Layered and regional visual detection algorithm for solar silicon wafer fragments

Abstract

The invention relates to a layered and regional visual detection algorithm for solar silicon wafer fragments, which belongs to the technical field of image processing, solves the technical problem of automatically detecting the solar silicon wafer fragments, and comprises the following steps of: 1, acquiring an image; 2, irregular fragment detection; 3, chamfering point notch detection; 4, chamfer edge notch detection; and 5, detecting a straight edge notch. According to the invention, automatic real-time high-precision detection of solar silicon wafer fragments can be realized.

Description

Technical Field

[0001] The invention belongs to the field of machine vision solar silicon wafer detection and provides a layered and regionalized visual detection algorithm for solar silicon wafer fragments. Background Art

[0002] Today, when humans are heavily dependent on fossil energy, solar photovoltaic power generation technology, with its clean, efficient and renewable characteristics, provides a new way to solve the energy crisis. As the basic component of the photovoltaic power generation system, the quality of solar silicon wafers will seriously affect the performance of the photovoltaic power generation system. In the production process of solar cells, silicon wafers are prone to breakage due to factors such as the instability of the production machine, which will greatly affect the efficiency of solar cells and cause serious losses to related companies. In order to ensure the quality of solar cells, it is necessary to accurately detect solar silicon wafer fragments.

[0003] Solar wafer fragment detection methods can be divided into manual detection methods, physical detection methods, and machine vision detection methods. Manual detection methods use the experience of workers on the production site to identify solar cell fragments. This method is easily interfered by external factors, has low efficiency, and a high misjudgment rate. Physical detection methods use physical methods such as ultrasound and laser to detect surface gaps in solar cells. This method is also easily interfered by external factors and has a high cost.

[0004] Machine vision detection methods have fast detection speed, high accuracy, strong real-time performance, and low cost, and are widely used in the field of solar cell fragment detection. Machine vision detection methods can be divided into methods based on traditional machine vision detection and detection methods based on deep learning. The method based on traditional machine vision detection has fast detection speed, high accuracy, and low cost, but this method also has disadvantages such as low fault tolerance. Especially in large-scale real-time production line applications, due to factors such as unstable production line operation and variable silicon wafer sources, the collected images have problems such as deformation and uneven brightness, which ultimately leads to high false detection and missed detection of traditional machine vision detection algorithms. The detection method based on deep learning has a low false detection rate and strong generalization ability, but this method has low accuracy, high cost and high design difficulty.

[0005] There are many types of solar wafer fragments, and their sizes vary greatly. Due to factors such as transportation vibration, unstable production machines, and fragility of silicon wafers, many solar wafers have large gaps, and some gaps are very small (less than 0.1mm). This is a major problem that needs to be solved urgently for solar wafer fragment detection. In addition, solar wafer fragment detection needs to ensure real-time performance to meet the production requirements of the actual production site.

[0006] At the same time, the gaps of solar silicon wafers will occur at the chamfers and straight edges of the silicon wafers. Among them, the gaps at the chamfers can be further divided into gaps at the chamfer points (the connection between the chamfers and the edges of the silicon wafers) and gaps at the chamfer edges (the internal positions of the chamfers of the silicon wafers). Different types of silicon wafer gaps have different pixel arrangements and distributions. In addition, the sizes of gaps in different silicon wafers vary greatly. Therefore, how to perform high-precision and real-time detection of silicon wafer gap types at different locations is also a major problem. Summary of the invention

[0007] The main purpose of the present invention is to overcome the problems of traditional image algorithms in real-time silicon wafer production line visual inspection applications, such as failure to distinguish between notch types, frequent missed detections at chamfers, and difficulty in detecting small notches, and to provide a layered and regionalized visual inspection algorithm for solar silicon wafer fragments. The present invention proposes to perform layered and regionalized visual inspection of solar silicon wafer fragments according to the size, position, and pixel distribution characteristics of solar silicon wafer notches.

[0008] The present invention is implemented by the following technical solution: a layered and regionalized visual detection algorithm for solar silicon wafer fragments, comprising the following steps:

[0009] S1, using backlighting method, using industrial camera to collect solar silicon wafer image, and obtain original image I;

[0010] S2, irregular fragment detection: binarize the original image I to obtain image I1, and use the area difference between the effective contour and the minimum circumscribed rectangle of image I1 to determine whether it is an irregular fragment, where the effective contour is the contour with the largest area among all contours;

[0011] S3, chamfer point gap detection; the chamfer area of ​​image I is cropped to obtain four chamfer images of the silicon wafer, the effective contour in the chamfer is converted into coordinates, the width of the coordinate points in the effective contour is calculated to obtain chamfer point 1 and chamfer point 2, all coordinate points in the effective contour are traversed and the x-axis coordinate difference and y-axis coordinate difference of the current point and its adjacent points are calculated, and then the sum of the x-axis coordinate difference and the y-axis coordinate difference of the current point is obtained PDiff i ; via PDiff i Take the value to determine the outlier point; i Classify and filter according to the rules to obtain abnormal points in the x direction and abnormal points in the y direction; traverse all abnormal points and determine whether there is a continuous gradient abnormality at the chamfer point based on the coordinate difference of adjacent abnormal points. If so, there is a gap at the chamfer point; calculate the area of ​​the effective contour. If the area of ​​the effective contour is greater than the area threshold, it is determined that there is an abnormal gap in the area of ​​the silicon wafer chamfer point;

[0012] S4, chamfer edge gap detection; perform 2x super-resolution and binarization on the four corner graphics respectively, perform coordinate transformation on the effective contour in the processed corner image, traverse all coordinate points of the effective contour after coordinate transformation and calculate the x-axis coordinate difference and y-axis coordinate difference between the current point and its adjacent points, and obtain the sum of the differences, judge the abnormal point according to the rules through the value of the sum of the differences, and finally judge whether there is a gap at the chamfer edge of the silicon wafer according to the coordinate difference of the adjacent positive and negative abnormal points;

[0013] S5, straight edge gap detection; crop image I, and perform 2 times super-resolution and high grayscale binarization with a binarization threshold of 150 to obtain four straight edge images of the silicon wafer, traverse the valid contours in the straight edge images, calculate the x-direction difference or y-direction difference between the current coordinate point and its adjacent coordinates, and then obtain the abnormal point according to the rules based on the value of the x-direction difference or y-direction difference, and finally determine whether the silicon wafer has a left straight edge gap or an upper straight edge gap based on the coordinate difference of the adjacent positive and negative abnormal points.

[0014] Furthermore, the specific process of irregular fragment detection in step S2 is as follows:

[0015] S2-1, perform a binarization operation on the original image I to obtain an image I1;

[0016] S2-2, find and traverse all contours in the image I1, and obtain the valid contour Contour and its minimum circumscribed rectangle Contour_rect;

[0017] S2-3. Calculate the area difference between Contour_rect and Contour using formula (1):

[0018] Area_diff = Contour_rect_area- Contour_area (1)

[0019] Among them, Contour_rect_area and Contour_area are the areas of Contour_rect and Contour respectively;

[0020] S2-4, if Area_diff satisfies formula (2), that is, the difference between the area of ​​the minimum circumscribed rectangle of the solar cell and its contour area is too large, it is considered that there are irregular fragments in the solar cell;

[0021] Area_diff > 200000 (2)

[0022] S2-5. If it is determined in step S2-4 that irregular fragments exist in the solar silicon wafer, the text "damaged slice!" is output on the image I.

