, the gradient change rate of the crack area is estimated by the gradient change of adjacent pixels: ; where: represents the gray gradient change rate.

[0016] Preferably, according to the stress distribution inside the silicon wafer, calculate the stress anomaly degree of the suspected crack area through infrared transmission data, specifically including: collecting the polarized infrared transmission image IIR(x,y), where IIR(x, y) represents the polarized infrared transmission light intensity at the image coordinates (x, y), and normalizing the image to obtain the normalized image , which is defined as: ; where, and are the minimum and maximum gray values of the original image respectively;

[0017] Due to the birefringence effect of polarized light caused by the internal stress of the silicon wafer, calculate the phase delay at each pixel: ; where: δ(x,y) is the phase delay, d(x,y) is the local thickness of the silicon wafer; Δn(x,y) is the refractive index change, which is proportional to the stress; λ is the infrared light wavelength, and the stress σ(x,y) and the phase delay δ(x,y) satisfy: ; where, is the photoelastic coefficient of the silicon wafer;

[0018] Perform Fourier transform on the polarized infrared image to identify the stress concentration area F(u,v). The low-frequency component represents the uniform stress distribution, and the high-frequency component represents the stress mutation area;

[0019] Extract the high-frequency stress anomaly area through a band-pass filter: ; where H(u,v) is the band-pass filter: ; 、 are the low-frequency and high-frequency cut-off values; perform inverse Fourier transform to restore the stress anomaly area , calculate the standard deviation of the stress anomaly distribution : ; where: is the number of pixels in the suspected crack area; is the average stress value of the area;

[0020] Calculate the stress change rate : ; Calculate the stress anomaly degree value , and the expression is: ; where: is the weighting coefficient, satisfying 。

[0021] Preferably, according to the real-time factor, combining the calculation results of the gray gradient change rate and the stress anomaly degree, a weighted scoring method is used to generate a crack credibility score, specifically including:

[0022] Normalize the gray gradient change rate and the stress anomaly degree value so that they are both within [0,1], and calculate the crack credibility score according to the normalized gray gradient change rate and stress anomaly degree value.

[0023] Preferably, adjust the illumination angle for secondary sampling to further confirm the defect type, specifically including:

[0024] In order to enhance the visibility of the crack or defect area, a dynamic illumination angle adjustment method is adopted: ; where: is the adjusted illumination angle, is the original sampling illumination angle; is the angle adjustment step size;

[0025] At the new illumination angle recalculate the gray gradient change rate: ; where: is the normalized gray gradient change rate at the new angle; 、 are the gray gradients in the X and Y directions calculated by the Sobel operator.

[0026] Preferably, after adjusting the illumination angle, use polarized infrared transmission detection to calculate the stress anomaly degree at the new angle: ; where: is the normalized stress anomaly degree at the new illumination angle, is the stress anomaly degree calculated at the new angle; is still the global normalization parameter.

[0027] Preferably, according to the results of the secondary sampling, recalculate the crack credibility score, compare the new crack credibility score with the crack detection threshold, and if the new crack credibility score is lower than the crack detection threshold, it is determined as a non-crack area.

[0028] In the above technical solution, the technical effects and advantages provided by the present invention:

[0029] 1. The present invention effectively solves the problem in the prior art that it is difficult to distinguish between hidden cracks and pseudo-defects induced by lattice stress by integrating multi-angle optical imaging and polarized infrared transmission analysis. By adopting a multi-angle illumination system, optical imaging at different incident angles enhances the visibility of different types of cracks, while infrared transmission imaging further provides stress information inside the silicon wafer, improving the accuracy of hidden crack detection. The present invention extracts the suspected crack regions on the silicon wafer surface through normalization processing, edge enhancement filtering, and adaptive threshold segmentation, calculates the gray gradient change rate at multiple illumination angles respectively, calculates the stress anomaly degree in combination with the infrared transmission data, generates a crack credibility score through a weighted scoring method, and performs secondary sampling optimization determination based on a dynamic light adjustment mechanism.

