CN113674197B - Method for dividing back electrode of solar cell - Google Patents
Method for dividing back electrode of solar cell Download PDFInfo
- Publication number
- CN113674197B CN113674197B CN202110753376.XA CN202110753376A CN113674197B CN 113674197 B CN113674197 B CN 113674197B CN 202110753376 A CN202110753376 A CN 202110753376A CN 113674197 B CN113674197 B CN 113674197B
- Authority
- CN
- China
- Prior art keywords
- electrode
- calculating
- value
- image
- points
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformation in the plane of the image
- G06T3/40—Scaling the whole image or part thereof
- G06T3/4007—Interpolation-based scaling, e.g. bilinear interpolation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/12—Edge-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/62—Analysis of geometric attributes of area, perimeter, diameter or volume
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
Abstract
The invention discloses a method for dividing a solar cell back electrode, which comprises the steps of firstly, sequentially obtaining a rough position and a more accurate position of an electrode by utilizing an edge intensity projection and template matching method; then extracting electrode edge points by adopting a threshold value method and a seed growth method; finally, a closed area is generated by using the edge points to represent the shape of the electrode. The invention can accurately collect various defect shapes, has good robustness and compatibility, and can be used for online detection.
Description
Technical Field
The invention relates to the technical field of machine vision detection, in particular to a method for segmenting a back electrode of a solar cell.
Background
Solar energy is a renewable energy source, and has the advantages of easy acquisition, sustainable utilization, environmental protection and the like. Based on the photovoltaic effect, the solar cell can convert solar energy into electric energy, so the solar cell has a great market prospect. However, the production of solar cells is a complex process, and any unexpected errors in the production process can lead to defects, which can cause significant damage to the cell. Therefore, a quality detection system is crucial to the solar cell production line.
In order to identify and segment various defects in solar cells or other industrial products, a number of machine vision based methods have been proposed over the past decade. Such as OTSU method, canny-based method, discrete Cosine Transform (DCT) method, support Vector Machine (SVM) method, but most of these methods are directed to specific types of defects, such as micro-cracks, stains, scratches, holes, etc. different from background gray scale.
Disclosure of Invention
The present invention is directed to solving the above-mentioned drawbacks of the prior art, and provides a method for dividing a back electrode of a solar cell.
The purpose of the invention can be achieved by adopting the following technical scheme:
a method of dividing a solar cell back electrode, the method comprising the steps of:
s1, coarse positioning: roughly positioning 54 back electrodes in the silicon wafer image of the cell by using a projection method and a polynomial fitting method, and cutting to obtain 54 roughly positioned images I with the size of 128 multiplied by 158 r R =1,2 …, each coarse positioning image containing one electrode;
s2, fine positioning: calculating rough positioning image I based on Sobel operator r Gradient map of (1) c Calculating a gradient map I by maximum projection c Maximum value P per row and column x And P y To P x And P y Calculating the weighted average value and obtaining an electrode fine positioning image based on the weighted average value and recording the electrode fine positioning image as I d ；
S3, edge segmentation: from the fine positioning image I d Selecting three threshold values to obtain a fine positioning image I d Corresponding binary image I db Selecting target edge points from the binary image I by using a seed growing method db Further performing edge segmentation on the electrode to obtain an edge segmentation result G 123 ；
S4, area generation: result of edge segmentation G 123 Drawing a closed contour to obtain a finally divided solar cell back electrode M 3 。
Further, the process of coarsely positioning the electrode in step S1 is as follows:
s1.1, calculating gradients of the image of the battery silicon wafer in the horizontal direction and the vertical direction by using a Sobel operator, and recording the gradients as G x And G y ：
Wherein, I r Is a coarse positioning image;
s1.2, adopting a mean projection method, and calculating G x Mean value of each column G x Projected onto the Y axis, denoted M y By calculating G y Mean value of each row G y Projected onto the X-axis, denoted M x ；
S1.3, fitting M by utilizing third-order polynomial x And the fitting result is recorded as M fx And defining an intermediate variable M p Comprises the following steps:
M p ＝maximum(M x -M fx ,0)
wherein maximum () represents taking a larger value;
s1.4, defining a threshold value T m ：
Wherein n is p Is M p Number of middle and larger than 0, M p (i T ) Is a one-dimensional vector M p Ith T Value i T Range of values from 1 to N x Writing i T ∈{1,2,…,N x }，N x Is M x Length of (d);
s1.5, using threshold T m Will M p Carry out binarization to obtain B x ＝[B x (1),B x (2),…,B x (i T ),…]：
Wherein B is x (i T ) Is a one-dimensional vector B x I th of (1) T An element;
s1.6, obtaining an electrode template T from a flawless battery silicon wafer sample p Will T p As a sliding window, in B x Up sliding, calculating T p And B x The position of the maximum intersection can obtain the row position of the electrode;