[0023] Furthermore, the specific process of chamfer point notch detection in step S3 is as follows:

[0024] S3-1, the lower right corner area, upper right corner area, lower left corner area and upper left corner area of ​​the image I are respectively cut and binarized to obtain the lower right chamfered image CP1, upper right chamfered image CP2, lower left chamfered image CP3 and upper left chamfered image CP4 of the solar silicon wafer, and the length and width of the cut area are both 300;

[0025] S3-2, find and traverse all contours in CP1 to obtain the valid contour ChamferP1;

[0026] S3-3, traverse the coordinate points on ChamferP1, and transform the coordinate points according to formula (3) and formula (4);

[0027] Chamfer P 1,X,i = ChamferP 1,x,i + (w-300) (3)

[0028] Chamfer P 1,Y,i = ChamferP 1,y,i + (h-300) (4)

[0029] Among them, ChamferP 1,x,i is the x-coordinate of the traversed point, ChamferP 1,y,i is the y coordinate of the traversed point, ChamferP 1,X,i and Chamfer P 1,Y,i are the x-coordinate and y-coordinate of the point after coordinate transformation, w is the width of image I, and h is the height of image I;

[0030] S3-4, perform the operation of step S3-2 on CP2, CP3 and CP4 respectively, and obtain the effective contours ChamferP2, ChamferP3 and ChamferP4. In particular, in the operation of step S3-3, for the points in ChamferP2, it is only necessary to perform coordinate transformation according to formula (3), for the points in ChamferP3, it is only necessary to perform coordinate transformation according to formula (4), and for the points in ChamferP4, no coordinate transformation is required;

[0031] S3-5, traverse the coordinate points after coordinate transformation in ChamferP1, and calculate the coordinate difference between two adjacent coordinate points according to formula (5):

[0032]

[0033] Among them, X_diff iIndicates the difference between two adjacent coordinate points in the x direction, Y_diff i Represents the difference between two adjacent coordinate points in the y direction;

[0034] S3-6, traverse the coordinate points in ChamferP1, and calculate the x-width Width_x of the current point according to formula (6) i , and calculate the average value Sumwidth_x of the x-direction width of the current point and the sum of the x-direction widths of the 20 points after the current point according to formula (7) i Finally, the difference ΔWidthx is calculated according to formula (8): i , if the difference satisfies formula (9), this point is the chamfer point Coner1 at the lower left of ChamferP1;

[0035] Width_x i = |ChamferP 1,X,i - 300| (6)

[0036]

[0037] ΔWidthx i = Widh_x i -Sumwidth_x i (8)

[0038] ΔWidthx i < 0.3 (9)

[0039] S3-7, the principle is the same as step S3-5, traverse the coordinate points in the contour ChamferP1 and calculate the y-direction width of the current traversal point to obtain the chamfer point Coner2 on the upper right of ChamferP1;

[0040] S3-8. In the contour ChamferP1, traverse each X_diff obtained by formula (5) i and Y_diff i , and calculate PDiff according to formula (10) i :

[0041] PDiff i = |X_diff i | + |Y_diff i | (10)

[0042] According to the calculation results, PDiff i The value is 1 or 2;

[0043] S3-9, starting from PDiff5, traverse the PDiff obtained in step S3-8 in sequence i, according to rules a and b, determine whether there are abnormal points at the chamfer point:

[0044] Rule a: PDiff i-3 PDiff i-2 PDiff i-1 PDiff i PDiff i+1 PDiff i+2 PDiff i+3 There is a value equal to 1 in

[0045] Rule b: PDiff i-4 PDiff i-3 PDiff i-2 PDiff i-1 PDiff i+1 PDiff i+2 PDiff i+3 There is a value equal to 1 in

[0046] When the traversed PDiff i When any of the above rules is met, an outlier is obtained. i ;

[0047] S3-10, Outlier i It is attributed to the chamfer points Coner1 and Coner2, and then the outlier point X_Outlier in the x direction is obtained. i and the outlier point in the y direction Y_Outlier i ;

[0048] S3-11, traverse X_Outlier separately i and Y_Outlier i , and calculate the difference ΔX_Outlier according to formula (11) and formula (12) i and ΔY_Outlier i :

[0049] ΔX_Outlier i = |X_Outlier x,i - X_Outlier x,i-1 | (11)

[0050] ΔY_Outlier i = |Y_Outlier y,i - Y_Outlier y,i-1 | (12)

[0051] Among them, X_Outlier x,iX_Outlier i x coordinate, Y_Outlier y,i Y_Outlier i The y coordinate of ΔX_Outlier i or ΔY_Outlier i If it is less than 3, then the X_Outlier i and Y_Outlier i Keep it as an outlier point, otherwise remove it from the outlier point set. After the traversal is completed, the outlier points X_Outlier in the x and y directions can be obtained again i and Y_Outlier i ;

[0052] S3-12, detect the abnormal gap of continuous gradient at the chamfer point; traverse the x-direction abnormal point X_Outlier i , calculate the difference between adjacent outliers according to formula (11). If there are two consecutive ΔX_outlier i If it is greater than 1, it is determined that there is a continuous abnormal gradient gap at the chamfer point. For the y-direction chamfer point, traverse the y-direction abnormal point Y_Outlier i , calculate the difference between adjacent outliers according to formula (12). If there are two consecutive ΔY_outlier i If it is greater than 1, it is determined that there is a continuous abnormal gradient gap at the chamfer point;

[0053] S3-13, calculate the contour ChamferP1 area C_area by formula (13);

[0054]

[0055] If the value of C_area satisfies formula (14), it is judged that there is an abnormal gap in the chamfer area of ​​the solar silicon wafer;

[0056] C_area > 15 (14)

[0057] S3-14, performing the processing of steps S3-5 to S3-13 on the contours ChamferP2, ChamferP3 and ChamferP4, and determining whether there are abnormal area gaps and abnormal continuous gradient gaps at the chamfer points;

[0058] S3-15. If a gap is detected in the above step, a frame is drawn to identify the position of the gap in the image.

