簡化骨架表達法及其在工件分類之應用

Abstract

本論文提出一個新的骨架表達法，稱為簡化骨架。此種骨架表達法是由一組連桿連接而成的樹狀結構，用以表達多邊形的整體外形資訊。簡化骨架與過去的表達法相比有兩個優點。第一個優點，簡化骨架比較簡單，骨架的每一段連桿都是直線段沒有曲線段。第二個優點，簡化骨架是比較好的整體外形資訊表達法，類似的多邊形其簡化骨架也會類似。 本研究最大的特色在提出骨架是由兩種連桿組成，分別是實連桿和虛連桿。本論文並證明只要實連桿骨架就可以描述整個多邊形，虛連桿只是表達實連桿的連接關係。過去的研究並沒有這種觀念，虛連桿被視為和實連桿相同，使得類似的多邊形其骨架不一定類似。如果骨架移除虛連桿直接以實連桿相連，則類似的多邊形其骨架的樹狀結構幾乎相同。因此簡化骨架是比較好的整體外形資訊表達法。 本研究並依據簡化骨架，利用類神經網路發展一個沖壓工件的分類系統，藉由分類結果來驗證本研究定義的骨架能確實的表達圖形之整體外形資訊。

This dissertation presents a new representation scheme for modeling the global shape of a general polygon, which is known as simplified skeleton. Compared to previous skeleton studies, the simplified skeleton has two distinctions. First, it is simpler in shape; that is, the simplified skeleton includes only straight-line segments and excludes curve segments. Second, the simplified skeleton is a better global shape descriptor for characterizing the skeleton links. The line segments of the skeleton can be classified into two types, real links and virtual links. In this paper, we prove that the set of all real links can be used to model a general polygon. The virtual links only represent the connection relationship of two real links. The skeletons proposed by previous research involve only one type of link; and even when a workpiece is slightly modified its skeleton might be quite dissimilar to its original one. Based on the simplified skeletons, a neural network approach is developed for classifying 2D punch parts. Testing results show that the proposed simplified skeleton is a good global shape descriptor.