[0030] 2. The present invention has higher detection robustness and reliability. Multi-angle illumination avoids false detection or missed detection caused by the limitations of single-angle imaging, and infrared transmission analysis makes up for the problem of internal defects that cannot be recognized by traditional visual inspection. The combined calculation of the gray gradient change rate and the stress anomaly degree makes crack identification more accurate. In addition, the present invention adopts intelligent dynamic light adjustment, automatically optimizes the detection angle when the crack credibility score is close to the threshold, improves the accuracy of defect identification, reduces the false rejection rate and missed detection rate, thereby improving the quality control level of solar silicon wafers, enhancing the long-term reliability of photovoltaic modules, and being of great significance for the efficient production and cost control of the photovoltaic industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required to be used in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0032] Figure 1 It is a flowchart of the method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0033] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0034] Embodiment. Please refer to Figure 1 As shown, the visual inspection method for solar silicon wafers in this embodiment includes:

[0035] Illuminate the solar silicon wafer with a multi-angle lighting device to generate optical images at multiple different incident angles, obtain the original detection images of the wafer at multiple lighting angles, and synchronously collect infrared transmission images;

[0036] Normalize the collected multi-angle images, enhance the image features through edge enhancement filtering, and perform binary processing using an adaptive threshold method to extract the suspected crack regions existing on the wafer surface;

[0037] Calculate the gray-scale gradient change rate of the suspected crack regions at multiple lighting angles respectively, and calculate the stress anomaly degree of the suspected crack regions based on the stress distribution inside the wafer through the infrared transmission data;

[0038] Combining the calculation results of the gray-scale gradient change rate and the stress anomaly degree, generate a crack credibility score using a weighted scoring method, and compare it with the crack detection threshold. If the crack credibility score is higher than the crack detection threshold, it is determined as a hidden crack defect, and the defect position and characteristics are recorded; if the crack credibility score is lower than the crack detection threshold, adjust the lighting angle for secondary sampling to further confirm the defect type.

[0039] For the detection of hidden cracks and defects in solar silicon wafers, a combination method of multi-angle lighting + infrared transmission imaging is adopted to improve the detection accuracy and robustness. Specifically: The multi-angle lighting system includes the following components:

[0040] Multiple adjustable-angle LED light source arrays: Configured above the wafer to irradiate the wafer surface at different angles; Programmable lighting control unit: Dynamically adjust the brightness and incident angle of the light source according to the reflection characteristics of the wafer surface to optimize the defect imaging effect;

[0041] Principle of multi-angle optical imaging: Low-angle incident light (10° - 30°): Enhance the visibility of surface micro-scratches and shallow cracks; Medium-angle incident light (30° - 60°): Provide uniform illumination, suitable for overall defect detection; High-angle incident light (60° - 90°): Enhance the optical contrast of deep cracks and hidden crack regions.

[0042] Collect the surface images of the wafer at different lighting angles through a high-resolution industrial camera and store them as multi-channel data. The specific process is as follows:

[0043] Synchronously control the light source and the camera, and sequentially collect the original images at each lighting angle;

[0044] Automatic image calibration to ensure the spatial alignment of multi-angle images and avoid detection deviation caused by equipment errors;

[0045] Preliminary defect region extraction, using edge enhancement processing and adaptive contrast adjustment to improve the visibility of defect features.

[0046] Synchronous acquisition of infrared transmission images:

[0047] Infrared transmission detection is an important innovation of this method and can be used to detect subsurface cracks or stress anomaly regions. By using a short-wave infrared (SWIR, 900 - 1700 nm) or mid-wave infrared (MWIR, 3 - 5 μm) light source, the silicon wafer is irradiated to obtain internal defect information;

[0048] The transmitted image is recorded by an infrared camera and fused with the visible light image for analysis; Preprocessing of infrared images: denoising, contrast enhancement, and edge detection to extract stress concentration regions and crack information.

[0049] Fusing the optical image and the infrared image for joint analysis to improve the detection accuracy; Using infrared transmission data to correct misjudgments in optical images and reduce interference caused by surface texture or contamination; Adopting the method of threshold comparison + stress analysis to ensure accurate distinction between cracks and pseudo-defects.

[0050] In this embodiment, for the multi-angle optical images and infrared transmission images collected, through normalization processing, edge enhancement filtering, and adaptive threshold binarization, the suspected crack regions on the silicon wafer surface are extracted, specifically:

[0051] Eliminating the problem of uneven image contrast caused by illumination angle, brightness change, or equipment noise to ensure the consistency of image data at different angles.

[0052] Calculating the average gray value of all collected images and performing normalization adjustment on each image I(x, y) so that its gray value range is between [0, 1]: ; where, and are the minimum and maximum gray values of the original image respectively.

[0053] Edge enhancement filtering processing: highlighting the edge features of the crack region to improve the accuracy of crack detection.

[0054] High-pass filtering to enhance crack edges: Using the Laplacian Filter to calculate the second derivative to enhance crack edges. This operation can strengthen crack edges while reducing surface noise interference.