[0059] Furthermore, the specific process of the S4 chamfer edge notch detection is as follows:

[0060] S4-1, perform 2-fold super-resolution and binarization on CP1, CP2, CP3 and CP4 respectively, and then obtain a lower right chamfer image CE1, an upper right chamfer image CE2, a lower left chamfer image CE3 and an upper left chamfer image CE4;

[0061] S4-2, find all contours in CE1; traverse all contours to find the valid contour ChamferEdge1, and perform the same operation on CE2, CE3 and CE4 to obtain the valid contours ChamferEdge2, ChamferEdge3 and ChamferEdge4 respectively;

[0062] S4-3, similar to the principle of steps S3-3 and S3-4, perform corresponding coordinate transformation on ChamferEdge1, ChamferEdge2, ChamferEdge3 and ChamferEdge4;

[0063] S4-4, traverse the coordinate points after coordinate transformation in the contour ChamferEdge1, and calculate the x and y coordinate differences between two adjacent coordinate points according to formula (15):

[0064]

[0065] Here, Chamfer Edge 1,x,i is the x-coordinate of the point after coordinate transformation on ChamferEdge1, 1,y,i Cx_diff is the y coordinate of the point after coordinate transformation on ChamferEdge1. i Indicates the difference between adjacent coordinate points in the x direction, Cy_diff i Indicates the difference between adjacent coordinate points in the y direction;

[0066] S4-5, traverse each Cx_diff obtained in step S4-4 i and Cy_diff i , and calculate EDiff according to formula (16) i :

[0067] EDiff i =|Cx_diff i |+|Cy_diff i | (16)

[0068] According to the calculation results, EDiff i The value is 1 or 2;

[0069] S4-6, starting from EDiff5, traverse the EDiff obtained in step S4-5 in sequence i, and judge the abnormal points according to rules c and d:

[0070] Rule c: EDiff i-3 EDiff i-2 , EDiff i-1 EDiff i , EDiff i+1 EDiff i+2 , EDiff i+3 There is a value equal to 1 in

[0071] Rule d: EDiff i-4 EDiff i-3 EDiff i-2 , EDiff i-1 , EDiff i+1 , EDiff i+2 EDiff i+3 There is a value equal to 1 in

[0072] When rule c or rule d is met, the anomaly point Anomaly is obtained i , Anomaly x,i Anomaly i x coordinate, Anomaly y,i Anomaly i The y coordinate of

[0073] S4-7. Traversing the Anomaly i , mark Anomaly x,i Not equal to 0, Anomaly y,i Anomalies equal to 0 are positive anomalies and are marked as Anomaly. y,i Not equal to 0, Anomaly x,i The outlier point equal to 0 is a negative outlier point. If there is a negative outlier point, search counterclockwise to see if there is a positive outlier point. If so, the coordinate difference between the two outliers is further calculated according to formula (17), and the chamfer edge depression is judged according to formula (18) to see if it meets the detection condition:

[0074]

[0075] Here, τ a is the detection threshold, and the threshold is 20. If formula (18) is satisfied, it is judged that there is a concave notch at the chamfered edge of the solar cell;

[0076] S4-8, for the points in the contours ChamferEdge2, ChamferEdge3 and ChamferEdge4, it is only necessary to repeat the process of steps S4-4 to S4-7 to determine whether there is a concave notch at the upper left chamfer edge of the solar silicon wafer; in particular, for the contours ChamferEdge2 and ChamferEdge3, if there are positive abnormal points, search counterclockwise to see if there are negative abnormal points. If so, further determine the coordinate difference between the two abnormal points according to formula (17). If the coordinate difference satisfies formula (18), it is determined that there is a concave notch at the chamfer edge.

[0077] S4-9. If a gap is detected in the above steps, the gap position is framed in the image.

[0078] Furthermore, the specific process of the straight edge gap detection in step S5 is as follows:

[0079] S5-1, cropping the image I according to the chamfer point obtained in the previous step, and performing 2-fold super-resolution and high-grayscale binarization with a binarization threshold of 150 on the cropped image, to obtain images e1, e2, e3, and e4 representing the left straight edge, right straight edge, upper straight edge, and lower straight edge of the solar cell, respectively;

[0080] S5-2, traverse all contours in each image in images e1, e2, e3 and e4, and then find out the valid contours in each image, which are Edge1, Edge2, Edge3 and Edge4 respectively;

[0081] S5-3, traverse the points in Edge1, and calculate the difference between adjacent coordinates in the x direction Ex_diff according to formula (19) i :

[0082] Ex_diff i =Edge 1,x,i+1 -Edge 1,x,i (19)

[0083] Here, Edge 1,x,i is the x coordinate of the midpoint of the traversed contour Edge1, Edge 1,x,i+1 For Edge 1,x,i The x coordinates of adjacent points; when Ex_diff i If it is not 0, search counterclockwise for the first non-0 Ex_diff within a distance of 5. i If both are positive, the point is a positive outlier point. If both are negative, the point is a negative outlier point. If they do not exist or the positive and negative values ​​are opposite, the point is a normal point. The contour points that meet the abnormal conditions are recorded as outliers. i ;

[0084] S5-4. Traversing the outlier points i , set the search distance threshold τ s is 30, if Exoutlier i For a positive outlier point, search the distance τ counterclockwise s Is there a negative outlier in the equation? If so, calculate the length ΔEy between the two outliers according to formula (20):

[0085] ΔEy = Exoutlier y,i+n - Exoutlier y,i (20)

[0086] ΔEy < τ e (twenty one)

[0087] Among them, Exoutlier y,i Exoutlier i The y coordinate of y,i+n Negative outlier point Exoutlier i+n The y coordinate of e is the detection threshold, which is set to 20. If ΔEy satisfies formula (21), it is determined that there is a left straight edge damage gap in the solar silicon wafer;

[0088] S5-5, performing the operations of steps S5-3 to S5-4 on the contour Edge2 to determine whether there is a broken gap on the right straight edge of the solar silicon wafer;

[0089] S5-6, traverse the points in the contour Edge3, and calculate the difference Ey_diff between adjacent coordinates in the y direction according to formula (22) i :

[0090] Ey_diff i = Edge 3,y,i+1 -Edge 3,y,i (twenty two)

[0091] Among them, Edge 3,y,i is the y coordinate of the point in the contour Edge3, 3,y,i+1 For Edge 3,y,i The y coordinates of adjacent points, when the Ey_diff of the i-th contour point i If it is not 0, search counterclockwise for the first non-0 Ey_diff within a distance of 5. i If both are positive, the point is a positive outlier point. If both are negative, the point is a negative outlier point. If they do not exist or the positive and negative values ​​are opposite, the point is a normal point. The contour points that meet the abnormal conditions are recorded as outliers. i ;

[0092] S5-7. Traversing the outlier point Eyoutlier i , if Eyoutlier i For a negative outlier point, search the distance τ counterclockwise s Is there a positive outlier in the system? If so, calculate the length ΔEx between the two outliers according to formula (23):

[0093] ΔEx=Eyoutlier x,i+n -Eyoutlier x,i (twenty three)

[0094] ΔEx < τ e (twenty four)

[0095] Here, Eyoutlier x,i For Eyoutlier i The x-coordinate of Eyoutlier x,i+n is a positive outlier point Eyoutlier i+n If ΔEx satisfies formula (24), it is determined that there is a straight edge damage gap on the solar silicon wafer;

[0096] S5-8, performing the operations of steps S5-6 to S5-7 for the contour Edge4, and determining whether there is a broken gap on the lower straight edge of the solar cell;

[0097] S5-9, cropping image I according to the chamfer points obtained in the previous step, and performing 2-fold super-resolution and low-grayscale binarization with a binarization threshold of 40 on the cropped image to obtain images e5, e6, e7, and e8, and performing steps S5-3 to S5-8 on images e5, e6, e7, and e8 to determine whether there is a damaged gap on the straight edge of the solar cell;

[0098] S5-10: If a gap is detected in the above steps, the gap position is framed in the image.