s1.7, repeating the steps S1.3-S1.6 to perform coarse positioning of electrode row positions, and cutting the silicon wafer according to the obtained electrode row and column positions to obtain a coarse positioning image I r 。
Further, the process of precisely positioning the electrode in step S2 is as follows:
s2.1, calculating a coarse positioning image I by using Sobel operator r Gradient diagram of (1), denoted as I c ；
S2.2, calculating a gradient map I by adopting a maximum projection method c The maximum value of each row is denoted as P x Calculating a gradient map I c The maximum value of each column is denoted as P y ；
S2.3 according to P y Calculating a weighted average value, denoted A w Obtained by the following formula:
wherein, P y (j) Is a one-dimensional vector P y The j element of (2), w j ＝abs(j-n y /2)，j∈{1,2,…,n y }，n y Is P y Abs () represents an absolute value operation;
s2.4, adding A w And P y Is represented in the same coordinate axis, A w And P y The area with the largest area in the closed area formed by the intersection is the position of the electrode,two end points P of the region 1 And P 2 To P y Moving the two sides to the nearest minimum value point;
s2.5, to P x Repeating the steps S2.3 and S2.4 to finally obtain a bounding rectangle of the electrode, and cutting out the fine positioning image I from the original image by using the bounding rectangle d 。
Further, the process of performing edge segmentation on the electrode in step S3 is as follows:
s3.1, obtaining a gradient chart I in the horizontal direction and the vertical direction by utilizing a Sobel operator x And I y ；
S3.2, after being blurred by a 5 multiplied by 5 Gaussian kernel, the gradient chart I is subjected to x And I y Refining the edges using non-maxima suppression;
s3.3, a gradient chart I y Normalizing to 0-255, and calculating histogram H y Then H is introduced y Transition to a probability range of 0 to 1, H y Is recorded as A sy Calculated from the following equation:
wherein H y (j H ) Is a one-dimensional vector H y J (d) of H Value i A ∈{1,2,…,255}；
S3.4, the threshold T meets the following conditions:
A sy (i A +1)>p&&A sy (i A )≤p
wherein the content of the first and second substances,&&representing a logical AND operation, p is a given scaling factor, where I is the ratio of the sum of the coefficients for an h x w image y Proportional coefficient p of y (i p ) Is defined as:
similarly, I x Proportional coefficient p of x (j p ) Is defined as:
setting i p Equal to 2, 3, 4,j, respectively p Respectively equal to 3, 4 and 5, and respectively obtain I according to the design of p y And I x Three respective thresholds, from high threshold to low threshold, from I y Obtaining 3 binary images, respectively recording as B y1 、B y2 And B y3 From I x Obtaining 3 binary images, respectively recording as B x1 、B x2 And B x3 ；
S3.5, combining the two corresponding binary images by using bitwise or operation, and connecting small gaps by using morphological closing operation to obtain G 1 、G 2 And G 3 Three images, formulated as:
G i ＝(B yi ∪B xi )⊙K
wherein U represents a bitwise OR operation, indicates a morphological OFF operation, K is a 3X 3 structural element, B xi Take values of B respectively x1 、B x2 And B x3 ，B yi Each value is B y1 、B y2 And B y3 ；
S3.6、G 1 In each column of (1), p f And p l Corresponding to the positions of the first and last white points, respectively. Let p be f And p l The points in between are 1 and the other points are 0, and the result is recorded as M c (ii) a For G 1 Each row in (a) performs the same operation to obtain M r Calculating the union of the two, M or ＝M c ∪M r Obtaining a mask M or Using a mask M or Elimination of G 2 And G 3 The noise point in (1) is obtained as G 2 ' and G 3 ′：
G 2 ′＝G 2 ·M or
G 3 ′＝G 3 ·M or
Wherein, represents a dot product operation;
s3.7, selecting target edge points by adopting a method based on seed growth:
G s ＝G 2 ′-G 12
G 3 ″＝G 3 ′-G s
wherein the content of the first and second substances,
denotes the seed growth Process, G 123 Is the final edge segmentation result, G 12 、G s 、G 3 "are different intermediate variables of the calculation process.
Further, the region generation process in step S4 is as follows:
s4.1, dividing the outline into an upper part, a lower part, a left part and a right part which are respectively expressed as P u 、P d 、P l And P r ；
S4.2、P u And P d Respectively record G 123 The position of the first and last white point in each column, the two white points being respectively marked
And
if it is not
Connected, i.e. the points between the two are all white points, mark f ud Is set to 1, otherwise is set to 0;
s4.3, utilization flag f ud And curve modification method, for P u Use minimumValue function modification, for P d Obtaining updated P using maximum function modification u And P d ；
S4.4, adding P u Eliminating P by convolution operation with one-dimensional vector k u The process of the method is as follows:
wherein the content of the first and second substances,
representing a convolution operation;
s4.5, setting a threshold value T u If g is u (i)>T u If the mark is 1, the mark of the corresponding position is set to 0,g otherwise u (i) Is g u The ith element of (1), finally using a mark and using a linear interpolation method to P u The information is further updated according to the received information,
s4.6, to P d 、P l And P r Repeating the steps S4.4 and S4.5 to carry out the same treatment;
s4.7, order P u And P d Point in between generates mask M of 1 1 Let P l And P r Point in between generates mask M of 1 2 And calculating the intersection of the two:
M 3 ＝M 1 ∩M 2
wherein, n represents a bitwise AND operation, M 3 The final result is shown, representing the area of the electrode.
