臺灣博碩士論文加值系統

本論文實現了濾波逆投影法及類神經網路法之電阻抗影像重建演算法，利
用此二種演算法影像重建速度快，且其數學運算簡單，為單純之矩陣運算
，因此適合發展為平行處理之架構。另外本研究整合建立了一套架構於個
人電腦之電阻抗影像重建系統，使系統校正、資料量測及影像之重建，均
在同一使用者界面下完成，避免以往必須於個人電腦量資料，再將資料送
至工作站運算之繁複過程。有限元素法求解過程，由於系統主矩陣具有稀
疏性，因此利用稀疏矩陣求解法對於大運算量之計算，可節省相當之計算
時間。由結果可知，二維等帶寬高斯稀疏矩陣求解法較高斯消去直接法及
高斯－西得疊代法快。類神經網路演算法訓練樣本之產生、模擬退火演算
法與基因演算法之計算時間快慢主要決定於有限元素法之計算快慢。本研
究採用較少節點與元素(77個節點、120個元素) 之有限元素網目於上述三
種方法，使有限元素法求解速度加快。另外在模擬退火演算法方面，溫度
下降方式採動態改變型式，隨疊代次數之增加而變慢下降速度，使求解過
程有快速收斂之效果。

In this thesis, the methods of filtered backprojection and
l network for electrical impedance image reconstruction
aretigated. Both methods have the advantages of fast speed
andicity. They have the potential in parallel implementation.
Inion, a PC-based electrical impedance imaging system is
alsooped. All the procedures such as data acquisition,
systemration, and image reconstruction are integrated in a
user-dly GUI software system under MS Windows 3.1. Due to the
Sparsity of the FEM system matrix, a metod isyed for
reducing the computation time. From the results, it isthat
the 2D Gauss-Sparse method is faster than Gauss-nation
method and Gauss-Seidel iterative method. In this studyFEM
containing 77 nodes and 120 elements is used to
generateraining data sets for neural-network method and the
theoreticalction data sets for SA and GA methods. In SA method,
decreasingrature dynamically would make the algorithm converge
rapidly.