[0055] Adaptive threshold binarization processing: Extracting the suspected crack regions on the silicon wafer surface and converting the image into a black-and-white binary image for subsequent analysis.

[0056] Local adaptive threshold (Adaptive Thresholding): Due to the uneven illumination on the silicon wafer surface, a local window adaptive method is used to calculate the binarization threshold: ; where, is the average gray value within the local window, σ(x, y) is the local standard deviation, and k is an adjustment parameter that controls the threshold sensitivity (usually taken as 0.5 - 1.0). If Ienhance(x, y) > T(x, y), it is set to white (crack area); otherwise, it is set to black (background).

[0057] Final output result: Generate a binary crack image, where the crack area is white and the background is black; extract the connected crack regions, calculate characteristic parameters such as crack length and width, providing a basis for subsequent analysis; combine with the infrared transmission image to further exclude false defects and ensure the accuracy of detection.

[0058] Calculate the gray gradient change rate of the suspected crack area at multiple illumination angles. The method for obtaining the gray gradient change rate is as follows:

[0059] The input original image is usually an RGB color image, which needs to be first converted to a gray image I(x, y). The calculation formula is: I(x, y) = 0.299R(x, y) + 0.587G(x, y) + 0.114B(x, y); where R, G, and B are the red, green, and blue channel values of the image pixels respectively.

[0060] The Sobel operator is used to calculate the gray gradients of the image in the horizontal (X - direction) and vertical (Y - direction). The following two 3×3 convolution kernels are used: Horizontal - direction gradient: ; Vertical - direction gradient: ; Perform convolution calculation on the image to separately obtain and : ; ; In the formula, represents the convolution operation; is the gradient in the X - direction; is the gradient in the Y - direction. Combine and to calculate the gradient magnitude of each pixel point: ; where: G(x, y) represents the total gradient magnitude of the pixel point, which can be used to measure the gray change rate of the crack area.

[0061] At multiple illumination angles, the gradient change rate of the crack area is estimated by the gradient change of adjacent pixel points: ; where: represents the gray gradient change rate; calculate the difference in the gradient values of adjacent pixel points to measure the degree of gradient mutation in the crack area.

[0062] According to the stress distribution inside the silicon wafer, calculate the stress anomaly degree of the suspected crack area through infrared transmission data, specifically including:

[0063] Polarized infrared imaging is used to detect the internal stress of silicon wafers. Due to the photoelastic effect, the silicon wafer will generate a phase delay under the action of internal stress, which affects the transmission characteristics of polarized light.

[0064] Set an infrared light source and a polarizer, and adjust the polarization direction to enhance the contrast of the stress area;

[0065] Collect the polarized infrared transmission image IIR(x,y), where IIR(x, y) represents the intensity of polarized infrared transmitted light at the image coordinates (x, y), which is affected by the internal stress distribution of the silicon wafer; to reduce the equipment response error, normalize the image to obtain the normalized image , which is defined as: ; where, and are the minimum and maximum gray values of the original image, respectively.

[0066] Due to the birefringence effect of polarized light caused by the internal stress of the silicon wafer, calculate the phase delay at each pixel: ; where: δ(x,y) is the phase delay, d(x,y) is the local thickness of the silicon wafer; Δn(x,y) is the refractive index change, which is proportional to the stress; λ is the infrared light wavelength (usually in the range of 900 - 1700nm). The stress σ(x,y) and the phase delay δ(x,y) satisfy: ; where, is the photoelastic coefficient of the silicon wafer.

[0067] Perform a Fourier transform on the polarized infrared image. The Fourier transform is used to extract the frequency characteristics of the stress distribution and identify the stress concentration area F(u,v). The low-frequency components represent the uniform stress distribution, and the high-frequency components represent the stress mutation area.

[0068] Extract the high-frequency stress anomaly area through a band-pass filter: ; where H(u,v) is the band-pass filter: ; , are the low-frequency and high-frequency cut-off values. Perform an inverse Fourier transform to restore the stress anomaly area , and calculate the standard deviation of the stress anomaly distribution : ; where: is the number of pixels in the suspected crack area; is the average stress value of the area.

[0069] Calculate the stress change rate : ; if is higher than the set threshold, it indicates that the stress in this area has mutated and there may be cracks. Calculate the stress anomaly degree value , the expression is: ; where: is the weighting coefficient, usually satisfying ; if is higher than the set threshold, then this area is a stress anomaly area.