Further, in the segmentation method, the curve C is modified by using the flag vector F, and the process is as follows: if in a continuous interval, e.g. [ a ] 1 ,a k ]Wherein a is 1 ，a k Is the two endpoints of the interval, and a k >a 1 To any i F ∈[a 1 ,a k ]Having F (i) F ) =1, for any
With F (i) F )＝0，F(i F ) I < th > of the flag vector F F One element, using adjacent points C (a) at both ends of the interval 1 -1)，C(a k -1) by a function f (C (a) 1 -1),C(a k -1)) to calculate a substitute value for the whole interval instead of the value of the curve C over this interval, wherein the function f is chosen as one of a minimum function, a maximum function and a linear interpolation function according to different requirements.
Compared with the prior art, the invention has the following advantages and effects:
(1) According to the invention, the electrodes in the silicon wafer are all cut out by using coarse positioning and fine positioning, so that most of uneven and complex backgrounds are removed;
(2) The most obvious feature of the electrode is its strong gradient strength around or inside. Therefore, the gradient is the main information for dividing the electrode, and in general, the intensity range of the gradient map is large, and it is difficult to select a threshold value to distinguish between the target and the noise. According to the method, firstly, the shape prior of an electrode is utilized, three thresholds are selected on the basis of a histogram to obtain a corresponding binary image, and then a seed growing method is used for selecting target edge points to obtain an edge segmentation result with a good effect;
(3) The invention provides a method for modifying a curve by using a mark in region generation, and a final segmentation detection result of an electrode is obtained on the basis of an edge segmentation result.
Drawings
Fig. 1 is a flow chart of a solar cell back electrode splitting method disclosed in the present invention;
FIG. 2 is a diagram illustrating an original image and a coarse positioning result according to the present invention;
FIG. 3 is a schematic diagram of the main variables of the coarse positioning algorithm of the present invention;
FIG. 4 is a schematic diagram of the result of the fine positioning algorithm of the present invention;
fig. 5 is a schematic diagram of intermediate variables in the edge segmentation algorithm of the present invention, in which fig. 5 (a), 5 (b), and 5 (c) show binary images obtained according to three thresholds, fig. 5 (d) shows the final result of edge segmentation, fig. 5 (e) and 5 (f) show masks obtained according to fig. 5 (a) and 5 (b), respectively, fig. 5 (g) shows the union of fig. 5 (e) and 5 (f), i.e., the final mask, and fig. 5 (h) shows the final electrode region;
FIG. 6 is an exemplary graph of the present invention using a flag modification curve;
FIG. 7 is a schematic diagram of a four-part contour processing according to the present invention, wherein FIG. 7 (a) is a schematic diagram of an upper and lower part contour processing; FIG. 7 (b) is a schematic front and rear view of the outline processing of the left and right portions;
fig. 8 is a comparison graph of segmentation results of different methods, and in fig. 8, the following are sequentially performed from the first row to the last row: test images, labels, results from DCT-T, CED, DCT-SVM, FCN and PM, respectively.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Examples
The embodiment mainly provides an intelligent segmentation technology for the back electrode of the solar cell, and the rough position and the accurate position of the electrode are sequentially obtained by utilizing an edge intensity projection and template matching method; and then extracting electrode edge points by adopting a threshold value method and a seed growing method, and finally generating a closed region by utilizing the edge points to represent the shape of the electrode.
Fig. 1 is a flowchart of a method for dividing a back electrode of a solar cell according to the present embodiment, and the following description is made by using an embodiment. A method for dividing a solar cell back electrode comprises the following specific steps:
s1, coarse positioning: roughly positioning 54 back electrodes in the battery silicon wafer image by using a projection method and a polynomial fitting method, and cutting to obtain 54 back electrodesCoarse positioning image I with size of 128 x 158 r R =1,2 …, each of the coarse positioning images contains an electrode, the coarse positioning result obtained in this step is shown in fig. 2, and it is apparent from fig. 2 that all the minimum rectangular positions are roughly positioned;
s2, fine positioning: calculating rough positioning image I based on Sobel operator r Gradient map of (1) c Calculating a gradient map I by maximum projection c Maximum value P per row and column x And P y To P x And P y Calculating a weighted average and obtaining a gradient histogram I containing the electrode fine positioning result based thereon d P in FIG. 4 x And P y All the results show that the final fine positioning result is also shown in fig. 4, the small rectangle in fig. 4 contains the whole electrode, meanwhile, almost no excessive background exists, and the fine positioning effect is very obvious;
s3, edge segmentation: from the fine positioning image I d Selecting three threshold values from the gradient histogram to obtain a fine positioning image I d Corresponding binary image
Selecting target edge points from binary images using seed growth methods
Further performing edge segmentation on the electrode to obtain an edge segmentation result G 123 Fig. 5 (a), 5 (b), and 5 (c) are preliminary edge segmentation images obtained by performing corresponding combination and morphological closing operations based on the selected three thresholds. FIG. 5 (d) is a final edge segmentation result obtained based on the preliminary edge segmentation result using a seed growth method;
s4, area generation: result of edge segmentation G 123 Drawing a closed contour to obtain a finally divided solar cell back electrode M 3 The effect is shown in fig. 5 (h), the result of edge segmentation and region generation is further improved based on the fine positioning, and the maximum possible effect is achieved under the condition of keeping the electrodeSome backgrounds were successfully removed.