[0099] Compared with the prior art, the present invention has the following advantages:

[0100] 1. Layered sequential detection, the algorithm is divided into two levels: irregular wafer large gap detection and regular wafer small gap detection. The former is designed to detect irregular broken silicon wafers, and the latter detects small gaps in different areas of regular silicon wafers, including chamfer point gap detection, chamfer edge gap detection and straight edge gap detection. Layered sequential detection can ensure that the upper layer large gap detection algorithm and the lower layer small gap detection algorithm do not interfere with each other.

[0101] 2. Divide silicon wafers with regular contours into regions, such as chamfered edges, chamfered points, and straight edge areas, and extract features by region to improve the performance of the detection algorithm.

[0102] 3. For the detection of different levels and different areas, the algorithm of the present invention analyzes the pixel arrangement rules of the gap in detail, and proposes and implements a pixel-level gap detection algorithm. In the detection of large gaps in irregular wafers, the present invention analyzes the concave features of irregular large-sized gaps, and uses the difference between the silicon wafer contour and its minimum circumscribed rectangle to realize the detection of irregular large-sized gaps. Before the detection of small gaps in regular wafers, the present invention uses the characteristics of the sudden change in pixel distribution at the chamfered edge and the straight edge of the silicon wafer to obtain the chamfer point of the silicon wafer. In the chamfer point detection area, the present invention obtains the abnormal point through the abnormal difference between adjacent pixel points, and determines whether there is a gap at the chamfer point through indicators such as the number of continuous abnormal points and the area of ​​the abnormal area. In the chamfer edge or straight edge detection area, the present invention searches for positive and negative abnormal points in each area through the relationship between adjacent pixels, compares the relationship between the positive and negative abnormal points of the gap in the current area to obtain suspected gaps, and detects the chamfer edge or straight edge gap based on whether the gap length meets the threshold.

[0103] 4. In the chamfered edge and straight edge areas, the algorithm of the present invention improves the detection accuracy of the gap and enhances the robustness of the detection algorithm by super-resolution processing of the image. BRIEF DESCRIPTION OF THE DRAWINGS

[0104] Figure 1 It is a flow chart of the present invention;

[0105] Figure 2 The original image of a solar cell captured by an industrial camera using backlighting;

[0106] Figure 3 The image is one in which a gap is detected by the irregular fragment detection step;

[0107] Figure 4 The chamfered image of the cut solar cell;

[0108] Figure 5 The image is an image with gaps detected after the chamfer point gap detection;

[0109] Figure 6 An image in which a gap is detected after the chamfer edge gap detection;

[0110] Figure 7 The image with the gap detected after the straight edge gap detection; DETAILED DESCRIPTION

[0111] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0112] like Figure 1 The layered and regionalized visual inspection algorithm for solar silicon wafer fragments shown includes the following steps:

[0113] S1, such as Figure 2 As shown, the backlighting method is adopted and the industrial camera is used to collect the image of the solar silicon wafer to obtain the original image I;

[0114] S2. Irregular fragment detection

[0115] In order to determine whether there are irregular fragments on the solar silicon wafer (large gaps at the chamfers or straight edges of the solar cell), irregular fragment detection is performed on the solar silicon wafer.

[0116] S2-1, perform a binarization operation on the image I to obtain an image I1;

[0117] S2-2, find and traverse all contours in the image I1, and obtain the valid contour Contour (the contour with the largest area among all contours) and its minimum circumscribed rectangle Contour_rect;

[0118] S2-3. Calculate the area difference between Contour_rect and Contour using formula (1):

[0119] Area_diff = Contour_rect_area- Contour_area (1)

[0120] Here, Contour_rect_area and Contour_area are the areas of Contour_rect and Contour respectively;

[0121] S2-4, if Area_diff satisfies formula (2), that is, the difference between the area of ​​the minimum circumscribed rectangle of the solar cell and its contour area is too large, it is considered that there are irregular fragments in the solar cell;

[0122] Area_diff > 200000 (2)

[0123] S2-5, such as Figure 3 As shown, if it is determined in step S2-4 that irregular fragments exist in the solar silicon wafer, the text "damaged slice!" is output on the image I.

[0124] S3, chamfer point notch detection

[0125] In order to determine whether there is a gap at the chamfer point of the solar silicon wafer (the junction of the chamfer edge and the straight edge of the solar cell), the chamfer point detection of the solar silicon wafer is performed;

[0126] S3-1, such as Figure 4As shown, the lower right corner area, upper right corner area, lower left corner area and upper left corner area of ​​image I are cropped and binarized respectively, and the image CP1 of the lower right chamfer of the solar silicon wafer, the image CP2 of the upper right chamfer, the image CP3 of the lower left chamfer and the image CP4 of the upper left chamfer are obtained. The length and width of the cropped area are both 300;

[0127] S3-2, find and traverse all contours in CP1 to obtain the valid contour ChamferP1;

[0128] S3-3, traverse the coordinate points on ChamferP1, and transform the coordinate points according to formula (3) and formula (4);

[0129] Chamfer P 1,X,i = ChamferP 1,x,i + (w-300) (3)

[0130] Chamfer P 1,Y,i = ChamferP 1,y,i + (h-300) (4)

[0131] Here, Chamfer P 1,x,i is the x-coordinate of the traversed point, ChamferP 1,y,i is the y coordinate of the traversed point, ChamferP 1,X,i and Chamfer P 1,Y,i are the x-coordinate and y-coordinate of the point after coordinate transformation, w is the width of image I, and h is the height of image I;

[0132] S3-4, perform the operation of step S3-2 on CP2, CP3 and CP4 to obtain the effective contours ChamferP2, ChamferP3 and ChamferP4. In particular, in step S3-3, for the points in ChamferP2, only the coordinate transformation needs to be performed according to formula (3), for the points in ChamferP3, only the coordinate transformation needs to be performed according to formula (4), and for the points in ChamferP4, no coordinate transformation is required;

[0133] S3-5, traverse the coordinate points after coordinate transformation in ChamferP1, and calculate the coordinate difference between two adjacent coordinate points according to formula (5):

[0134]

[0135] Here, X_diff i Indicates the difference between two adjacent coordinate points in the x direction, Y_diff i Represents the difference between two adjacent coordinate points in the y direction;

[0136] S3-6, traverse the coordinate points in ChamferP1, and calculate the x-width Width_x of the current point according to formula (6) i , and calculate the average value Sumwidth_x of the x-direction width of the current point and the sum of the x-direction widths of the 20 points after the current point according to formula (7) i Finally, the difference ΔWidthx is calculated according to formula (8): i , if the difference satisfies formula (9), this point is considered to be the chamfer point Coner1 at the lower left of ChamferP1;

[0137] Width_x i = |ChamferP 1,X,i - 300| (6)

[0138]

[0139] ΔWidthx i = Widh_x i -Sumwidth_x i (8)

[0140] ΔWidthx i < 0.3 (9)

[0141] S3-7, the principle is the same as step S3-5, traverse the coordinate points in the contour ChamferP1 and calculate the y-direction width of the current traversal point to obtain the chamfer point Coner2 on the upper right of ChamferP1;

[0142] S3-8. In the contour ChamferP1, traverse each X_diff obtained by formula (5) i and Y_diff i , and calculate PDiff according to formula (10) i :

[0143] PDiff i = |X_diff i | + |Y_diff i | (10)

[0144] According to the calculation results, PDiff i The value is 1 or 2;

[0145] S3-9, starting from PDiff5, traverse the PDiff obtained in step S3-8 in sequence i , according to rules a and b, determine whether there are abnormal points at the chamfer point:

[0146] Rule a: (PDiff i-3PDiff i-2 PDiff i-1 PDiff i PDiff i+1 PDiff i+2 PDiff i+3 ) has a value equal to 1.