[0070] Combining the calculation results of the gray gradient change rate and the stress anomaly degree, a weighted scoring method is used to generate the crack credibility score, specifically including:

[0071] Normalize the gray gradient change rate and the stress anomaly degree value so that they are both in the range of [0, 1], and calculate the crack credibility score according to the normalized gray gradient change rate and stress anomaly degree value.

[0072] For example, the present invention can use the following formula to calculate the crack credibility score, and the calculation expression is: ; in the formula, is the crack credibility score, represents the gray gradient change rate, is the stress anomaly degree value, are the weight coefficients of the gray gradient change rate and the stress anomaly degree value (which can be optimized according to experimental experience or machine learning), and are all greater than 0.

[0073] Compare the obtained crack credibility score with the crack detection threshold. If the crack credibility score is higher than the crack detection threshold, it is determined as a hidden crack defect, and the defect location and characteristics are recorded; if the crack credibility score is lower than the crack detection threshold, adjust the illumination angle for secondary sampling to further confirm the defect type.

[0074] Adjust the illumination angle for secondary sampling to further confirm the defect type, specifically including:

[0075] In order to enhance the visibility of the crack or defect area, a dynamic illumination angle adjustment method is adopted: ; where: is the adjusted illumination angle, is the original sampling illumination angle; is the angle adjustment step size. If the crack credibility score is low, increase the incident angle to improve the shadow contrast of the crack area; if stress anomaly is detected but there is no obvious crack, reduce the illumination angle to reduce the surface reflection interference.

[0076] At the new illumination angle , recalculate the gray gradient change rate: ; where: is the normalized gray gradient change rate at the new angle; , The gray - scale gradients in the X and Y directions calculated by the Sobel operator (at the new illumination angle). If increases, it indicates that this area may be a crack; if decreases, it may be noise or surface texture changes.

[0077] After adjusting the illumination angle, use polarized infrared transmission detection to calculate the degree of stress anomaly at the new angle: ; where: is the normalized degree of stress anomaly at the new illumination angle, is the degree of stress anomaly calculated at the new angle; is still the global normalization parameter. If significantly increases, it may be a hidden crack; if shows no obvious change, it may be a normal stress distribution.

[0078] According to the results of the secondary sampling, recalculate the crack credibility score, and compare the new crack credibility score with the crack detection threshold. If the new crack credibility score is lower than the crack detection threshold, it is determined as a non - crack area, which may be a stress pseudo - defect or normal texture.

[0079] The above - mentioned formulas are all dimensionless and take their numerical values for calculation. The formula is obtained by collecting a large amount of data for software simulation to get a formula closest to the real situation. The preset parameters in the formula are set by those skilled in the art according to the actual situation.

[0080] The above - mentioned embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination. When implemented using software, the above - mentioned embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general - purpose computer, a special - purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer - readable storage medium, or transmitted from one computer - readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center in a wired or wireless (such as infrared, wireless, microwave, etc.) manner. The computer - readable storage medium can be any available medium that the computer can access, or a data storage device such as a server or data center that contains one or more collections of available media. The available medium can be a magnetic medium (such as a floppy disk, hard disk, magnetic tape), an optical medium (such as a DVD), or a semiconductor medium. The semiconductor medium can be a solid - state drive.

[0081] It should be understood that the term "and / or" in this text is merely a description of the association relationship between associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. Here, A and B can be singular or plural. Additionally, the character " / " in this text generally represents an "or" relationship between the associated objects before and after, but it may also represent an "and / or" relationship, which can be understood specifically by referring to the context. Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed in this text can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0082] As described above, the above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed in this application can easily think of changes or substitutions, which should all be covered by the protection scope of this application.

Claims

1. A solar silicon wafer visual inspection method, characterized in that: include: A multi-angle lighting device is used to illuminate the solar silicon wafer to generate multiple optical images at different incident angles, obtain the original detection images of the silicon wafer at multiple lighting angles, and simultaneously collect infrared transmission images; The collected multi-angle images are normalized, the image features are enhanced through edge enhancement filtering, and the image is binarized using an adaptive threshold method to extract the suspected crack area on the silicon wafer surface; The grayscale gradient change rate of the suspected crack area is calculated under multiple illumination angles, and the stress anomaly degree of the suspected crack area is calculated through infrared transmission data based on the stress distribution inside the silicon wafer. The weighted scoring method is used to generate a crack credibility score based on the calculation results of the grayscale gradient change rate and the stress anomaly degree, and the score is compared with the crack detection threshold. If the crack credibility score is higher than the crack detection threshold, it is determined to be a hidden crack defect, and the defect location and characteristics are recorded. If the crack credibility score is lower than the crack detection threshold, the illumination angle is adjusted for secondary sampling to further confirm the defect type.