The invention provides a solar cell back electrode segmentation method, which is further realized by the following technical scheme:
in this embodiment, the specific process of coarsely positioning the electrode in step S1 is as follows:
s1.1, calculating gradients of the image of the battery silicon wafer in the horizontal direction and the vertical direction by using a Sobel operator, and recording the gradients as G x And G y ：
Wherein, I r For roughly positioning the image, the line and column positions of the electrodes can be roughly determined by utilizing a Sobel operator due to the strong gradient strength around the electrodes;
s1.2, adopting a mean value projection method to calculate G x Mean value of each column and x projected onto the Y-axis, the result is noted as M y By calculating G y Average of each row and y projected onto the X-axis, the result is noted as M x The projection method is a common method for converting a two-dimensional matrix into a one-dimensional vector;
s1.3, fitting M by utilizing third-order polynomial x And the fitting result is recorded as M fx In practice to M x Sampling at intervals of 10 to reduce the computational burden and defining an intermediate variable M p Comprises the following steps:
M p ＝maximum(M x -M fx ,0)
wherein maximum () represents taking a larger value;
s1.4, defining a threshold value T m ：
Wherein n is p Is M p Number of middle and larger than 0, M p (i T ) Is a one-dimensional vector M p Ith T Value i T Range of values from 1 to N x Writing i T ∈{1,2,…,N x }，N x Is M x Length of (d);
s1.5, using threshold T m A 1, M p Carry out binarization to obtain B x ＝[B x (1),B x (2),…,B x (i T ),…]：
Wherein B is x (i T ) Is a one-dimensional vector B x I th of (1) T And (4) each element. In B x The region of 1 is the possible electrode column position, B x The results are shown in FIG. 3;
s1.6, obtaining an electrode template T from a defect-free battery silicon wafer sample p Will T p As a sliding window, in B x Up sliding, calculating T p And B x The position of the maximum intersection can obtain the row position of the electrode;
s1.7, repeating the steps S1.3-S1.6 to perform coarse positioning of electrode row positions, and cutting the silicon wafer according to the obtained electrode row and column positions to obtain a coarse positioning image I r 。
In this embodiment, the specific process of performing the fine positioning on the electrode in step S2 is as follows:
s2.1, calculating a coarse positioning image I by using Sobel operator r Gradient diagram of (1), denoted as I c ；
S2.2, calculating a gradient map I by adopting a maximum projection method c The maximum value of each row is denoted as P x Calculating a gradient map I c The maximum value of each column is marked as P y ；
S2.3 according to P y Calculating a weighted average value, denoted A w Obtained by the following formula:
wherein, P y (j) Is a one-dimensional vector P y The j element of (2), w j ＝abs(j-n y /2)，j∈{1,2,…,n y }，n y Is P y Abs () represents an absolute value operation;
s2.4, mixing A w And P y Is represented in the same coordinate axis, A w And P y The area with the largest area in the closed area formed by intersection is the position of the electrode, and two end points P of the area 1 And P 2 To P y Moving the two sides to the nearest minimum value point;
s2.5, to P x Repeating the steps S2.4 and S2.4 to finally obtain a surrounding rectangle of the electrode, and cutting out the fine positioning image I from the original image by using the surrounding rectangle d . This allows the electrodes to be positioned within a small area, thus removing most of the uneven, complex background and further bedding the subsequent segmentation.