[0147] Rule b: (PDiff i-4 PDiff i-3 PDiff i-2 PDiff i-1 PDiff i+1 PDiff i+2 PDiff i+3 ) has a value equal to 1.

[0148] When the traversed PDiff i When any of the above rules is met, an outlier is obtained. i ;

[0149] S3-10, Outlier i It is attributed to the chamfer points Coner1 and Coner2, and then the outlier point X_Outlier in the x direction is obtained. i and the outlier point in the y direction Y_Outlier i ;

[0150] S3-11, traverse X_Outlier separately i and Y_Outlier i , and calculate the difference ΔX_Outlier according to formula (11) and formula (12) i and ΔY_Outlier i :

[0151] ΔX_Outlier i = |X_Outlier x,i - X_Outlier x,i-1 | (11)

[0152] ΔY_Outlier i = |Y_Outlier y,i - Y_Outlier y,i-1 | (12)

[0153] Here, X_Outlier x,i X_Outlier i x coordinate, Y_Outlier y,i Y_Outlieri The y coordinate of ΔX_Outlier i or ΔY_Outlier i If it is less than 3, then the X_Outlier i and Y_Outlier i The outlier point is retained, otherwise the point is removed from the outlier point set. After the traversal is completed, the outlier points X_Outlier in the x-direction and y-direction can be obtained again. i and Y_Outlier i ;

[0154] S3-12, detect the abnormal gap of continuous gradient at the chamfer point. Traverse the x-direction abnormal point X_Outlier i , calculate the difference between adjacent outliers according to formula (11). If there are two consecutive ΔX_outlier i If it is greater than 1, it is determined that there is a continuous abnormal gradient gap at the chamfer point. For the y-direction chamfer point, traverse the y-direction abnormal point Y_Outlier i , calculate the difference between adjacent outliers according to formula (12). If there are two consecutive ΔY_outlier i If it is greater than 1, it is determined that there is a continuous abnormal gradient gap at the chamfer point;

[0155] S3-13. To detect abnormal notches at the chamfer points of solar silicon wafers, the contour ChamferP1 area C_area is calculated using formula (13).

[0156]

[0157] If the value of C_area satisfies formula (14), it is judged that there is an abnormal gap in the chamfer area of ​​the solar silicon wafer;

[0158] C_area > 15 (14)

[0159] S3-14, performing the processing of steps S3-5 to S3-13 on the contours ChamferP2, ChamferP3 and ChamferP4, and determining whether there are abnormal area gaps and abnormal continuous gradient gaps at the chamfer points;

[0160] S3-15, such as Figure 5 As shown, if a gap is detected in the above steps, the gap position is framed in the image.

[0161] S4, chamfer edge notch detection

[0162] In order to determine the gap of the chamfered edge of the solar silicon wafer (the straight edge inside the chamfer of the solar cell), the chamfered edge of the solar cell is detected.

[0163] S4-1, perform 2-fold super-resolution and binarization on CP1, CP2, CP3 and CP4 respectively, and then obtain a lower right chamfer image CE1, an upper right chamfer image CE2, a lower left chamfer image CE3 and an upper left chamfer image CE4;

[0164] S4-2, find all contours in CE1. Traverse all contours and find the valid contour ChamferEdge1. Perform the same operation on CE2, CE3 and CE4 to obtain the valid contours ChamferEdge2, ChamferEdge3 and ChamferEdge4 respectively;

[0165] S4-3, similar to the principle of steps S3-3 and S3-4, perform corresponding coordinate transformation on ChamferEdge1, ChamferEdge2, ChamferEdge3 and ChamferEdge4;

[0166] S4-4, traverse the coordinate points after coordinate transformation in the contour ChamferEdge1, and calculate the x and y coordinate differences between two adjacent coordinate points according to formula (15):

[0167]

[0168] Here, Chamfer Edge 1,x,i is the x coordinate of the point after coordinate transformation on ChamferEdge1, 1,y,i Cx_diff is the y coordinate of the point after coordinate transformation on ChamferEdge1. i Indicates the difference between adjacent coordinate points in the x direction, Cy_diff i Indicates the difference between adjacent coordinate points in the y direction;

[0169] S4-5, traverse each Cx_diff obtained in step S4-4 i and Cy_diff i , and calculate EDiff according to formula (16) i :

[0170] EDiff i =|Cx_diff i |+|Cy_diff i | (16)

[0171] According to the calculation results, EDiff i The value is 1 or 2;

[0172] S4-6, starting from EDiff5, traverse the EDiff obtained in step S4-5 in sequence i, and judge the abnormal points according to rules c and d:

[0173] Rule c: (EDiff i-3 , EDiff i-2 EDiff i-1 , EDiff i , EDiff i+1 EDiff i+2 EDiff i+3 ) has a value equal to 1.

[0174] Rule d: (EDiff i-4 EDiff i-3 EDiff i-2 EDiff i-1 EDiff i+1 , EDiff i+2 EDiff i+3 ) has a value equal to 1.

[0175] When rule c or rule d is met, the anomaly point is obtained. i , Anomaly x,i Anomaly i x coordinate, Anomaly y,i Anomaly i The y coordinate of

[0176] S4-7. Traversing the Anomaly i , mark Anomaly x,i Not equal to 0, Anomaly y,i Anomalies equal to 0 are positive anomalies and are marked as Anomaly. y,i Not equal to 0, Anomaly x,i The outlier point equal to 0 is a negative outlier point. If there is a negative outlier point, search counterclockwise to see if there is a positive outlier point. If so, calculate the coordinate difference between the two outliers according to formula (17), and judge whether the chamfer edge depression meets the detection condition according to formula (18):

[0177]

[0178] Here, τ a is the detection threshold, which is generally set to 20. If formula (18) is satisfied, it is determined that there is a concave notch at the chamfered edge of the solar cell;

[0179] S4-8, for the points in the contours ChamferEdge2, ChamferEdge3 and ChamferEdge4, it is only necessary to repeat the process of steps S4-4 to S4-7 to determine whether there is a concave notch at the upper left chamfer edge of the solar silicon wafer; in particular, for the contours ChamferEdge2 and ChamferEdge3, if there are positive abnormal points, search counterclockwise to see if there are negative abnormal points. If so, further determine the coordinate difference between the two abnormal points according to formula (17). If the coordinate difference satisfies formula (18), it is determined that there is a concave notch on the chamfer edge.

[0180] S4-9, such as Figure 6 As shown, if a gap is detected in the above steps, the gap position is framed in the image.

[0181] S5, straight edge notch detection

[0182] In order to determine whether there is a gap at the straight edge of the solar cell, the straight edge of the solar cell is tested.

