隨著工業檢測的應用需求，檢測技術愈來愈多樣化。以太陽能電池板為檢測對象的研究還不是很多，但一樣有著準確與快速的需求。在本論文的研究中，我們提出自動視覺檢測方法來量測太陽能電池板的尺寸並檢測線路上的瑕疵。 我們的系統包含幾項檢測: 尺寸量測、外形破損檢測、及網印線路瑕疵檢測。在尺寸量測中，我們利用次像素邊偵測、霍氏轉換、和最小平方誤差法來偵測電池板邊緣，利用偵測到的邊線來判定破損瑕疵。在線路瑕疵檢測中，我們將線路分為兩類：粗線 (busbar) 與細線 (finger)。我們使用追蹤檢驗的方法來檢測粗線上的瑕疵；用樣板比對的方法來檢測細線上的瑕疵。我們的檢測機制能夠檢測粗線上的四類瑕疵：凹、凸、破洞、和斷線瑕疵；能夠檢測細線上的三類瑕疵：凹、凸、和斷線瑕疵。 我們以2048×2048影像來驗證我們的方法。並以三維CCD測量出真正電池板的尺寸做比較，以判定系統的正確性。我們以三張電池板在相同情況下連續拍十張影像做尺寸量測，以測驗系統的重覆性。我們的方法幾乎可以百分之百的檢驗出所有瑕疵。在Intel Pentium 4 2.4GHz CPU的執行環境下平均一張電池板影像的執行時間約耗費1.86秒。實驗結果顯示，我們所提的方法幾乎已達到實用的階段。 Recently, the inspection of manufactured products is an important industrial activity. The inspection of solar cells is not widespread and the advanced techniques are still pursued. In this paper, the studies automatic optical inspection methods for detecting defects and scale measurement on the solar cells are proposed. The proposed system includes several inspection items: scale measurement, V-shape defect detection, and line defect detection of printed solar cells. In scale measurement, we use subpixel edge detection, Hough transform, and LSM to detect and accurately allocate edges on solar cells. The detected edges are also used for detecting the V-shape defect on the solar cell boundaries. We proposed a subpixel edge detection method based on the Catmull-Rom spline to accurately locate the edges. In line defect detection, there are two types of lines: busbars and fingers. We use a tracking method to inspect the defects on busbars and use a template matching method to inspect the defects on fingers. The detector can extract four-type defects on busbars: concave, convex, pinhole, and interruption and extract three-type defects on fingers: concave, convex, and interruption. In the experiments, the solar images of size 2048×2048 are evaluated. We measure the real size of solar cells by a 3D CCD for comparison with the measured scales to test the accuracy of the proposed measurement methods. Three pieces of solar cells are taken to evaluate the repeatability of the measurement. Almost 100 percentage defects can be detected by the proposed methods; moreover, the average inspection time is only 1.86 seconds for all defect types run on a PC with Intel Pentium 4 2.4GHz CPU. The experimental results show that the proposed approaches almost reach the practical stage.
