利用二維投影視圖進行工件分類之研究

Abstract

本論文提出一種利用二維視圖進行工件分類的方法，此一分類方法是 先利用工件二維視圖之外維輪廓，再利用其內部訊息而來，因此稱之為二 階段工件自動分類法。 工件分類方法在第一階段是利用Wu和Jen在1995 年所提出的分類方法，先將三維方形工件依其二維視圖之外圍輪廓予以外 形上的分類。接著第二階段再依工件二維視圖所產生的內部訊息予以更細 部的分類。 在第二階段中，為了能完全表達一個工件的內部訊息，本 文以六個實線二維視圖，及三個虛線二維視圖來表達。而後將代表工件的 九個二維視圖以網狀圖結構來表達，網狀圖結構中節點表示二維視圖中實 線或虛線所分割出的不同區域，而弧線表示分割區域的銜接情形，弧線上 的權值則表示分割區域的銜接邊數，並以一環狀碼來表示分割區域的形狀 ；而後再將網狀圖結構以一規則轉換成參考樹的樹狀架構，接著把樹狀架 構轉換成以向量的方式來表達，此向量即為類神經網路之輸入向量，最後 再藉由此類神經網路對二維視圖作分類，並且根據上述分類的結果將工件 形成工件族。 This paper present a two-stage approach to the classification of 3Dprismatic parts based on their 2D projections. In the first stage, workpieces are classified into groups based on the contours of their three projections,this portion of work is essentially adopted from Jeng's work(1995). In thesecond stage, the workpieces in a pariticular group can be further classifiedinto subgroups by referring to the interior information of nine projectionviews. Of the nine views, six denote the visible projections and the other three represent the invisible projections. Any of such 2D views in appearance is a 2D region subdivided into seveal subregions; therefore itcan be described as a graph s tructure by modeling subregions as graphnodes and their neighboring relationships as graph arcs. The graphstructure for representing a 2D view is subsequently converted into a tree structure, and finally converted into a vector format for application of the back-propagation neural network model, a widely known vehicle for pattern classification. Extensive empirical tests have been performed; the pros and cons of the approach are also investigated.