[0183] S5-1, cropping the image I according to the chamfer point obtained in the previous step, and performing 2-fold super-resolution and high-grayscale binarization with a binarization threshold of 150 on the cropped image, to obtain images e1, e2, e3, and e4 representing the left straight edge, right straight edge, upper straight edge, and lower straight edge of the solar cell, respectively;

[0184] S5-2, traverse all contours in each image in images e1, e2, e3 and e4, and then find out the valid contours in each image, which are Edge1, Edge2, Edge3 and Edge4 respectively;

[0185] S5-3, traverse the points in Edge1, and calculate the difference between adjacent coordinates in the x direction Ex_diff according to formula (19) i :

[0186] Ex_diff i =Edge 1,x,i+1 -Edge 1,x,i (19)

[0187] Here, Edge 1,x,i is the x coordinate of the midpoint of the traversed contour Edge1, Edge 1,x,i+1 For Edge 1,x,i The x coordinates of adjacent points. i If it is not 0, search counterclockwise for the first non-0 Ex_diff within a distance of 5. iIf both are positive, the point is a positive outlier. If both are negative, the point is a negative outlier. If they do not exist or the positive and negative values ​​are opposite, the point is a normal point. The contour points that meet the abnormal conditions are recorded as outliers. i ;

[0188] S5-4. Traversing the outlier points i , set the search distance threshold τ s is 30, if Exoutlier i For a positive outlier point, search the distance τ counterclockwise s Is there a negative outlier in the equation? If so, calculate the length ΔEy between the two outliers according to formula (20):

[0189] ΔEy = Exoutlier y,i+n - Exoutlier y,i (20)

[0190] ΔEy < τ e (twenty one)

[0191] Here, Exoutlier y,i Exoutlier i The y coordinate of y,i+n Negative outlier point Exoutlier i+n The y coordinate of e is the detection threshold, which is generally set to 20. If ΔEy satisfies formula (21), it is judged that there is a left straight edge damage gap in the solar silicon wafer;

[0192] S5-5, performing the operations of steps S5-3 to S5-4 on the contour Edge2 to determine whether there is a broken gap on the right straight edge of the solar silicon wafer;

[0193] S5-6, traverse the points in the contour Edge3, and calculate the difference Ey_diff between adjacent coordinates in the y direction according to formula (22) i :

[0194] Ey_diff i = Edge 3,y,i+1 -Edge 3,y,i (twenty two)

[0195] Here, Edge 3,y,i is the y coordinate of the point in the contour Edge3, 3,y,i+1 For Edge 3,y,i The y coordinates of adjacent points. When the Ey_diff of the i-th contour point iIf it is not 0, search counterclockwise for the first non-0 Ey_diff within a distance of 5. i If both are positive, the point is a positive outlier point. If both are negative, the point is a negative outlier point. If they do not exist or the positive and negative values ​​are opposite, the point is a normal point. The contour points that meet the abnormal conditions are recorded as outliers. i ;

[0196] S5-7. Traversing the outlier point Eyoutlier i , if Eyoutlier i For a negative outlier point, search the distance τ counterclockwise s Is there a positive outlier in the system? If so, calculate the length ΔEx between the two outliers according to formula (23):

[0197] ΔEx=Eyoutlier x,i+n -Eyoutlier x,i (twenty three)

[0198] ΔEx < τ e (twenty four)

[0199] Here, Eyoutlier x,i For Eyoutlier i The x-coordinate of Eyoutlier x,i+n is a positive outlier point Eyoutlier i+n If ΔEx satisfies formula (24), it is determined that there is a broken gap on the upper straight edge of the solar silicon wafer;

[0200] S5-8, performing the operations of steps S5-6 to S5-7 for the contour Edge4, and determining whether there is a broken gap on the lower straight edge of the solar cell;

[0201] S5-9, cropping image I according to the chamfer points obtained in the previous step, and performing 2x super-resolution and low grayscale binarization with a binarization threshold of 40 on the cropped image to obtain images e5, e6, e7 and e8. Perform steps S5-3 to S5-8 on images e5, e6, e7 and e8 to determine whether there is a broken gap on the straight edge of the solar cell;

[0202] S5-10, such as Figure 7 As shown, if a gap is detected in the above steps, the gap position is framed in the image.

[0203] The present invention accurately and rapidly locates the gap of the solar silicon wafer through machine vision when detecting solar silicon wafer fragments, meeting the requirements of the actual solar cell production site. The present invention locates the gaps at the chamfer and straight edge of the solar silicon wafer through machine vision and image processing algorithms to achieve the purpose of automatic detection. The present invention divides the silicon wafer gap types into irregular gaps, chamfer point gaps, chamfer edge gaps and straight edge gaps according to the position, size and pixel distribution of the gaps, and detects the gaps by methods suitable for their respective gap characteristics.

[0204] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A layered and regionalized visual inspection algorithm for solar silicon wafer fragments, characterized in that: The following steps are involved: S1, using backlighting method, using industrial camera to collect solar silicon wafer image, and obtain original image I; S2, irregular fragment detection: binarize the original image I to obtain image I1, and use the area difference between the effective contour and the minimum circumscribed rectangle of image I1 to determine whether it is an irregular fragment, where the effective contour is the contour with the largest area among all contours; S3, chamfer point gap detection; the chamfer area of ​​image I is cropped to obtain four chamfer images of the silicon wafer, the effective contour in the chamfer is converted into coordinates, the width of the coordinate points in the effective contour is calculated to obtain chamfer point 1 and chamfer point 2, all coordinate points in the effective contour are traversed and the x-axis coordinate difference and y-axis coordinate difference of the current point and its adjacent points are calculated, and then the sum of the x-axis coordinate difference and the y-axis coordinate difference of the current point is obtained PDiff i , through PDiff i Determine the abnormal points by taking values; classify and filter the abnormal points according to the rules to obtain the abnormal points in the x direction and the abnormal points in the y direction; traverse all the abnormal points, and determine whether there is a continuous gradient abnormality at the chamfer point according to the coordinate difference of the adjacent abnormal points. If so, there is a gap at the chamfer point; calculate the area of ​​the effective contour. If the area of ​​the effective contour is greater than the area threshold, it is determined that there is an abnormal gap in the area of ​​the silicon wafer chamfer point; S4, chamfer edge gap detection; perform 2x super-resolution and binarization on the four chamfer images respectively, perform coordinate transformation on the effective contour in the processed chamfer image, traverse all coordinate points of the effective contour after coordinate transformation and calculate the x-axis coordinate difference and y-axis coordinate difference between the current point and its adjacent points, and obtain the sum of the differences, judge the abnormal point according to the rules through the value of the sum of the differences, and finally judge whether there is a gap at the chamfer edge of the silicon wafer according to the coordinate difference of the adjacent positive and negative abnormal points; S5, straight edge gap detection; crop image I, and perform 2 times super-resolution and high grayscale binarization with a binarization threshold of 150 to obtain four straight edge images of the silicon wafer, traverse the valid contours in the straight edge images, calculate the x-direction difference or y-direction difference between the current coordinate point and its adjacent coordinates, and then obtain the abnormal point according to the rules based on the value of the x-direction difference or y-direction difference, and finally determine whether the silicon wafer has a left straight edge gap or an upper straight edge gap based on the coordinate difference of the adjacent positive and negative abnormal points.

