連接式模型於震測水平連線

Abstract

在地球物理科學的訊號資料處理過程中，震測水平連接(seismic horizon linking)是一重要的處理程序。一水平震測訊號是由許多的peak組合而成。在本論文中，我們應用兩種連接式模型(connectionist model)用於連接所偵測到的震測訊號。利用強度(strength)和方向(direction)做為每一個處理單元(processing element)的輸入，而每一個處理單元都藉由與鄰近的處理單元的互動來調整自身狀態值。在第一個連接模型中，每一個處理單元與其八個鄰近的處理單元相連接互動；而另一個模型則是擴大連接的範圍，使每個處理單元能與更多鄰近的處理單元相連接互動。當整個網路收達到收斂後，輸出的結果即為連接的結果。此二模型接能有效的連接破碎的震測水平訊號。我們將此兩模型應用在實際(real)及模擬(simulated)的震測圖型(seismogram)中。從實驗結果中，我們能發現擴大與鄰近處理單元的連接模型所得到的實作結果較只與鄰近八個處理單元的連接模型的實作結果要佳。

Seismic horizon linking is an important task in geophysical data processing. A seismic horizon is composed to a set of peaks joined together. In this paper, two connectionist models are applied for seismic peaks linking and the horizons can be found in every linked pattern. The models take the strength and corresponding direction of every processing element as the input. The state of each processing element is update by its neighboring processing elements. In first one model, each processing element interacts with its eight neighbors. In another model, each processing element interacts over a large neighborhood. After convergence, the output of network is the linked result. These models can link the broken seismic horizon and delete noise in seismogram. We apply these two models on simulated and real seismogram. Results show that the connectionist model with large neighborhood outperforms the connectionist model with 1-neighborhood on seismic horizon linking problem.