摘要 隧道開挖工程,由於大量隧道影像的判斷,都需經由專業的地質師來分析隧道的資料,才能進行開挖的工程,因此,其工作量通常較龐大,而分析的工作,也多受限於概略性之數值分析,而有欠精確性。 本論文的目的為利用電腦的自動辨識技術來分辨:(1) 分辨隧道影像中不同的岩層,並劃分擷取出不同岩層的區域;(2) 分辨隧道的紋理,也就是隧道影像上的弱面分析,以供專業地質師利用影像上弱面的座標數據,套入專業的隧道數據分析程式,期能輔助地質師分析龐大的隧道影像資料。 本論文所採用之方法,利用一階微分遮罩與數學形態學,分別利用快速傅利葉轉換套入碎形分析,再利用Karhunen-Loeve來合併取得岩層的分類圖,可辨識出隧道中不同的材質的岩層。利用高斯平滑法與一階微分遮罩的技術,來分析出隧道影像的弱面。就模型圖庫而言,本實驗結果顯示提出的方法是有效的。另外,隧道影樣套用結果亦有初步的成果。 隧道影像由於受外界干擾訊號繁雜,加上取得影像品質不一,而有些隧道影像雖然清晰,但是相似度極高的問題,以致於影響到辨識結果,都是隧道影像辨識未來需要考慮的問題。
Abstract The tunnel excavation project, that is associated with a large number of tunnel images, requires the professional geologist to analyze these images such that the project could be implemented. Therefore, the amount of work load is generally massive. In addition, the analysis is limited to brief data analysis, and may subject to imprecision. The purpose of this thesis is to use computer to automatically distinguish the following: First, distinguish different rock texture in the tunnel images, and classify different area of rock masses. Second, distinguish rock discontinuities using tunnel images. Such rock mass analysis of the tunnel image, can be used to provide professional geologist to utilize data of rock mass coordinates on the image, and set up the specialized tunnel data analysis procedures. The goal of this thesis is to help geologists analyze the massive amount of data. This thesis utilizes a first-order differential mask and Mathematical Morphology, the Fast Fourier Transform and Fractal Analysis, and the Karhunen-Loeve expansion to obtain a classification graph of the rock masses. In addition, Gaussian Smoothing and a differential mask are used to analyze rock discontinuities on the tunnel images. The experimental results show that our method is effective in classifying a set of model images. For the tunnel images, our method also achieved a promising result. The tunnel image can be corrupted by unexpected interference. The problem is also coupled with the fact that some images are of bad quality in uncontrolled environment. Although some tunnel images are quite clear, but inherent similarity is high, thus may affect the classification result. Future investigation must address these problems.