2. The layered and region-based visual detection algorithm for solar silicon wafer fragments according to claim 1 is characterized by: The specific process of irregular fragment detection in step S2 is as follows: S2-1, perform a binarization operation on the original image I to obtain an image I1; S2-2, find and traverse all contours in the image I1, and obtain the valid contour Contour and its minimum circumscribed rectangle Contour_rect; S2-3. Calculate the area difference between Contour_rect and Contour using formula (1): Area_diff=Contour_rect_area-Contour_area (1) Among them, Contour_rect_area and Contour_area are the areas of Contour_rect and Contour respectively; S2-4, if Area_diff satisfies formula (2), that is, the difference between the area of ​​the minimum circumscribed rectangle of the solar cell and its contour area is too large, it is considered that there are irregular fragments in the solar cell; Area_diff > 200000 (2) S2-5. If it is determined in step S2-4 that irregular fragments exist in the solar silicon wafer, the text "damagedslice!" is output on the image I.

3. The layered and region-based visual detection algorithm for solar silicon wafer fragments according to claim 2 is characterized by: The specific process of chamfer point notch detection in step S3 is as follows: S3-1, the lower right corner area, upper right corner area, lower left corner area and upper left corner area of ​​the image I are respectively cut and binarized to obtain the lower right chamfered image CP1, upper right chamfered image CP2, lower left chamfered image CP3 and upper left chamfered image CP4 of the solar silicon wafer, and the length and width of the cut area are both 300; S3-2, find and traverse all contours in CP1 to obtain the valid contour ChamferP1; S3-3, traverse the coordinate points on ChamferP1, and transform the coordinate points according to formula (3) and formula (4); ChamferP 1,X,i = ChamferP 1,x,i + (w-300) (3) ChamferP 1,Y,i = ChamferP 1,y,i + (h-300) (4) Among them, ChamferP 1,x,i is the x-coordinate of the traversed point, ChamferP 1,y,i is the y coordinate of the traversed point, ChamferP 1,X,i and Chamfer P 1,Y,i are the x-coordinate and y-coordinate of the point after coordinate transformation, w is the width of image I, and h is the height of image I; S3-4, perform the operation of step S3-2 on CP2, CP3 and CP4 respectively, and obtain the effective contours ChamferP2, ChamferP3 and ChamferP4. In the operation of step S3-3, for the points in ChamferP2, it is only necessary to perform coordinate transformation according to formula (3), for the points in ChamferP3, it is necessary to perform coordinate transformation according to formula (4), and for the points in ChamferP4, it is not necessary to perform coordinate transformation; S3-5, traverse the coordinate points after coordinate transformation in ChamferP1, and calculate the coordinate difference between two adjacent coordinate points according to formula (5): Among them, X_diff i Indicates the difference between two adjacent coordinate points in the x direction, Y_diff i Represents the difference between two adjacent coordinate points in the y direction; S3-6, traverse the coordinate points in ChamferP1, and calculate the x-width Width_x of the current point according to formula (6) i , and calculate the average value Sumwidth_x of the x-direction width of the current point and the sum of the x-direction widths of the 20 points after the current point according to formula (7) i Finally, the difference ΔWidthx is calculated according to formula (8): i , if the difference satisfies formula (9), this point is the chamfer point Coner1 at the lower left of ChamferP1; Width_x i =|ChamferP 1,X,i -300| (6) ΔWidthx i =Widh_x i -Sumwidth_x i (8) ΔWidthx i <0.3 (9) S3-7, the principle is the same as step S3-5, traverse the coordinate points in the contour ChamferP1 and calculate the y-direction width of the current traversal point to obtain the chamfer point Coner2 on the upper right of ChamferP1; S3-8. In the contour ChamferP1, traverse each X_diff obtained by formula (5) i and Y_diff i , and calculate PDiff according to formula (10) i : PDiff i =|X_diff i |+|Y_diff i | (10) According to the calculation results, PDiff i The value is 1 or 2; S3-9, starting from PDiff5, traverse the PDiff obtained in step S3-8 in sequence i , according to rules a and b, determine whether there are abnormal points at the chamfer point: Rule a: PDiff i-3 PDiff i-2 PDiff i-1 PDiff i PDiff i+1 PDiff i+2 PDiff i+3 There is a value equal to 1 in Rule b: PDiff i-4 PDiff i-3 PDiff i-2 PDiff i-1 PDiff i+1 PDiff i+2 PDiff i+3 There is a value equal to 1 in When the traversed PDiff i When any of the above rules is met, an outlier is obtained. i ; S3-10, Outlier i It is attributed to the chamfer points Coner1 and Coner2, and then the outlier point X_Outlier in the x direction is obtained. i and the outlier point in the y direction Y_Outlier i ; S3-11, traverse X_Outlier separately i and Y_Outlier i , and calculate the difference ΔX_Outlier according to formula (11) and formula (12) i and ΔY_Outlier i : ΔX_Outlier i =|X_Outlier x,i -X_Outlier x,i-1 | (11) ΔY_Outlier i =|Y_Outlier y,i -Y_Outlier y,i-1 | (12) Among them, X_Outlier x,i X_Outlier i x coordinate, Y_Outlier y,i Y_Outlier i The y coordinate of ΔX_Outlier i or ΔY_Outlier i If it is less than 3, then the X_Outlier i and Y_Outlier i Keep it as an outlier point, otherwise remove it from the outlier point set. After the traversal is completed, the outlier points X_Outlier in the x and y directions can be obtained again i and Y_Outlier i ; S3-12, detect the abnormal gap of continuous gradient at the chamfer point; traverse the x-direction abnormal point X_Outlier i , calculate the difference between adjacent outliers according to formula (11). If there are two consecutive ΔX_outlier i If it is greater than 1, it is determined that there is a continuous abnormal gradient gap at the chamfer point. For the y-direction chamfer point, traverse the y-direction abnormal point Y_Outlier i , calculate the difference between adjacent outliers according to formula (12). If there are two consecutive ΔY_outlier i If it is greater than 1, it is determined that there is a continuous abnormal gradient gap at the chamfer point; S3-13, calculate the contour ChamferP1 area C_area by formula (13); If the value of C_area satisfies formula (14), it is judged that there is an abnormal gap in the chamfer area of ​​the solar silicon wafer; C_area>15 (14) S3-14, performing the processing of steps S3-5 to S3-13 on the contours ChamferP2, ChamferP3 and ChamferP4, and determining whether there are abnormal area gaps and abnormal continuous gradient gaps at the chamfer points; S3-15. If a gap is detected in the above step, a frame is drawn to identify the position of the gap in the image.

