應用影像視覺於超薄型表面載式電感器之瑕疵品辨識

A Defect-Inspection Technique for a Low-Profile Power Inductor Using Machine Vision

陳昭雄(Chaio-Shiung Chen) ； 鄭光宏(Wgoun-Hon Zuwn)

《科學與工程技術期刊》 2卷2期 (2006/06) Pp. 65-77

https://doi.org/10.7117/JSET.200606.0065

摘要

本文主要發展一影像處理技術包括影像定位、色彩分離、字型辨識及影像歸類等，以應用於超薄型表面載式電感器之瑕疵品線上檢測。瑕疵檢測主要分成兩部分進行，首先利用正交色彩空間轉換技術，以降低光線強度變化對色彩之影響，並歸納出銅色彩像素點空間分佈之區域，以辨識出電感器細小之鐵心破裂、斷線和線圈外露等瑕疵；再者，利用字元切割與細線化技術處理電感器上之型號，再歸納出各型號字元之端點和交叉點特徵，以辨識電感器印字模糊和混料等瑕疵。另外，本文為了提高型號字元之辨識率，發展一全新自適應性類神經網路系統以當做字型之歸類器，利用幾何慣性法則萃取字型之形狀特徵值，以當做此類神經系統之輸入，克服字型位移、旋轉和縮放等問題，並推導出一非線性最小平方誤差之學習法則，能自動調整隱藏層神經元之個數和所有參數，此學習法則能確保所有系統參數快速收斂，具備高度之字型辨識強健度。最後，將所發展出之影像檢測技術應用於實際之超薄型表面載式電感器生產線，以驗證所提方法之有效性，瑕疵品辨識率可達98%以上。

關鍵字

機械視覺 ； 影像辨識 ； 超薄型表面載式電感器 ； 自適應性類神經網路系統

並列摘要

This study develops image-processing techniques including image positioning, color separation, symbol recognition and image classification for on-line inspection of the defects in a power inductor by the use of machine vision. The inspection process is divided into two stages. In the first stage, color separation is applied to the RGB color space of copper. The spread region of normalized copper colors is then incorporated to inspect defects such as broken iron, appearance of coil, etc. In the second stage, the symbols on the inductor are recognized by the use of a node-extraction technique. The extraction features consist of the ends and nodes of a symbol. By using these features to identify the characters on the inductor, the copper-mixing problems can be solved. Moreover, a self-adaptive neural network is developed to improve the symbol recognition. Geometric moments are employed to extract the symbol features, which are used as the inputs for the neural networks. A recursive nonlinear least-square algorithm is implemented to auto-tune all parameters in the neural network, thereby accelerating the learning convergence and improving the approximation accuracy. Finally, the developed techniques are applied to practical power inductor manufacturing processes to confirm the validity of the proposed method.

並列關鍵字

machine vision ； defect recognition ； power inductor ； self-adaptive neural network system