In this embodiment, the process of edge segmentation of the electrode in step S3 is as follows:
s3.1, obtaining a gradient chart I in the horizontal direction and the vertical direction by utilizing a Sobel operator x And I y ；
S3.2, after being blurred by a 5 multiplied by 5 Gaussian kernel, the gradient chart I is subjected to x And I y Refining the edges using non-maxima suppression;
s3.3, gradient chart I y Normalizing to 0-255, and calculating histogram H y Then, H is introduced y Transition to the probability range of 0 to 1. H y Is recorded as A sy Calculated from the following equation:
wherein N is y (j H ) Is a one-dimensional vector H y J (d) of H Value i A ∈{0,1,…,255}；
S3.4, the threshold T meets the following conditions:
A sy (i A +1)>p&&A sy (i A )≤p
wherein the content of the first and second substances,&&representing a logical AND operation, p is a given scaling factor, where I is the ratio of the sum of the coefficients for an h x w image y Proportional coefficient p of y (i p ) Is defined as:
similarly, I x Proportional coefficient p of x (j p ) Is defined as:
setting i p Equal to 2, 3, 4,j, respectively p Respectively equal to 3, 4 and 5, and respectively obtain I according to the design of p y And I x Three respective thresholds, from high threshold to low threshold, from I y Obtaining 3 binary images, respectively recording as B y1 、B y2 And B y3 From I x Obtaining 3 binary images, respectively recording as B x1 、B x2 And B x3 ；
S3.5, combining the two corresponding binary images by using bitwise or operation, and connecting small gaps by using morphological closing operation to obtain G 1 、G 2 And G 3 Three binary images are expressed by a formula as follows:
G i ＝(B yi ∪B xi )⊙K
wherein U represents an bitwise or operation, indicates a morphological close operation, and K is a 3 × 3 structural element;
s3.6, binary image G 1 In each column of (1), p f And p l Corresponding to the positions of the first and last white points, respectively. Let p be f And p l The points in between are 1 and the other points are 0, and the result is recorded as M c . For G 1 Each row in (a) performs the same operation to obtain M r . Calculate the union of the two, M or ＝M c ∪M r A mask M is obtained or Using a mask M or Elimination of G 2 And G 3 The noise point in (1) is obtained as G 2 ' and G 3 ′：
G 2 ′＝G 2 ·M or
G 3 ′＝G 3 ·M or
Where, represents a dot product operation;
s3.7, selecting target edge points by adopting a method based on seed growth:
G s ＝G 2 ′-G 12
G 3 ″＝G 3 ′-G s
wherein the content of the first and second substances,
denotes the seed growth process, G 123 Is the final edge segmentation result, shown in FIG. 5 (d), G 12 、G s 、G 3 "are different intermediate variables of the calculation process.
In this embodiment, the process of generating the area in step S4 is as follows:
s4.1, dividing the outline into an upper part, a lower part, a left part and a right part which are respectively expressed as P u 、P d 、P l And P r ；
S4.2、P u And P d Respectively record G 123 The first and last white point of each columnThe two white dots are respectively recorded as
And
if it is not
And with
Connected, i.e. the points between the two are all white points, mark f ud Is set to 1, otherwise is set to 0;
s4.3, utilization flag f ud And curve modification method, for P u Modified using a minimum function, P d Obtaining updated P using maximum function modification u And P d An exemplary diagram of curve modification using flags is shown in FIG. 6, where FIG. 6 illustrates the process of modifying curve C using a linear interpolation function y =1.25x +2, but here, the minimum function and the maximum function are selected for modification;
s4.4, adding P u Eliminating P by convolution operation with one-dimensional vector k u The process of the method is as follows:
wherein the content of the first and second substances,
where convolution is represented, the value of k is set to k = [ -1,2, -1]；
S4.5, setting a threshold value T u If g is u (i g )>T u If so, setting the mark of the corresponding position to be 1, otherwise, setting the mark to be 0,g u (i g ) Is g u I th of (1) g An element of, wherein T u In the experiment, P was interpolated by linear interpolation to 10 u Go toUpdating;
s4.6, to P d 、P l And P r Repeating the steps S4.4 and S4.5 to carry out the same treatment;
s4.7, order P u And P d Point in between generates mask M of 1 1 Let P l And P r Point in between generates mask M of 1 2 And calculating the intersection of the two:
M 3 ＝M 1 ∩M 2
wherein, n represents a bitwise AND operation, M 3 The final result, representing the area of the electrode, is shown in fig. 5 (h);
in this example, 50 defective images and 50 non-defective images were selected as the test set, and the training set contained 25 defective images and 25 non-defective images. All methods in the examples were performance evaluated using a test set, and four methods were compared to the method proposed by the present invention, the first method being OTSU plus DCT, denoted as DCT-T. The second method is the Canny edge detection method, denoted CED, the joint DCT and Support Vector Machine (SVM) classifier is the third method, and the Fully Connected Network (FCN) is the last method. The results of the different measurements are shown in table 1 below:
TABLE 1 evaluation index Table of two forms
The metrics are defined as follows:
wherein FP is false positive and true positive in the experiment for TP, FN and TN are false negative and true negative respectively. Thus, FPR is the proportion of experimentally negative pixels that are misjudged positive, FNR is the proportion of experimentally positive pixels that are misjudged negative, and MAE is the proportion of misclassified pixels to all pixels. In addition, three evaluation indexes are calculated in two forms, one is in a pixel-wise form (pixel-wise form) for evaluating the pixel-wise accuracy of the segmentation method, and the other is in a sample-wise form (sample-wise form) for evaluating the ability of the method to identify defects. Using the segmentation results, the presence or absence of defects in the sample can be identified. In the sample-level format, FPR is the proportion of negative samples that are misjudged as positive samples, FNR is the proportion of positive samples that are misjudged as negative samples, and MAE is the proportion of misclassified samples to all samples.
