TWI445944B - Method for classifying color of a solar cell - Google Patents

Description

Solar cell wafer color separation method

BACKGROUND OF THE INVENTION 1. Field of the Invention This invention relates to a method of color separation of solar cell wafers, and more particularly to a method of efficiently classifying the color of a solar cell wafer surface with an automatic learning system.

In recent years, energy and environmental issues have received attention, making countries all over the world develop new and less polluting energy sources, such as wind, tide, geothermal, solar, and biomass energy. Among them, solar energy has become widely used due to its wide range of applications and mature development. One of the most promising new alternative energy sources.

The use of solar energy generally converts the energy contained in sunlight into electrical energy through photoelectric conversion through a solar cell wafer. Solar cell wafers can be divided into twin crystal solar cell wafers and thin film solar cell wafers according to different processes. Among them, twin crystal solar cell wafers have the longest development and the most mature technology. The basic structure of a twinned solar cell wafer is a combination of a P-type semiconductor and an N-type semiconductor. When sunlight illuminates a solar cell wafer, its light energy excites electrons in the germanium atom to form a convection of electrons and holes. The generated electrons and holes are attracted by the P-type semiconductor and the N-type semiconductor to be collected on different sides by the built-in potential, and the electrodes can extract electrons and holes to generate electric energy.

Since the solar cell chip generates electricity by receiving sunlight on its surface, whether the surface of the solar cell wafer can effectively absorb the light source will become the key to improving energy conversion efficiency. In addition, the surface of the solar cell wafer has an anti-reflection layer which can be used to reduce the reflectance of light irradiated to the surface, in other words, to increase the light absorption efficiency. The thickness of the anti-reflection layer will affect the efficiency of anti-reflection, and the thickness of the anti-reflection layer can be distinguished by the color of the surface of the solar cell wafer. Generally, the thicker the anti-reflection layer, the lighter the color, and vice versa. The difference in color depends mainly on the control of the process. Depending on the depth and uniformity of the color, it is also possible to judge whether the process capability is insufficient and the process must be improved.

As mentioned above, solar cell wafer surface color classification can be used as a basis for improving the manufacturing process. The surface color of each piece of solar cell wafer is not only one kind, but a mixture of several different colors. For example, solar cell chips with blue-black and dark-blue colors, the ratio of the two colors on each wafer is usually not It's exactly the same. Conventional solar cell wafer surface color classification methods are classified by the human eye directly viewing the surface thereof, and the human eye classification criterion is to perform color separation according to the ratio of the main color of the surface to the secondary color. However, the general classification method only uses the main color of the surface of the solar cell wafer for classification, and the importance of the color to be used last time is neglected, which is easy to cause an error with the color separation of the human eye. In addition, it is limited that each person can't be uniform in color definition, and the human eye is easily subject to fatigue and other factors, so the human eye color separation judgment method will have considerable error.

Accordingly, it is an object of the present invention to provide a method of color separation of solar cell wafers that can be categorized by automated learning techniques to address the surface color of solar cell wafers.

According to a specific embodiment, the method for color separation of a solar cell wafer of the present invention comprises the steps of: obtaining an image of a surface of a solar cell wafer; analyzing the image to obtain a plurality of coordinate points in a color space; classifying each coordinate point into Among the plurality of color groups; analyzing the proportion of coordinate points in each color group to obtain a color feature vector; and inputting the color feature vector into a machine learning classifier to obtain a classification result of the surface color of the solar cell wafer.

In this embodiment, the machine learning classifier is subjected to color separation training on the surface color of the solar cell wafer. The color separation training includes the following steps: analyzing each training image in the training image set to obtain corresponding training in the color space. a set of coordinate points of the image; a plurality of color groups are classified from all coordinate point sets by a grouping algorithm, and each set of punctuation points is respectively classified into a plurality of color groups; respectively, each coordinate point is collected in each color The ratio of the training image coordinate points in the group to obtain a plurality of training image color feature vectors; respectively providing the desired color values to the respective training image color feature vectors to obtain a plurality of training samples; and then classifying the machine learning with the training samples The device performs color separation training.

The advantages and spirit of the present invention will be further understood from the following detailed description of the invention.

Referring to FIG. 1, FIG. 1 is a flow chart showing the steps of a method for color separation of a solar cell wafer according to an embodiment of the present invention. The method of this embodiment can utilize the automated learning system to classify the color of the surface of the solar cell wafer while solving the problems of the prior art.

As shown in FIG. 1 , the method for color separation of a solar cell wafer according to the embodiment includes the following steps: in step S10, obtaining an image of a surface of a solar cell wafer to be classified; and in step S12, analyzing the image to obtain a color space. a plurality of coordinate points; in step S14, each coordinate point is separately classified into a plurality of color groups; in step S16, the ratio of coordinate points in each color group is analyzed to obtain a color feature vector of the image; Step S18, providing the obtained color feature vector to the machine learning classifier as an input value of the machine learning classifier, thereby obtaining the result of color classification of the surface of the solar cell wafer.

