基于視覺的硅太陽能電池檢測方法的研究
發(fā)布時(shí)間：2019-06-10 20:55
【摘要】：隨著能源危機(jī)的出現(xiàn)，太陽能在新能源開發(fā)和利用中扮演著重要角色，光伏發(fā)電作為近期發(fā)展起來具有很大潛力的新能源技術(shù)，其核心部分為太陽能電池。太陽能電池生產(chǎn)過程中產(chǎn)生的瑕疵，會(huì)導(dǎo)致光電轉(zhuǎn)換效率的下降和產(chǎn)品生產(chǎn)成本的提高，因此在生產(chǎn)過程中需對(duì)其進(jìn)行檢測。傳統(tǒng)的太陽能電池檢測依賴于人工判別，存在著人工檢測易出現(xiàn)偏差誤判且成本偏高等弊端。本文將基于視覺的檢測算法應(yīng)用在硅太陽能電池生產(chǎn)的各階段，依次為對(duì)太陽能硅片瑕疵檢測算法的研究、對(duì)硅太陽能電池片瑕疵檢測算法的研究以及對(duì)硅太陽能電池組件瑕疵檢測算法和分類算法的研究。 針對(duì)太陽能硅片瑕疵檢測，本文采用近紅外LED作為光源設(shè)備和近紅外CCD相機(jī)作為圖像采集設(shè)備，得到太陽能硅片表面和內(nèi)部的隱裂圖像。在隱裂瑕疵檢測算法設(shè)計(jì)上，根據(jù)隱裂瑕疵在硅片上呈現(xiàn)出低灰度級(jí)和高梯度級(jí)的特點(diǎn)，傳統(tǒng)邊緣檢測法或二值化法對(duì)于低對(duì)比度圖像并不適用。因此本文采用各向異性擴(kuò)散算法進(jìn)行隱裂檢測，算法根據(jù)圖像中不同的梯度分布，對(duì)高梯度值—瑕疵區(qū)域進(jìn)行銳化處理，對(duì)低梯度值—無瑕疵區(qū)域進(jìn)行平滑處理，即實(shí)現(xiàn)在銳化瑕疵的同時(shí)抑制噪聲的目的。但在算法中，擴(kuò)散函數(shù)、銳化參數(shù)等數(shù)值的改變直接影響到瑕疵檢測的結(jié)果，目前也缺乏統(tǒng)一的擴(kuò)散函數(shù)表達(dá)方式。因此本文提出將改進(jìn)的異性擴(kuò)散算法作為圖像銳化算子進(jìn)行圖像邊緣提取后，根據(jù)確定的種子像素利用區(qū)域生長算法將隱裂瑕疵從背景中分割出來，經(jīng)試驗(yàn)算法運(yùn)算精度高，可滿足在線太陽能硅片瑕疵檢測的要求。 針對(duì)太陽能電池片瑕疵檢測，本文在分析了各種圖像分割算法后，提出了一種改進(jìn)的Otsu算法作為瑕疵特征提取方法。在線瑕疵檢測分為兩種情況：第一種為判斷電池片是否出現(xiàn)瑕疵，如果出現(xiàn)瑕疵則立即丟棄，不影響生產(chǎn)；第二種為檢測出瑕疵電池片后需要對(duì)瑕疵進(jìn)行定位回溯，以使用戶發(fā)現(xiàn)造成瑕疵的原因，進(jìn)行產(chǎn)品質(zhì)量追蹤。針對(duì)第一種情況，本文在研究了典型瑕疵特征后，采用空間域的方法進(jìn)行瑕疵檢測。針對(duì)第二種情況，本文采用“空間域-頻率域-空間域”的圖像重建方法得到了無瑕疵圖像，再將原始圖像和重建圖像進(jìn)行差分運(yùn)算，進(jìn)而實(shí)現(xiàn)圖像瑕疵定位。根據(jù)單晶硅和多晶硅表面紋理呈現(xiàn)出的差異性，對(duì)具有規(guī)律紋理特征的單晶硅太陽能電池采用多解析度、多分辨率的小波變換進(jìn)行圖像重建；對(duì)具有隨機(jī)紋理特征的多晶硅太陽能電池采用代表全局頻域信息的傅里葉圖像重建算法。 針對(duì)太陽能電池組件瑕疵檢測，，由于太陽能電池組件是由一系列太陽能電池片經(jīng)過串聯(lián)或并聯(lián)形式得到的，但瑕疵并不會(huì)出現(xiàn)在太陽能電池組件中每個(gè)子部分。因此如果以單一電池片作為模板應(yīng)用前面介紹的圖像重建算法進(jìn)行逐一搜索，數(shù)據(jù)冗余大、并且效率低。針對(duì)這一問題，本文采用通過訓(xùn)練得到的獨(dú)立成份分析（ICA）分離矩陣重構(gòu)待檢圖像，以增強(qiáng)瑕疵信息并濾除組件圖像的規(guī)律性紋理。ICA方法之一的FastICA具有了收斂速度快等優(yōu)點(diǎn),但也存在當(dāng)初始點(diǎn)遠(yuǎn)離極值點(diǎn)而無法收斂等缺點(diǎn)，弱化了ICA算法的瑕疵檢測能力。本文提出將粒子群優(yōu)化算法（PSO）引入到FastICA算法中，由PSO算法得到的全局最佳位置作為最佳分類矩陣，并求出獨(dú)立分量IC，最后重建檢測圖像以判斷其是否存在瑕疵。針對(duì)太陽能電池瑕疵分類算法的研究，本文提出了基于AdaBoost分類器的支持向量機(jī)（SVM）算法進(jìn)行樣本訓(xùn)練，之后對(duì)輸入的待檢測圖像應(yīng)用SVM分類器進(jìn)行瑕疵分類，輸出分類結(jié)果。針對(duì)太陽能電池組件瑕疵檢測和分類都需要事先進(jìn)行樣本訓(xùn)練的弊端，本文提出了一種無需參考樣本的自適應(yīng)閾值的瑕疵檢測和分類方法，速度得到明顯優(yōu)化，并且處理效果令人滿意，對(duì)于目前依賴人工檢測的太陽能電池生產(chǎn)行業(yè)有著非常大的應(yīng)用前景。
[Abstract]:With the emergence of the energy crisis, the solar energy plays an important role in the development and utilization of the new energy, and the photovoltaic power generation, as a new energy technology with great potential for recent development, is the core part of the solar cell. The defects generated in the production process of the solar cell can lead to the reduction of the photoelectric conversion efficiency and the improvement of the production cost of the product, so that the production process needs to be detected. The traditional detection of the solar cell depends on the artificial discrimination, and the defects that the manual detection is easy to be misjudged and the cost is high is high. In this paper, a visual detection algorithm is applied to each stage of the production of silicon solar cell, in order to study the defect detection algorithm of the solar silicon wafer, the research on the defect detection algorithm of the silicon solar cell, and the research on the defect detection algorithm and the classification algorithm of the silicon solar cell. In the light of the defect detection of the solar silicon wafer, the near-infrared LED is used as the light source device and the near-infrared CCD camera as the image acquisition equipment to