4. The layered and region-based visual detection algorithm for solar silicon wafer fragments according to claim 3 is characterized by: The specific process of S4 chamfer edge notch detection is as follows: S4-1, perform 2-fold super-resolution and binarization on CP1, CP2, CP3 and CP4 respectively, and then obtain a lower right chamfer image CE1, an upper right chamfer image CE2, a lower left chamfer image CE3 and an upper left chamfer image CE4; S4-2, find all contours in CE1; traverse all contours to find the valid contour ChamferEdge1, and perform the same operation on CE2, CE3 and CE4 to obtain the valid contours ChamferEdge2, ChamferEdge3 and ChamferEdge4 respectively; S4-3, similar to the principle of steps S3-3 and S3-4, perform corresponding coordinate transformation on ChamferEdge1, ChamferEdge2, ChamferEdge3 and ChamferEdge4; S4-4, traverse the coordinate points after coordinate transformation in the contour ChamferEdge1, and calculate the x and y coordinate differences between two adjacent coordinate points according to formula (15): Among them, ChamferEdge 1,x,i is the x-coordinate of the point after coordinate transformation on ChamferEdge1, 1,y,i Cx_diff is the y coordinate of the point after coordinate transformation on ChamferEdge1. i Indicates the difference between adjacent coordinate points in the x direction, Cy_diff i Indicates the difference between adjacent coordinate points in the y direction; S4-5, traverse each Cx_diff obtained in step S4-4 i and Cy_diff i , and calculate EDiff according to formula (16) i : EDiff i =|Cx_diff i |+|Cy_diff i | (16) According to the calculation results, EDiff i The value is 1 or 2; S4-6, starting from EDiff5, traverse the EDiff obtained in step S4-5 in sequence i , and judge the abnormal points according to rules c and d: Rule c: EDiff i-3 EDiff i-2 EDiff i-1 EDiff i , EDiff i+1 EDiff i+2 EDiff i+3 There is a value equal to 1 in Rule d: EDiff i-4 EDiff i-3 EDiff i-2 , EDiff i-1 , EDiff i+1 EDiff i+2 EDiff i+3 There is a value equal to 1 in When rule c or rule d is met, the anomaly point is obtained. i , Anomaly x,i Anomaly i x coordinate, Anomaly y,i Anomaly i The y coordinate of S4-7. Traversing the Anomaly i , mark Anomaly x,i Not equal to 0, Anomaly y,i Anomalies equal to 0 are positive anomalies and are marked as Anomaly. y,i Not equal to 0, Anomaly x,i The outlier point equal to 0 is a negative outlier point. If there is a negative outlier point, search counterclockwise to see if there is a positive outlier point. If so, the coordinate difference between the two outliers is further calculated according to formula (17), and the chamfer edge depression is judged according to formula (18) to see if it meets the detection condition: Here, τ a is the detection threshold, the threshold is 20, if it satisfies formula (18), it is judged that there is a concave notch at the chamfered edge of the solar cell; S4-8, for the points in the contours ChamferEdge2, ChamferEdge3 and ChamferEdge4, it is only necessary to repeat the process of steps S4-4 to S4-7 to determine whether there is a concave notch at the upper left chamfer edge of the solar silicon wafer; for the contours ChamferEdge2 and ChamferEdge3, if there are positive abnormal points, search counterclockwise to determine whether there are negative abnormal points. If there are, further determine the coordinate difference between the two abnormal points according to formula (17). If the coordinate difference satisfies formula (18), it is determined that there is a concave notch on the chamfer edge; S4-9. If a gap is detected in the above steps, the gap position is framed in the image.

5. The layered and region-based visual detection algorithm for solar silicon wafer fragments according to claim 4 is characterized by: The specific process of the straight edge gap detection in step S5 is as follows: S5-1, cropping the image I according to the chamfer point obtained in the previous step, and performing 2-fold super-resolution and high-grayscale binarization with a binarization threshold of 150 on the cropped image, to obtain images e1, e2, e3, and e4 representing the left straight edge, right straight edge, upper straight edge, and lower straight edge of the solar cell, respectively; S5-2, traverse all contours in each image in images e1, e2, e3 and e4, and then find out the valid contours in each image, which are Edge1, Edge2, Edge3 and Edge4 respectively; S5-3, traverse the points in Edge1, and calculate the difference between adjacent coordinates in the x direction Ex_diff according to formula (19) i : Ex_diff i =Edge 1,x,i+1 -Edge 1,x,i (19) Here, Edge 1,x,i is the x coordinate of the midpoint of the traversed contour Edge1, Edge 1,x,i+1 For Edge 1,x,i The x coordinates of adjacent points; when Ex_diff i If it is not 0, search counterclockwise for the first non-0 Ex_diff within a distance of 5. i If both are positive, the point is a positive outlier point. If both are negative, the point is a negative outlier point. If they do not exist or the positive and negative values ​​are opposite, the point is a normal point. The contour points that meet the abnormal conditions are recorded as outliers. i ; S5-4. Traversing the outlier points i , set the search distance threshold τ s is 30, if Exoutlier i For a positive outlier point, search the distance τ counterclockwise s Is there a negative outlier in the equation? If so, calculate the length ΔEy between the two outliers according to formula (20): ΔEy=Exoutlier y,i+n -Exoutlier y,i (20) ΔEy<τ e (21) Among them, Exoutlier y,i Exoutlier i The y coordinate of y,i+n Negative outlier point Exoutlier i+n The y coordinate of e is the detection threshold, which is set to 20. If ΔEy satisfies formula (21), it is determined that there is a left straight edge damage gap in the solar silicon wafer; S5-5, performing the operations of steps S5-3 to S5-4 on the contour Edge2 to determine whether there is a broken gap on the right straight edge of the solar silicon wafer; S5-6, traverse the points in the contour Edge3, and calculate the difference Ey_diff between adjacent coordinates in the y direction according to formula (22) i : Ey_diff i =Edge 3,y,i+1 -Edge 3,y,i (22) Among them, Edge 3,y,i is the y coordinate of the point in the contour Edge3, 3,y,i+1 For Edge 3,y,i The y coordinates of adjacent points, when the Ey_diff of the i-th contour point i If it is not 0, search counterclockwise for the first non-0 Ey_diff within a distance of 5. i If both are positive, the point is a positive outlier point. If both are negative, the point is a negative outlier point. If they do not exist or the positive and negative values ​​are opposite, the point is a normal point. The contour points that meet the abnormal conditions are recorded as outliers. i ; S5-7. Traversing the outlier point Eyoutlier i , if Eyoutlier i For a negative outlier point, search the distance τ counterclockwise s Is there a positive outlier in the system? If so, calculate the length ΔEx between the two outliers according to formula (23): ΔEx=Eyoutlier x,i+n -Eyoutlier x,i (23) ΔEx<τ e (24) Here, Eyoutlier x,i For Eyoutlier i The x-coordinate of Eyoutlier x,i+n is a positive outlier point Eyoutlier i+n If ΔEx satisfies formula (24), it is determined that there is a straight edge damage gap on the solar silicon wafer; S5-8, performing the operations of steps S5-6 to S5-7 for the contour Edge4, and determining whether there is a broken gap on the lower straight edge of the solar cell; S5-9, cropping image I according to the chamfer points obtained in the previous step, and performing 2-fold super-resolution and low-grayscale binarization with a binarization threshold of 40 on the cropped image to obtain images e5, e6, e7, and e8, and performing steps S5-3 to S5-8 on images e5, e6, e7, and e8 to determine whether there is a damaged gap on the straight edge of the solar cell; S5-10: If a gap is detected in the above steps, the gap position is framed in the image.

Images