As can be seen from table 1, PM performance is generally better than all comparative methods in the experiment.
The above embodiments are preferred embodiments of the present invention, but the present invention is not limited to the above embodiments, and any other changes, modifications, substitutions, combinations, and simplifications which do not depart from the spirit and principle of the present invention should be construed as equivalents thereof, and all such modifications are intended to be included in the scope of the present invention.
Claims (5)
1. A method for dividing a back electrode of a solar cell, the method comprising:
s1, coarse positioning: roughly positioning 54 back electrodes in the battery silicon wafer image by using a projection method and a polynomial fitting method, and cutting to obtain 54 roughly positioned images I with the size of 128 multiplied by 158 r R =1,2 …, each coarse positioning image containing one electrode;
s2, fine positioning: calculating rough positioning image I based on Sobel operator r Gradient map I of c Calculating a gradient map I by maximum projection c Maximum value P per row and column x And P y To P x And P y Calculating the weighted average value and obtaining an electrode fine positioning image based on the weighted average value and recording the electrode fine positioning image as I d ；
S3, edge segmentation: from the fine positioning image I d Selecting three threshold values to obtain a fine positioning image I d Corresponding binary image
Selecting target edge points from binary images using seed growth methods
Further performing edge segmentation on the electrode to obtain an edge segmentation result G 123 ；
S4, region generation: result of edge segmentation G 123 Drawing a closed contour to obtain a finally divided solar cell back electrode M 3 ；
Wherein, the process of coarsely positioning the electrode in the step S1 is as follows:
s1.1, calculating gradients of the image of the cell silicon wafer in the horizontal direction and the vertical direction by using a Sobel operator, and recording the gradients as G x And G y ：
Wherein, I r Is a coarse positioning image;
s1.2, adopting a mean projection method, and calculating G x Mean value of each column G x Projected onto the Y axis, denoted M y By calculating G y Mean value of each row G y Projection onto X-axisAbove, is marked as M x ；
S1.3, fitting M by utilizing third-order polynomial x And the fitting result is recorded as M fx And defining an intermediate variable M p Comprises the following steps:
M p ＝maximum(M x -M fx ,0)
wherein maximum () represents taking a larger value;
s1.4, defining a threshold value T m ：
Wherein n is p Is M p Number of middle and larger than 0, M p (i T ) Is a one-dimensional vector M p Ith (i) T Value i T Range from 1 to N x Writing i T ∈{1,2,…,N x }，N x Is M x Length of (d);
s1.5, using threshold T m A 1, M p Carry out binarization to obtain B x ＝[B x (1),B x (2),…,B x (i T ),…]：
Wherein B is x (i T ) Is a one-dimensional vector B x I th of (1) T An element;
s1.6, obtaining an electrode template T from a flawless battery silicon wafer sample p Will T p As a sliding window, in B x Up sliding, calculating T p And B x The position of the maximum intersection can obtain the position of the row where the electrode is positioned;
s1.7, repeating the steps S1.3-S1.6 to carry out coarse positioning on electrode row positions, and cutting the silicon wafer according to the obtained electrode row and column positions to obtain a coarse positioning image I r 。
2. The method for dividing the back electrode of the solar cell according to claim 1, wherein the step S2 of finely positioning the electrode comprises the following steps:
s2.1, calculating a coarse positioning image I by using Sobel operator r Gradient diagram of (2), marked as I c ；
S2.2, calculating a gradient map I by adopting a maximum projection method c The maximum value of each row is denoted as P x Calculating a gradient map I c The maximum value of each column is marked as P y ；
S2.3 according to P y Calculating a weighted average value, denoted A w Obtained by the following formula:
wherein, P y (j) Is a one-dimensional vector P y The j element of (2), w j ＝abs(j-n y /2)，j∈{1,2,…,n y }，n y Is P y Abs () represents an absolute value operation;
s2.4, adding A w And P y Is represented in the same coordinate axis, A w And P y The area with the largest area in the closed area formed by intersection is the position of the electrode, and two end points P of the area 1 And P 2 To P y Moving the two sides to the nearest minimum value point;