obtain the hidden crack of the surface and the interior of the solar silicon chip. The traditional edge detection method or the binary method does not apply to the low contrast image according to the characteristics of low gray scale and high gradient level on the silicon wafer according to the hidden flaw detection algorithm. In this paper, the anisotropic diffusion algorithm is used to detect the crack, and the algorithm is used to sharpen the defect area of the high gradient value according to the different gradient distribution in the image, and the non-defective area of the low gradient value is smoothed, that is, the purpose of suppressing the noise while sharpening the defect is realized. However, in the algorithm, the change of the diffusion function and the sharpening parameter directly affects the result of the flaw detection, and there is also a lack of uniform diffusion function expression party. In this paper, the improved anisotropic diffusion algorithm is used as the image sharpening operator to carry out image edge extraction, and the hidden flaws are separated from the background by the regional growth algorithm according to the determined seed pixels, and the accuracy of the algorithm is calculated by the test algorithm. high, can meet that defect detection of the on-line solar silicon chip, In order to detect the defect of solar cell, an improved Otsu algorithm is put forward as a feature of the flaw detection after various image segmentation algorithms are analyzed. The method comprises the following steps of: first, judging whether the battery piece is defective or not, and immediately discarding the defect if the defect is present, and does not affect the production; and the second is to carry out positioning and backtracking on the defect after the defective battery piece is detected, so that the user can find the defect For the reason, make the product quality In the first case, after the typical flaw feature is studied, the method of space domain is used to make the flaw. In the second case, the image reconstruction method of the "space domain-frequency domain-space domain" is used to obtain the defect-free image, then the original image and the reconstructed image are subjected to differential operation, and the image-vanishing point is realized. The multi-resolution and multi-resolution wavelet transform is applied to the single-crystal silicon solar cell with regular texture characteristics according to the difference between the single-crystal silicon and the surface texture of the polysilicon. image reconstruction; a polycrystalline silicon solar cell with a random texture feature using a Fourier image representative of global frequency domain information The invention relates to a building algorithm, aiming at the defect detection of a solar cell module, because the solar cell module is obtained by series or parallel connection of a series of solar cell pieces, the defects do not appear in the solar cell module, therefore, the data redundancy is large if the image reconstruction algorithm described above is applied as a template by a single battery slice, And the efficiency is low. Aiming at the problem, an independent component analysis (ICA) separation matrix obtained by training is adopted to reconstruct the image to be detected, so that the defect information is enhanced and the image of the component is filtered out. The FastICA of one of the ICA methods has the advantages of fast convergence speed and the like, but also has the disadvantage that the