In step S10, the surface image of the solar cell wafer to be classified is obtained. In practice, the image capturing device such as a Charge-Coupled Device (CCD) camera can be used to capture the surface image. In step S12, after the surface image of the solar cell wafer is removed, the effective point in the image can be first found through the image processing system, and then the color of the image at each effective point is converted to the coordinate point in the color space. In practice, the effective point can be obtained by removing the busbar image, the fine line and the background from the solar cell wafer image, and the pixel on the effective area after the removal can be used as an effective point. The color distribution of the image at each effective point can be represented by coordinate points in the color space. For example, a solar cell wafer with a dark blue color indicates that the color of each effective point is mostly dark blue, so that the coordinates corresponding to the effective points after conversion mostly fall in the area representing the dark blue in the color space. The color space used in this embodiment can be the CIE Lab color space. However, in practice, the CIE XYZ color space or the CIE Luv color space can also be used, depending on the needs of the user or the designer.

After the color of each valid point is converted to the corresponding coordinate point of the color space in step S12, then, in step S14, each coordinate point is respectively classified into a plurality of preset color groups. For example, the method of classifying each coordinate point into a color group is as follows. Each color group occupies an area in the color space, and each area can be represented by a gravity center position, and the above-mentioned respective coordinate points can be classified by the European geometric distance. The basis, that is, a punctuation will be classified into the color group represented by the position of the center of gravity closest to its Euclidean distance. Please note that the present invention is not limited to the above-described European distance classification method, and any method for classifying coordinate points into color groups can be applied to step S14.

Through the above step S14, the color of all the effective points on the image of one solar cell wafer can be divided into the respective color groups, and then, in step S16, the proportion of the coordinate points in each color group is analyzed to obtain the solar cell with the solar cell. The color feature vector associated with the color of the wafer surface. For example, if the surface of a solar cell wafer has a dark blue color and a blue-black color, the number of coordinate points included in the dark blue color group will account for the largest proportion of all coordinate points. The proportion of coordinate points included in the blue-black color group is second.

Step S16 is further illustrated by another specific embodiment. If each punctuation point analyzed by a solar cell wafer image can be divided into N color groups, a ratio of the number of coordinate points included in each color group to the total number of all coordinate points of the image can be obtained. The vector values (P ₁ , P ₂ , . . . , P _N ) are used as color eigenvectors for the image of the solar cell wafer, wherein the numbers 1~N represent N different color groups, and P is in the color group. The number of coordinate points included or the number of coordinate points included and the total number of all coordinate points.

After step S16 obtains a color feature vector for the solar cell wafer image, in step S18, the color feature vector may be input to the machine learning classifier, and the machine learning classifier judges the output value to be the surface color of the solar cell chip. Classification results. It should be noted that the machine learning classifier has been subjected to color separation training first, and the content of the color separation training will be described in detail in the following embodiments.

Please refer to FIG. 2 . FIG. 2 is a flow chart showing the steps of the color separation training of the machine learning classifier used in the method for color separation of solar cell wafers according to FIG. 1 .

As shown in FIG. 2, the color separation training of the machine learning classifier of the embodiment includes the following steps: in step S20, analyzing a plurality of training images in the training image set respectively to obtain corresponding training images in the color space respectively. a complex array of coordinate points, wherein each set of punctuation points respectively includes a plurality of training image coordinate points; then, in step S22, a plurality of color groups are classified from all training image coordinate points by a grouping algorithm, and Each training image coordinate point of each set of punctuation points is respectively classified into each color group; in step S24, the proportion of the training image coordinate points of each coordinate point group collected in each color group is analyzed to obtain related training images. Training the image color feature vector; in step S26, respectively providing the desired color value to each training image color feature vector to obtain a plurality of training samples; and, in step S28, inputting the training samples into the machine learning classifier to machine learning The classifier performs color separation training.

The training image set analyzed in step S20 may be composed of images of solar cell chips of different colors, that is, the aforementioned training images. In practice, each solar cell wafer as a training sample can be classified by its other color separation method, for example, by color separation of the human eye to establish the color category of the solar cell wafer. When a training image set is formed, a certain number of images of the solar cell wafer can be taken from different color categories. For example, a total of more than 5,000 image images can be taken from the six color categories as a training image set, and each training is performed. Only one solar cell wafer is shown in the image.