s2.5, to P x Repeating steps S2.3 and S2.4 to perform the same processing, finally obtaining a bounding rectangle of one electrode, and cutting out the fine positioning image I from the original image by using the bounding rectangle d 。
3. The method as claimed in claim 1, wherein the step S3 of edge-dividing the electrode comprises:
s3.1, obtaining a gradient chart I in the horizontal direction and the vertical direction by utilizing a Sobel operator x And I y ；
S3.2, after being blurred by a 5 multiplied by 5 Gaussian kernel, the gradient image I is processed x And I y Refining the edges using non-maxima suppression;
s3.3, a gradient chart I y Normalizing to 0-255, and calculating histogram H y Then, H is introduced y Transition to a probability range of 0 to 1, H y Is recorded as A sy Calculated from the following equation:
wherein H y (j H ) Is a one-dimensional vector H y J (d) of H Value i A ∈{1,2,…,255}；
S3.4, the threshold T meets the following conditions:
A sy (i A +1)>p&&A sy (i A )≤p
wherein, the first and the second end of the pipe are connected with each other,&&representing a logical AND operation, p is a given scaling factor, where I is the ratio of the sum of the coefficients for an h x w image y Proportional coefficient p of y (i p ) Is defined as:
similarly, I x Proportional coefficient p of x (j p ) Is defined as:
setting i p Equal to 2, 3, 4,j, respectively p Respectively equal to 3, 4 and 5, and respectively obtain I according to the design of p y And I x Three respective thresholds, from high threshold to low threshold, from I y Obtaining 3 binary images, respectively recording as B y1 、B y2 And B y3 From I x Obtaining 3 binary images, respectively recordingIs B x1 、B x2 And B x3 ；
S3.5, combining the two corresponding binary images by using bitwise or operation, and connecting small gaps by using morphological closing operation to obtain G 1 、G 2 And G 3 Three images, formulated as:
G i ＝(B yi ∪B xi )⊙K
wherein U represents a bitwise OR operation, indicates a morphological OFF operation, K is a 3X 3 structural element, B xi Take values of B respectively x1 、B x2 And B x3 ，B yi Each value is B y1 、B y2 And B y3 ；
S3.6、G 1 In each column of (a), p f And p l Let p correspond to the positions of the first and last white dots, respectively f And p l The points in between are 1, the other points are 0, and the results are recorded as M c (ii) a For G 1 Each row in (a) performs the same operation to obtain M r Calculating the union of the two, M or ＝M c ∪M r Obtaining a mask M or Using a mask M or Elimination of G 2 And G 3 The noise point in (1) is obtained as G 2 ' and G 3 ′：
G 2 ′＝G 2 ·M or
G 3 ′＝G 3 ·M or
Where, represents a dot product operation;
s3.7, selecting target edge points by adopting a method based on seed growth:
G s ＝G 2 ′-G 12
G 3 ″＝G 3 ′-G s
wherein, the first and the second end of the pipe are connected with each other,
denotes the seed growth Process, G 123 Is the final edge segmentation result, G 12 、G s 、G 3 "are different intermediate variables of the calculation process.
4. The method for dividing a back electrode of a solar cell according to claim 3, wherein the region generation process in step S4 is as follows:
s4.1, dividing the outline into an upper part, a lower part, a left part and a right part which are respectively expressed as P u 、P d 、P l And P r ；
S4.2、P u And P d Respectively record G 123 The position of the first and last white point in each column, the two white points being respectively marked
And
if it is not
Connected, i.e. the points between the two are all white points, mark f ud Is set to 1, otherwise is set to 0;
s4.3, utilization flag f ud And curve modification method, for P u Using minimum function modification on P d Obtaining updated P using maximum function modification u And P d ；
S4.4, adding P u Eliminating P by convolution operation with one-dimensional vector k u The process of the method is as follows:
wherein, the first and the second end of the pipe are connected with each other,
representing a convolution operation;
s4.5, setting a threshold value T u If g is u (i)>T u If so, setting the mark of the corresponding position to be 1, otherwise, setting the mark to be 0,g u (i) Is g u The ith element of (1), finally using a mark and using a linear interpolation method to P u The information is further updated according to the received information,
s4.6, to P d 、P l And P r Repeating the steps S4.4 and S4.5 to carry out the same treatment;
s4.7, order P u And P d Point in between generates mask M of 1 1 Let P l And P r Point in between generating mask M for 1 2 And calculating the intersection of the two:
M 3 ＝M 1 ∩M 2
wherein, n represents a bitwise AND operation, M 3 The final result is shown, representing the area of the electrode.