2. The solar silicon wafer visual inspection method according to claim 1, characterized in that: The multi-angle lighting device comprises a plurality of adjustable angle LED light source arrays, the incident angle range of the LED light source array is , where: low angle incident light For enhanced visibility of fine surface scratches and shallow cracks; medium angle incident light Used to provide uniform illumination, suitable for overall defect detection; high angle incident light Used to enhance the optical contrast of deep cracks and hidden crack areas.

3. The solar silicon wafer visual inspection method according to claim 2, characterized in that: The grayscale gradient change rate of the suspected crack area is calculated at multiple lighting angles. The grayscale gradient change rate is obtained as follows: The original input image is an RGB color image, which is converted to a grayscale image I(x,y). The Sobel operator is used to calculate the horizontal and vertical grayscale gradients of the image using two 3×3 convolution kernels: Horizontal gradient: ; Vertical gradient: ; Perform convolution calculation on the image and calculate and : ; ; In the formula, Represents the convolution operation; is the gradient in the X direction; is the gradient in the Y direction, combined with and , calculate the gradient amplitude of each pixel: ; Where: G(x,y) represents the total gradient size of the pixel. Under multiple lighting angles, the gradient change rate of the crack area is estimated by the gradient change of adjacent pixels: ;in: Indicates the grayscale gradient change rate.

4. The solar silicon wafer visual inspection method according to claim 3, characterized in that: According to the stress distribution inside the silicon wafer, the degree of stress anomaly in the suspected crack area is calculated through infrared transmission data, which specifically includes: collecting polarized infrared transmission images IIR(x, y), where IIR(x, y) represents the intensity of polarized infrared transmission light at the image coordinates (x, y), normalizing the image, and obtaining a normalized image , which is defined as: ;in, and are the minimum and maximum grayscale values ​​of the original image respectively; Due to the birefringence effect of polarized light caused by internal stress in the silicon wafer, the phase delay at each pixel is calculated: ; Where: δ(x,y) is the phase delay, d(x,y) is the local thickness of the silicon wafer; Δn(x,y) is the refractive index change, which is proportional to the stress; λ is the wavelength of infrared light, and the stress σ(x,y) and the phase delay δ(x,y) satisfy: ;in, is the photoelastic coefficient of the silicon wafer; Perform Fourier transform on the polarized infrared image to identify the stress concentration area F(u,v). The low-frequency component represents the uniform stress distribution, and the high-frequency component represents the stress mutation area. Extract high-frequency stress anomaly areas through bandpass filtering: ; where H(u,v) is a bandpass filter: ; , is the low-frequency and high-frequency cutoff value; perform inverse Fourier transform to restore the stress abnormal area , calculate the standard deviation of stress anomaly distribution : ;in: is the number of pixels in the suspected crack area; is the average stress value of the region; Calculate the rate of change of stress : ; Calculate the stress anomaly value , the expression is: ;in: is the weighting coefficient, satisfying .

5. The solar silicon wafer visual inspection method according to claim 4, characterized in that: According to the real-time factor, combined with the calculation results of the gray gradient change rate and the stress anomaly degree, a weighted scoring method is used to generate the crack credibility score, which includes: The grayscale gradient change rate and stress anomaly value are normalized to be between [0, 1], and the crack credibility score is calculated based on the normalized grayscale gradient change rate and stress anomaly value.

6. The solar silicon wafer visual inspection method according to claim 5, characterized in that: Adjust the illumination angle for secondary sampling to further confirm the defect type, including: To enhance the visibility of cracks or defective areas, a dynamic lighting angle adjustment method is used: ;in: is the adjusted lighting angle, is the original sampling illumination angle; Adjust the step size for the angle; At the new light angle Next, recalculate the grayscale gradient change rate: ;in: is the normalized gray gradient change rate at the new angle; , The grayscale gradient in the X and Y directions calculated by the Sobel operator.

7. The solar silicon wafer visual inspection method according to claim 6, characterized in that: After adjusting the illumination angle, polarized infrared transmission detection is used to calculate the degree of stress anomaly at the new angle: ;in: is the normalized stress anomaly degree under the new illumination angle, is the stress anomaly degree calculated at the new angle; It is still a global normalization parameter.

8. The solar silicon wafer visual inspection method according to claim 7, characterized in that: According to the result of secondary sampling, the crack credibility score is recalculated, and the new crack credibility score is compared with the crack detection threshold. If the new crack credibility score is lower than the crack detection threshold, it is determined to be a non-crack area.

Images