The above five thousand training images can be separately analyzed, and the color at each effective point is converted to the training image coordinate point in the color space. Please note that since the color separation is based on a solar cell wafer, the plurality of training image coordinate points converted from each training image can form different sets of coordinate points. The color space in step S20 can also be a CIE Lab color space, a CIE XYZ color space, or a CIE Luv color space. Although the step S20 and the step S12 of the above specific embodiment can apply different color spaces for conversion, the color space used in the step S20 should be consistent with the color space used in step S12, for example, both are adopted. CIE Lab color space.

In step S22, through the grouping algorithm, a plurality of color groups can be classified from all the training image coordinate points in all coordinate point sets, and each training image coordinate point of each coordinate point set can be classified into each In the color group. As described above, since the color of a solar cell wafer is not unique, the converted image pixel points of the solar cell wafer taken out from the color category of a solar cell wafer may be classified into different colors, respectively. In the group. The grouping algorithm used in step S22 may be, but is not limited to, the K-means grouping algorithm.

After the set of coordinate points of each training image is formed, in step S24, the proportion of the training image coordinate points in each color group may be separately analyzed for each coordinate point set, thereby obtaining the training image color related to each training image. Feature vector. For example, if there are N color groups, the training image coordinate points of a training image can obtain the vector value (P ₁ , P ₂ , ..) of the training image according to the proportion in each color group. , P _N ), that is, the training image color feature vector F (feature vector).

Since each training image is taken from the classified color categories, each training image has a desired classification color. If the classification color expected by each training image is represented by the desired color value B, each training image can provide a desired color value to its training image color feature vector F, and then obtain a set (F, B) as a training sample. , as described in step S26.

After each training image obtains its (F, B) training samples, in step S28, all (F, B) training samples can be input to the input of the machine learning classifier for training. Through the training of step S28, the machine learning classifier can be used as a tool for classification in the method of the specific embodiment of Fig. 1. Please note that the machine learning classifier trained by the specific embodiment, the results classified by the solar cell wafers to be classified are substantially consistent with the results classified by the color separation method for pre-classifying each training sample, Because the desired color values used during training are set according to the pre-classification method. For example, if the desired color value B of the above specific embodiment is based on the human eye color separation method, the machine learning classifier classifies the surface color of a solar cell wafer to be classified, which is roughly related to the human eye color separation method. The results of the classification are the same. It should be noted that each desired color value can be set by the user or the designer, so the color separation training of the machine learning classifier can be adjusted according to actual needs, so that the classification result of the machine learning classifier is closer to the requirements of the user or the designer. .

In practice, the machine learning classifier may include, but is not limited to, a neural network, a support vector machine (SVM), or a Gaussian Mixture Models (GMM). Taking a neural network as an example, a neural network usually includes an input layer, a hidden layer, and an output layer. The input layer is available for the input vector, and then the result is outputted in the output layer after learning or classifying the hidden layer. In the above specific embodiment, the training samples of the color separation training and the color feature vectors obtained when actually performing color classification can be input by the input layer, and the results are obtained at the output layer. The hidden layer contains a plurality of nodes, and the number of nodes can be determined according to the needs of the user or the designer.

Referring to FIG. 3, FIG. 3 is a flow chart showing the steps of a method for color separation of a solar cell wafer according to another embodiment of the present invention. As shown in FIG. 3, the method for color separation of a solar cell wafer in the embodiment of the present invention first performs color separation training on a machine learning classifier by using a plurality of training images, and then converts and calculates color characteristics of the solar cell wafer to be classified. The vector is input into a machine learning classifier, and the machine learning classifier classifies the surface color of the solar cell wafer.

In this embodiment, the method for color separation of a solar cell wafer includes the following steps: in step S30, analyzing a plurality of training images in the training image set to obtain a set of complex array coordinate points in the color space, wherein Each set of punctuation points respectively includes a plurality of coordinate points in the color space; then, in step S32, a plurality of color groups are classified from all coordinate points by a grouping algorithm, and each of the punctuation points is set The coordinate points are respectively classified into the color groups; in step S34, the proportions of the coordinate points of each coordinate point set in each color group are analyzed to further obtain a plurality of color feature vectors, wherein each color feature vector is correlated For each training image; in step S36, each color feature vector is respectively provided with a desired color value to obtain a plurality of training samples; in step S38, the machine learning classifier is trained with the obtained plurality of training samples; and Step S40, input the first color feature vector converted and calculated by the solar cell wafer to be classified to after training Machine learning classifier, thereby to obtain a classification result of the surface color of the solar cell wafer.

In this embodiment, the machine learning classifier trained in steps S30 to S38 can be used to perform color classification of solar cell wafers. Please note that the steps of the color separation training method of the machine learning classifier of steps S30 to S38 of this embodiment are substantially the same as the steps corresponding to the color separation training method of the above specific embodiment. No longer. In addition, in the step S40, the manner of obtaining the first color feature vector of the solar cell wafer to be classified is also described in the above specific embodiment, and details are not described herein again.

In summary, the method for color separation of a solar cell wafer of the present invention first trains a machine learning classifier that can be used for color separation by an automatic learning method, which converts each training image into a color space by using a color space conversion and a grouping algorithm. Train the samples and use these training samples to train the machine learning classifier. The solar cell wafers whose surface colors are to be classified are also subjected to color space conversion, and the converted coordinate points are classified into the color groups classified by the above-described grouping algorithm, thereby obtaining color feature vectors. Then, the color feature vector is input to the trained machine learning classifier to obtain the classification result. Compared with the prior art, the method for color separation of the solar cell wafer of the present invention can not only separate the solar cell wafer according to the main color and the secondary color of the surface, but also perform color separation according to the color that may appear on the surface. At the same time, the use of machine learning classifiers for color separation can avoid personal subjective color recognition and errors caused by human eye fatigue. Thereby, the surface color of the solar cell wafer can be classified more effectively.

The features and spirit of the present invention will be more apparent from the detailed description of the preferred embodiments. On the contrary, the intention is to cover various modifications and equivalents within the scope of the invention as claimed. Therefore, the scope of the patented scope of the invention should be construed as broadly construed in the

S10~S18, S20~28, S30~S40‧‧‧ Process steps

1 is a flow chart showing the steps of a method for color separation of a solar cell wafer according to an embodiment of the present invention.

FIG. 2 is a flow chart showing the steps of the color separation training of the machine learning classifier applied in the method for color separation of solar cell wafers according to FIG. 1.

3 is a flow chart showing the steps of a method for color separation of a solar cell wafer according to another embodiment of the present invention.

S30~S40. . . Process step

Claims (11)

A solar cell wafer color separation method for classifying a surface color of a solar cell wafer, the method comprising the steps of: obtaining an image of a surface of the solar cell wafer; analyzing the image to obtain a plurality of coordinates in a color space Points; classifying the coordinate points into a plurality of color groups; analyzing a ratio of the number of coordinate points classified into each of the color groups to the total number of coordinate points of the coordinate points to obtain a color feature vector; And providing the color feature vector as an input value of a machine learning classifier to obtain a classification result of the surface color of the solar cell wafer, wherein the machine learning classifier is subjected to a color separation training. The method of claim 1, wherein the color space is a CIE Lab color space. The method of claim 1, wherein the analyzing the image to obtain the coordinate points in the color space comprises: analyzing the image to obtain a plurality of valid points; and validating the image on the image The colors on the dots are converted to the coordinate points in the color space, respectively. The method of claim 1, wherein the color separation training of the machine learning classifier comprises the steps of separately analyzing a plurality of training images in a training image set to obtain a plurality of colors in the color space. a set of coordinate points, each of which includes a plurality of training image coordinate points in the color space; and a grouping algorithm classifies the training image coordinate points from the training points a color group, and classifying the training image coordinate points into the color groups; respectively analyzing the proportions of the coordinate points of the training images collected in the color groups to obtain a plurality of Training the image color feature vector; respectively providing a plurality of desired color values to the training image color feature vectors to obtain a plurality of training samples; and training the machine learning classifier with the training samples. The method of claim 4, wherein the grouping algorithm is a K-means grouping algorithm. The method of claim 1, wherein the machine learning classifier is a type of neural network having a hidden layer and the hidden layer has a plurality of nodes. The method of claim 1, wherein the machine learning classifier is one of a Support Vector Machine (SVM) and a Gaussian Mixture Models (GMM). A solar cell wafer color separation method for classifying a surface color of a solar cell wafer, the method comprising the steps of separately analyzing a plurality of training images in a training image set to obtain a complex array coordinate point in a color space a collection, the set of coordinate points respectively comprise a plurality of coordinate points in the color space; a plurality of color groups are classified from the coordinate points by a grouping algorithm, and the coordinate points are respectively classified into a plurality of color points In the color group; respectively analyzing the ratio of the number of coordinate points classified into each color group to the total number of coordinate points of the coordinate points in the set of coordinate points For example, obtaining a plurality of color feature vectors; respectively providing a plurality of desired color values to the color feature vectors to obtain a plurality of training samples; training a machine learning classifier with the training samples; and providing the solar cell A first color feature vector of the wafer is used as an input value of the machine learning classifier to obtain a classification result of the surface color of the solar cell wafer. The method of claim 8, wherein the clustering algorithm is a K-means grouping algorithm. The method of claim 8, wherein the color space is a CIE Lab color space. The method of claim 8, wherein separately analyzing the training images in the training image set to obtain the set of coordinate points in the color space comprises: separately analyzing the images to obtain a plurality of sets of valid points, each of the set of valid points respectively comprising a plurality of valid points; and the colors of the images on the corresponding effective points are respectively converted to the coordinate points in the color space.