5. The method for dividing the back electrode of the solar cell according to claim 4, wherein the curve C is modified by using the flag vector F as follows: in a continuous interval [ a ] 1 ,a k ]Wherein a is 1 ，a k Is two endpoints of a range, and a k >a 1 To any i F ∈[a 1 ,a k ]Having F (i) F ) =1, for any
With F (i) F )＝0，F(i F ) I-th representing a flag vector F F One element, using adjacent points C (a) at both ends of the interval 1 -1)，C(a k -1) by a function f (C (a) 1 -1),C(a k -1)) to calculate a substitute value for the whole interval instead of the value of the curve C over this interval, wherein the function f is chosen as one of a minimum function, a maximum function and a linear interpolation function according to different requirements.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110753376.XA CN113674197B (en) | 2021-07-02 | 2021-07-02 | Method for dividing back electrode of solar cell |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110753376.XA CN113674197B (en) | 2021-07-02 | 2021-07-02 | Method for dividing back electrode of solar cell |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113674197A CN113674197A (en) | 2021-11-19 |
| CN113674197B true CN113674197B (en) | 2022-10-04 |
Family
ID=78538503
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110753376.XA Active CN113674197B (en) | 2021-07-02 | 2021-07-02 | Method for dividing back electrode of solar cell |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113674197B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102044071A (en) * | 2010-12-28 | 2011-05-04 | 上海大学 | Single-pixel margin detection method based on FPGA |
| EP3174008A1 (en) * | 2015-11-26 | 2017-05-31 | Thomson Licensing | Method and apparatus for determining a sharpness metric of an image |
| CN112163587A (en) * | 2020-09-30 | 2021-01-01 | 北京环境特性研究所 | Feature extraction method and device of target object and computer readable medium |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101540040B (en) * | 2008-03-21 | 2012-12-12 | 深圳迈瑞生物医疗电子股份有限公司 | Method and device for automatically detecting boundary of beam-limiting device |
| CN101465001B (en) * | 2008-12-31 | 2011-04-13 | 昆山锐芯微电子有限公司 | Method for detecting image edge based on Bayer RGB |
| US20130022288A1 (en) * | 2011-07-20 | 2013-01-24 | Sony Corporation | Image processing apparatus and method for reducing edge-induced artefacts |
| CN103925893B (en) * | 2014-04-17 | 2017-01-11 | 广东正业科技股份有限公司 | Quality detection method of battery cells |
| US10345046B2 (en) * | 2017-05-25 | 2019-07-09 | Northeastern University | Fault diagnosis device based on common information and special information of running video information for electric-arc furnace and method thereof |
| CN110544231B (en) * | 2019-07-24 | 2021-05-11 | 华南理工大学 | Lithium battery electrode surface defect detection method based on background standardization and centralized compensation algorithm |
| CN112819827B (en) * | 2021-04-16 | 2021-08-06 | 高视科技(苏州)有限公司 | LED electrode offset detection method and device based on polar coordinate transformation and storage medium |
-
2021
- 2021-07-02 CN CN202110753376.XA patent/CN113674197B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102044071A (en) * | 2010-12-28 | 2011-05-04 | 上海大学 | Single-pixel margin detection method based on FPGA |
| EP3174008A1 (en) * | 2015-11-26 | 2017-05-31 | Thomson Licensing | Method and apparatus for determining a sharpness metric of an image |
| CN112163587A (en) * | 2020-09-30 | 2021-01-01 | 北京环境特性研究所 | Feature extraction method and device of target object and computer readable medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113674197A (en) | 2021-11-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110866924B (en) | Line structured light center line extraction method and storage medium | |
| CN109507192B (en) | Magnetic core surface defect detection method based on machine vision | |
| CN105913396A (en) | Noise estimation-based image edge preservation mixed de-noising method | |
| CN109241973B (en) | Full-automatic soft segmentation method for characters under texture background | |
| WO2021109697A1 (en) | Character segmentation method and apparatus, and computer-readable storage medium | |
| CN114529459B (en) | Method, system and medium for enhancing image edge | |
| CN108830857B (en) | Self-adaptive Chinese character copy label image binarization segmentation method | |
| CN111754538B (en) | Threshold segmentation method for USB surface defect detection | |
| CN111598856A (en) | Chip surface defect automatic detection method and system based on defect-oriented multi-point positioning neural network | |
| CN110738139A (en) | NIN license plate recognition method fusing Min-Max targets | |
| CN111476758A (en) | Defect detection method and device for AMO L ED display screen, computer equipment and storage medium | |
| CN111738931B (en) | Shadow removal algorithm for aerial image of photovoltaic array unmanned aerial vehicle | |
| CN112668725A (en) | Metal hand basin defect target training method based on improved features | |
| CN112308872A (en) | Image edge detection method based on multi-scale Gabor first-order derivative | |
| CN115880683B (en) | Urban waterlogging ponding intelligent water level detection method based on deep learning | |
| CN113674197B (en) | Method for dividing back electrode of solar cell | |
| CN108053402B (en) | Defect image segmentation method | |
| CN116433978A (en) | Automatic generation and automatic labeling method and device for high-quality flaw image | |
| CN113763404B (en) | Foam image segmentation method based on optimization mark and edge constraint watershed algorithm | |
| CN112101058B (en) | Automatic identification method and device for test paper bar code | |
| CN111915633A (en) | Casting edge detection method and device | |
| Han et al. | Single image dehazing method via sky-regions segmentation and dark channel prior | |
| CN117541582B (en) | IGBT insulation quality detection method for high-frequency converter | |
| CN109271986B (en) | Digital identification method based on Second-Confirm | |
| CN116818778B (en) | Rapid and intelligent detection method and system for automobile parts |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |