標題: 循序式模擬退火系統於圖形偵測與震測圖形識別之應用Sequential Simulated Annealing System for Pattern Detection and Seismic Applications 作者: 周映良黃國源多媒體工程研究所 關鍵字: 模擬退火;霍夫演算法;震測圖形;Simulated Annealing;Hough transform;Seismic pattern 公開日期: 2007 摘要: Hough transform (HT) 通常用來偵測影像中的直線、圓、橢圓、和雙曲線。但此方法需要大量的計算，且耗費大量記憶體空間。 我們提出一個運用模擬退火演算法的循序式偵測系統來偵測影像中的直線、圓、橢圓、和雙曲線之參數。循序式模擬退火圖形偵測系統可搜尋到一組圖形參數，使得影像上的點到這組圖形的距離為最小。在偵測影像中的圖形時，我們不同時偵測所有圖形總類，而是以循序式的觀念，一次偵測一個種類的圖形，並將偵測到的圖形所包含的點移除，再對剩餘的資料重複偵測。 我們先將循序式模擬退火圖形偵測系統應用於模擬的影像資料上，偵測模擬影像成功之後，再將之應用於偵測模擬的與真實的單炸點震測圖形上，偵測直接波圖形 (直線) 與反射波圖形 (雙曲線) ，對於瞭解地層特性及後續的研究有很大的幫助。我們也將循序式偵測系統應用於偵測CDP圖形中的雙曲線，並利用偵測出來的雙曲線參數，推算出地層中每層波的旅行速度 (interval velocity)，此結果可使速度分析達到自動化的效果。Usually Hough transform (HT) was used to detect lines, circles, ellipses, and hyperbolas in an image. But it needed huge calculation and consumed the memory. Using simulated annealing (SA), we propose the sequential detection system to detect the parameters of lines, circle, ellipse, and hyperbola in images. The proposed sequential simulated annealing pattern detection system has the capability of searching a set of parameter vectors with global minimal error. In pattern detection, instead of detection the patterns of all types at one step, we use a sequential concept to detect the patterns of a type at one step. Then we remove the detected patterns. The reaming data will be used in the SA to detect the patterns of other type. Using sequential simulated annealing pattern detection system to detect simulated images can get good detections. The method is also applied to detect the parameters of direct wave (line) and reflected wave pattern (hyperbola) in simulated and real one-shot seismograms, the results can improve seismic interpretation and further seismic data processing. We also use the system to detect the hyperbolic patterns in the CDP image. The detected parameters can be used to calculate the interval velocity in each layer. The results makes velocity analysis automated. URI: http://140.113.39.130/cdrfb3/record/nctu/#GT009557531http://hdl.handle.net/11536/39684 Appears in Collections: Thesis

Files in This Item: