A method and system for detecting color difference of solar cell
The method uses a classification network with convolutional autoencoders and clustering to accurately and efficiently detect color differences in solar cells, addressing inefficiencies and inaccuracies of existing methods.
Patent Information
- Application Number
- CN202111468227.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-12-03
- Publication Date
- 2024-12-27
- Estimated Expiration
- 2041-12-03
AI Technical Summary
Current methods for detecting color differences in solar cells are inefficient, inaccurate, and can damage the cells during manual inspection, while automated solutions lack a unified standard and are not precise.
A method involving a classification network built using convolutional autoencoders and clustering algorithms to process solar cell images, determining intra-chip chromatic aberration and constructing a classification network for accurate sorting based on multi-dimensional feature vectors and center points.
The method provides accurate and efficient color difference detection without damaging the cells, outperforming manual and existing automated methods by ensuring faster and more precise intra-chip and inter-chip classification.
Smart Images
Figure CN114387466B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of solar cell panels, and in particular to a method and system for detecting color difference of solar cell panels. Background Art
[0002] In the current context of energy shortage and the promotion of low-carbon economy, all countries have accelerated the pace of developing solar energy. In recent years, the scale of my country's photovoltaic industry has gradually expanded and the technology has also improved, in order to reduce dependence on non-renewable resources such as coal and oil and reduce carbon emissions. In the future, the domestic photovoltaic industry is expected to develop further.
[0003] Solar cells are photoelectric semiconductor sheets that can convert light energy into electrical energy by absorbing sunlight and using the photovoltaic effect. Under certain conditions of sunlight, solar cells can output voltage instantly and generate current in the presence of a circuit. However, due to the influence of the manufacturing process, solar cells will have color differences, which will not only cause uneven color distribution inside a single cell, but also color differences between cells. Using cells with too different colors together will not only affect the overall appearance, but also the power generation performance of photovoltaic modules. Therefore, before making photovoltaic modules, it is generally necessary to sort the cells according to their color.
[0004] At present, the color sorting of battery cells mainly relies on manual inspection, but this method is not only inefficient and inaccurate, but also easily causes damage to the battery cells during the sorting process; some are based on grayscale histograms or through BP algorithms for color classification, but because of the need to select standard samples as templates, the experiment cannot determine a unified standard and the results are not accurate. Summary of the invention
[0005] In order to overcome the shortcomings of the above-mentioned prior art, the present invention provides a method for detecting color difference of solar cell sheets, comprising the following steps:
[0006] S1. Building a classification network, and step S1 specifically includes:
[0007] S1a, respectively collecting images of multiple solar cells and preprocessing each image;
[0008] S1b, performing block processing on the preprocessed image of each solar cell, dividing it into a plurality of block images of preset sizes, inputting the plurality of block images of each solar cell into a convolutional autoencoder for training, and extracting a multi-dimensional feature vector in a hidden layer of the convolutional autoencoder;
[0009] S1c, for any solar cell, calculate the similarity of the multi-dimensional feature vectors between any two block images thereof, and determine whether the solar cell has intra-chip chromatic aberration according to the calculated similarity, and complete the calculation of all solar cells to eliminate the solar cells with intra-chip chromatic aberration;
[0010] S1d, taking image data of multiple solar cells without intra-cell chromatic aberration, taking an average value of multi-dimensional feature vectors of multiple block images of each solar cell, and obtaining a multi-dimensional average feature vector of each solar cell;
[0011] S1e, using a clustering algorithm to obtain a multi-dimensional average feature vector of a plurality of solar cells, and determining a plurality of center points according to the clustering results;
[0012] S1f, constructing a classification network using the multiple center points determined in step S1e as classification criteria;
[0013] S2. After the construction of the classification network is completed, for the solar cell to be classified, its image is collected and pre-processed according to the same scheme as in step S1a, and then the block processing is performed according to the same scheme as in step S1b, and then a plurality of block images are respectively input into the classification network, and the classification network is used to determine the category to which each block image belongs. If the categories to which the plurality of block images belong are inconsistent, it is determined that the detection result of the solar cell has intra-chip color difference. If the categories to which the plurality of block images belong are consistent, the solar cell is classified into this category.
[0014] In some embodiments, in step S1a, the step of preprocessing the image specifically includes: removing the background in the image, extracting the region of interest of the image to extract the region where the solar cell is located, and compressing the extracted solar cell region image to a preset size.
[0015] In some embodiments, in step S1a, in the step of preprocessing the image, the extracted solar cell region image is compressed to a size of 1280*1280 pixels;
[0016] In step S1b, when performing block processing, the image is divided into 10*10 block images with a size of 128*128 pixels.
[0017] In some embodiments, in step S1c, the following method is used to determine whether there is intra-chip color difference in the solar cell:
[0018] For a solar cell, the multiple multidimensional feature vectors corresponding to its multiple block images are recorded as a set X = {x1, x2, ..., x N}, where N represents the number of block images;
[0019] Calculate any two multidimensional feature vectors x in the set according to the following formula i With x j The Euclidean distance d is:
[0020] d=||x i -x j ||2
[0021] The Euclidean distance d is used as the similarity standard, and a threshold m is preset. If d>m exists in all the calculated Euclidean distances d, it is determined that the solar cell has intra-chip color difference, otherwise it is determined that the solar cell does not have intra-chip color difference.
[0022] In some embodiments, in step S1d, image data of at least 1000 solar cells without intra-cell chromatic aberration are obtained.
[0023] In some embodiments, in step S1e, a multi-dimensional average feature vector of a plurality of solar cells is obtained by processing with a PROCLUS clustering algorithm, and a plurality of center points are determined according to the clustering result.
[0024] In some embodiments, step S1f specifically includes the following steps:
[0025] The multiple center points determined in step S1e are used as classification criteria and labeled to obtain a training set. Next, the convolutional autoencoder trained in step S1b is adjusted by removing the decoder part and adding a classification layer after the hidden layer. The classification layer is established based on the obtained training set, thereby completing the construction of the classification network.
[0026] On the other hand, the present invention provides a solar cell color difference detection system, including an image acquisition module, a central control module and an operating element; the image acquisition module is used to take pictures to collect images of solar cells, and the central control module is used to implement color difference detection of solar cells according to the above-mentioned solar cell color difference detection method, and after completing the color difference detection, control the operating element to perform subsequent operations according to the detection results.
[0027] Compared with the prior art, the present invention has the following beneficial effects:
[0028] The solar cell color difference detection method and system provided by the present invention are more accurate and efficient than manual detection methods, and will not pollute or damage the solar cells; compared with existing automated detection solutions, the present invention only needs to use one classification neural network to quickly and accurately solve the intra-chip color difference and inter-chip classification problems, which is faster and more convenient, and the results are more accurate. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] Figure 1The present invention provides a flow chart of the method for detecting color difference of solar cells. DETAILED DESCRIPTION
[0030] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the following further describes how the present invention is implemented in conjunction with the accompanying drawings and specific implementation methods.
[0031] Reference Figure 1 As shown, the present invention provides a method for detecting color difference of solar cell sheets, comprising the following steps:
[0032] A method for detecting color difference of a solar cell comprises the following steps:
[0033] S1. Building a classification network, and step S1 specifically includes:
[0034] S1a, respectively collecting images of multiple solar cells and preprocessing each image;
[0035] S1b, performing block processing on the preprocessed image of each solar cell, dividing it into a plurality of block images of preset sizes, inputting the plurality of block images of each solar cell into a convolutional autoencoder for training, and extracting a multi-dimensional feature vector in a hidden layer of the convolutional autoencoder;
[0036] S1c, for any solar cell, calculate the similarity of the multi-dimensional feature vectors between any two block images thereof, and determine whether the solar cell has intra-chip chromatic aberration according to the calculated similarity, and complete the calculation of all solar cells to eliminate the solar cells with intra-chip chromatic aberration;
[0037] S1d, taking image data of multiple solar cells without intra-cell chromatic aberration, taking an average value of multi-dimensional feature vectors of multiple block images of each solar cell, and obtaining a multi-dimensional average feature vector of each solar cell;
[0038] S1e, using a clustering algorithm to obtain a multi-dimensional average feature vector of a plurality of solar cells, and determining a plurality of center points according to the clustering results;
[0039] S1f, constructing a classification network using the multiple center points determined in step S1e as classification criteria;
[0040] S2. After the construction of the classification network is completed, for the solar cell to be classified, its image is collected and pre-processed according to the same scheme as in step S1a, and then the block processing is performed according to the same scheme as in step S1b, and then a plurality of block images are respectively input into the classification network, and the classification network is used to determine the category to which each block image belongs. If the categories to which the plurality of block images belong are inconsistent, it is determined that the detection result of the solar cell has intra-chip color difference. If the categories to which the plurality of block images belong are consistent, the solar cell is classified into this category.
[0041] Preferably, in step S1a, the step of preprocessing the image specifically includes: removing the background in the image, extracting the region of interest from the image to extract the region where the solar cell is located, and compressing the extracted solar cell region image to a preset size.
[0042] In a specific embodiment, in step S1a, in the step of preprocessing the image, the extracted solar cell area image is uniformly compressed to a size of 1280*1280 pixels. Through such processing, the amount of calculation can be reduced without losing too much image information;
[0043] Correspondingly, in step S1b, when performing block processing, the image is divided into 10*10 block images of 128*128 pixels, and then these block images are input into the convolutional autoencoder for training. Since the convolutional autoencoder belongs to unsupervised learning, the original input image and the reconstructed image are used for loss training respectively. By reducing the loss, the features in the hidden layer of the convolutional autoencoder, that is, the multi-dimensional feature vector, can be extracted.
[0044] Preferably, in step S1c, whether there is color difference within the solar cell can be determined according to the following method:
[0045] For a solar cell, the multiple multidimensional feature vectors corresponding to its multiple block images are recorded as a set X = {x1, x2, ..., x N}, where N represents the number of block images. For a 10*10 block image, N=100;
[0046] Calculate any two multidimensional feature vectors x in the set according to the following formula i With x j The Euclidean distance d is:
[0047] d=||x i -x j ||2
[0048] It is understandable that according to the definition of Euclidean distance, we have:
[0049]
[0050] Where n represents the multidimensional feature vector x i With x j The number of dimensions, Respectively represent x i The components in the first, second, ... nth dimensions, Respectively represent x j Components in the first, second ... nth dimensions; using the Euclidean distance d as the similarity standard, and presetting a threshold m, if d>m exists in all the calculated Euclidean distances d, it is determined that the solar cell has intra-chip color difference, otherwise it is determined that the solar cell does not have intra-chip color difference.
[0051] In addition, it is understood that in order to ensure accuracy, a large number of solar cell samples should be taken for processing when constructing the classification network. For example, in step S1, about 1500 solar cells should be taken initially, and in the subsequent step S1d, image data of at least 1000 solar cells without intra-cell chromatic aberration should be taken.
[0052] Preferably, in step S1e, the multi-dimensional average feature vectors of the plurality of solar cells are processed by the PROCLUS clustering algorithm, and a plurality of center points are determined according to the clustering results. When the PROCLUS clustering algorithm is used for processing, a superset of center points is selected, the feature dimension of each center point is determined, the clustering results are analyzed, and the center points are continuously replaced iteratively to obtain the optimal center point set, the center point dimension is optimized, and an improved clustering result is obtained.
[0053] Furthermore, step S1f specifically includes the following steps:
[0054] The multiple center points determined in step S1e are used as classification criteria and labeled to obtain a training set. Next, the convolutional autoencoder trained in step S1b is adjusted by removing the decoder part and adding a classification layer after the hidden layer. The classification layer is established based on the obtained training set, thereby completing the construction of the classification network.
[0055] Finally, in step S2, after the construction of the classification network is completed, for the solar cell to be classified, its image is collected and preprocessed according to the same scheme as in step S1a, and then the block processing is performed according to the same scheme as in step S1b, and then the multiple block images are respectively input into the classification network, and the classification network is used to determine the category to which each block image belongs. If the categories to which the multiple block images belong are inconsistent, the detection result of the solar cell is determined to have intra-chip color difference. If the categories to which the multiple block images belong are consistent, the solar cell is classified into this category.
[0056] On the other hand, the present invention provides a solar cell color difference detection system, including an image acquisition module, a central control module and an operating element; the image acquisition module is used to take pictures to collect images of solar cells, and the central control module is used to implement color difference detection of solar cells according to the above-mentioned solar cell color difference detection method, and after completing the color difference detection, control the operating element to perform subsequent operations according to the detection results.
[0057] In summary, the solar cell color difference detection method and system provided by the present invention are more accurate and efficient than manual detection methods, and will not pollute or damage the solar cells; compared with existing automated detection solutions, the present invention only needs to use a classification neural network to quickly and accurately solve the intra-chip color difference and inter-chip classification problems, which is faster and more convenient, and the results are more accurate.
[0058] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution of the present invention, which should be included in the scope of the claims of the present invention.
Claims
1. A method for detecting color difference of a solar cell, characterized in that: The steps include: S1. Building a classification network, and step S1 specifically includes: S1a, respectively collecting images of multiple solar cells and preprocessing each image; S1b, performing block processing on the preprocessed image of each solar cell, dividing it into a plurality of block images of preset sizes, inputting the plurality of block images of each solar cell into a convolutional autoencoder for training, and extracting a multi-dimensional feature vector in a hidden layer of the convolutional autoencoder; S1c, for any solar cell, calculate the similarity of the multi-dimensional feature vectors between any two block images thereof, and determine whether the solar cell has intra-chip chromatic aberration according to the calculated similarity, and complete the calculation of all solar cells to eliminate the solar cells with intra-chip chromatic aberration; S1d, taking image data of multiple solar cells without intra-cell chromatic aberration, taking an average value of multi-dimensional feature vectors of multiple block images of each solar cell, and obtaining a multi-dimensional average feature vector of each solar cell; S1e, using a clustering algorithm to obtain a multi-dimensional average feature vector of a plurality of solar cells, and determining a plurality of center points according to the clustering results; S1f, constructing a classification network using the multiple center points determined in step S1e as classification criteria; S2. After the construction of the classification network is completed, for the solar cell to be classified, its image is collected and pre-processed according to the same scheme as in step S1a, and then the block processing is performed according to the same scheme as in step S1b, and then a plurality of block images are respectively input into the classification network, and the classification network is used to determine the category to which each block image belongs. If the categories to which the plurality of block images belong are inconsistent, it is determined that the detection result of the solar cell has intra-chip color difference. If the categories to which the plurality of block images belong are consistent, the solar cell is classified into this category.
2. The method for detecting color difference of a solar cell according to claim 1, characterized in that: In step S1a, the step of preprocessing the image specifically includes: removing the background in the image, extracting the region of interest of the image to extract the region where the solar cell is located, and compressing the extracted solar cell region image to a preset size.
3. The method for detecting color difference of a solar cell according to claim 2, characterized in that: In step S1a, in the step of image preprocessing, the extracted solar cell area image is compressed to a size of 1280*1280 pixels; In step S1b, when performing block processing, the image is divided into 10*10 block images with a size of 128*128 pixels.
4. The method for detecting color difference of a solar cell according to claim 1, characterized in that: In step S1c, the following method is used to determine whether there is color difference within the solar cell: For a solar cell, the multiple multidimensional feature vectors corresponding to its multiple block images are recorded as a set X = {x1, x2, ..., x N }, where N represents the number of block images; Calculate any two multidimensional feature vectors x in the set according to the following formula i With x j The Euclidean distance d is: d=||x i -x j ||2 The Euclidean distance d is used as the similarity standard, and a threshold m is preset. If d>m exists in all the calculated Euclidean distances d, it is determined that the solar cell has intra-chip color difference, otherwise it is determined that the solar cell does not have intra-chip color difference.
5. The method for detecting color difference of a solar cell according to claim 1, characterized in that: In step S1d, image data of at least 1000 solar cells without intra-cell chromatic aberration are obtained.
6. The method for detecting color difference of a solar cell according to claim 1, characterized in that: In step S1e, the multi-dimensional average feature vectors of the plurality of solar cells are processed using the PROCLUS clustering algorithm, and a plurality of center points are determined according to the clustering results.
7. The method for detecting color difference of a solar cell according to claim 1, characterized in that: Step S1f specifically includes the following steps: The multiple center points determined in step S1e are used as classification criteria and labeled to obtain a training set. Next, the convolutional autoencoder trained in step S1b is adjusted by removing the decoder part and adding a classification layer after the hidden layer. The classification layer is established based on the obtained training set, thereby completing the construction of the classification network.
8. A solar cell color difference detection system, characterized in that: It comprises an image acquisition module, a central control module and an operating element; the image acquisition module is used to take photos to collect images of solar cells, the central control module is used to implement color difference detection of solar cells according to the solar cell color difference detection method according to any one of claims 1 to 7, and after completing the color difference detection, control the operating element to perform subsequent operations according to the detection results.