彩色影像分類和檢測框法在電路板元件檢測上的應用

Abstract

電路板裝配完後，現今多靠人工檢測電阻、電容、積體電路等元件是否在正確位置。近來也有少數文獻探討自動化元件檢測，可是文獻中提及的方法多是對影像中每個像素逐一檢查，因此檢測時間過長。由於我們已知元件的大小形狀，也知道元件在電路板上的位置，因此在檢測元件是否在正確位置時，可以利用元件大小形狀給我們的資訊，不必檢查影像中每一個像素。本文為了更加容易分離電路底板和元件，使用彩色影像在RGB 3D的彩色座標中分類，只要顏色相同就視為同一類。元件缺陷判斷的部分，我們發展了一套以檢測方框為基礎的檢測法，設定一個像素寬度的長方形檢測方框。檢測方框上待測影像彩色分類後的類別和樣本類別相同的部份定義為T區段，不同的是F區段，根據T區段和F區段的數目和排列順序，可以知道缺件、插錯元件和如果元件有偏移時元件偏移的位置等元件裝配的缺陷。實驗結果證明彩色影像在RGB 3D的彩色座標中分類的確有存在的必要，而本論文元件缺陷判斷做法在檢測速度上比參考文獻表現優異。此外，本文在元件型號辨識部分，亦改良了與樣板比較的方式，克服元件型號印刷粗細不一的問題。

This paper presents a new technique for the inspection of defects on the Surface Mounted Devices (SMD) on printed circuit boards (PCB). There are four types of typical defects, namely, missing component, misalignment, wrong orientation of the IC chip and wrong parts. The inspection tasks are usually carried out by human inspectors visually and there are few references on automatic inspection method for components on PCB. Most of these references compare the PCB*s image with a given reference pattern pixel by pixel. As a result, the inspection process is very time consuming. Because we know the size, the shape and the position of the components, we can utilize these information and needn*t to compare the images pixel by pixel. In order to separate the components and the PCB more effectively, we use color image segmentation based on RGB 3D clustering, which classifies similar colors into the same class. On defects Inspection, we developed an inspection frame based method, which inspects the image on an one-pixel-wide rectangle frame. The experimental results show that the color image segmentation based on 3D clustering is more effective than the gray-scale based method in separating components for the PCB. Moreover, our inspection frame based method is faster than the method presented in other references. In addition, we also improved the mapping rule between the text on the sample and the template to overcome the non-uniformity problem of the oil-printed texts.