摘要 本研究針對PCB上的SMDs的缺陷有效結合數種演算法,以發展準確快速的印刷電路板SMDs檢測系統為目標。首先在待測影像方面,本研究是以彩色影像做為輸入,並分析在彩色影像中各色頻於瑕疵檢測時,所能達到之表現,再擷取出該色頻之影像加以分析。影像處理方面,本研究採用了區間式的二值化並結合統計方法,去掉不需要的背景。另外輔以中值濾波器、開合、閉合等運算減低雜訊。分析方面,本研究以倒傳遞類神經網路分析影像中的特徵,做初步的判斷且減少前處理的程序以達到快速檢測、分析的目的。最後本研究並整合上述理論技術建構成一印刷電路板SMT元件檢測系統,處理電阻、電容及IC等元件的瑕疵。
ABSTRACT This research uses several algorithms to develop an inspection system that efficiently detects the defects of the surface mount devices (SMDs) on printed circuit boards. Using the characteristics of color image, the inspected image will be analyzed for the performance of each band for inspect on, then the best performance band image is chosen. A dual threshold method with statistical data analysis is used in this research to separate the object and background. The medium filter, open, and close operations are used to reduce noises. Besides, this research uses the method of Back-propagation neural network for a pre-process to separate the defect types. Finally, these algorithms are integrated to develop an inspection system to detect the defects of resistor, capacitor, and IC.