initial point is far from the extreme point and cannot be converged, and the defect of the ICA algorithm is weakened. In this paper, a particle swarm optimization algorithm (PSO) is introduced into the FastICA algorithm, the global optimal position obtained by the PSO algorithm is used as the best classification matrix, and the independent component IC is obtained. In this paper, a support vector machine (SVM) algorithm based on AdaBoost classifier is proposed for sample training. In order to detect and classify the defect of the solar cell module, the defect of the sample training is required in advance. In this paper, the defect detection and classification method of the self-adaptive threshold without reference sample is proposed, the speed is obviously optimized, and the processing effect The fruit is satisfactory, and is very large for the solar cell production industry which is currently relying on manual detection
【學(xué)位授予單位】：河北農(nóng)業(yè)大學(xué)
【學(xué)位級(jí)別】：博士
【學(xué)位授予年份】：2014
【分類號(hào)】：TP391.41;TM914.4
本文編號(hào)：2496724
[Abstract]:With the emergence of the energy crisis, the solar energy plays an important role in the development and utilization of the new energy, and the photovoltaic power generation, as a new energy technology with great potential for recent development, is the core part of the solar cell. The defects generated in the production process of the solar cell can lead to the reduction of the photoelectric conversion efficiency and the improvement of the production cost of the product, so that the production process needs to be detected. The traditional detection of the solar cell depends on the artificial discrimination, and the defects that the manual detection is easy to be misjudged and the cost is high is high. In this paper, a visual detection algorithm is applied to each stage of the production of silicon solar cell, in order to study the defect detection algorithm of the solar silicon wafer, the research on the defect detection algorithm of the silicon solar cell, and the research on the defect detection algorithm and the classification algorithm of the silicon solar cell. In the light of the defect detection of the solar silicon wafer, the near-infrared LED is used as the light source device and the near-infrared CCD camera as the image acquisition equipment to obtain the hidden crack of the surface and the interior of the solar silicon chip. The traditional edge detection method or the binary method does not apply to the low contrast image according to the characteristics of low gray scale and high gradient level on the silicon wafer according to the hidden flaw detection algorithm. In this paper, the anisotropic diffusion algorithm is used to detect the crack, and the algorithm is used to sharpen the defect area of the high gradient value according to the different gradient distribution in the image, and the non-defective area of the low gradient value is smoothed, that is, the purpose of suppressing the noise while sharpening the defect is realized. However, in the algorithm, the change of the diffusion function and the sharpening parameter directly affects the result of the flaw detection, and there is also a lack of uniform diffusion function expression party. In this paper, the improved anisotropic diffusion algorithm is used as the image sharpening operator to carry out image edge extraction, and the hidden flaws are separated from the background by the regional growth algorithm according to the determined seed pixels, and the accuracy of the algorithm is calculated by the test algorithm. high, can meet that defect detection of the on-line solar silicon chip, In order to detect the defect of solar cell, an improved Otsu algorithm is put forward as a feature of the flaw detection after various image segmentation algorithms are analyzed. The method comprises the following steps of: first, judging whether the battery piece is defective or not, and immediately discarding the defect if the defect is present, and does not affect the production; and the second is to carry out positioning and backtracking on the defect after the defective battery piece is detected, so that the user can find the defect For the reason, make the product quality In the first case, after the typical flaw feature is studied, the method of space domain is used to make the flaw. In the second case, the image reconstruction method of the "space domain-frequency domain-space domain" is used to obtain the defect-free image, then the original image and the reconstructed image are subjected to differential operation, and the image-vanishing point is realized. The multi-resolution and multi-resolution wavelet transform is applied to the single-crystal silicon solar cell with regular texture characteristics according to the difference between the single-crystal silicon and the surface texture of the polysilicon. image reconstruction; a polycrystalline silicon solar cell with a random texture feature using a Fourier image representative of global frequency domain information The invention relates to a building algorithm, aiming at the defect detection of a solar cell module, because the solar cell module is obtained by series or parallel connection of a series of solar cell pieces, the defects do not appear in the solar cell module, therefore, the data redundancy is large if the image reconstruction algorithm described above is applied as a template by a single battery slice, And the efficiency is low. Aiming at the problem, an independent component analysis (ICA) separation matrix obtained by training is adopted to reconstruct the image to be detected, so that the defect information is enhanced and the image of the component is filtered out. The FastICA of one of the ICA methods has the advantages of fast convergence speed and the like, but also has the disadvantage that the initial point is far from the extreme point and cannot be converged, and the defect of the ICA algorithm is weakened. In this paper, a particle swarm optimization algorithm (PSO) is introduced into the FastICA algorithm, the global optimal position obtained by the PSO algorithm is used as the best classification matrix, and the independent component IC is obtained. In this paper, a support vector machine (SVM) algorithm based on AdaBoost classifier is proposed for sample training. In order to detect and classify the defect of the solar cell module, the defect of the sample training is required in advance. In this paper, the defect detection and classification method of the self-adaptive threshold without reference sample is proposed, the speed is obviously optimized, and the processing effect The fruit is satisfactory, and is very large for the solar cell production industry which is currently relying on manual detection
【學(xué)位授予單位】：河北農(nóng)業(yè)大學(xué)
【學(xué)位級(jí)別】：博士
【學(xué)位授予年份】：2014
【分類號(hào)】：TP391.41;TM914.4
【參考文獻(xiàn)】
相關(guān)期刊論文 前1條
1 張舞杰;李迪;葉峰;;硅太陽能電池紋理缺陷檢測[J];計(jì)算機(jī)應(yīng)用;2010年10期
本文編號(hào)：2496724
本文鏈接：http://sikaile.net/kejilunwen/dianlilw/2496724.html
最近更新
教材專著
