網格結構和向量結構分別代表不同的資料抽象化概念，兩種結構各有其特 色，一般資料庫建構過程需要兼採這兩種方法來儲存地理資料。由於網格圖層的 資料量龐大，難以由人工直接數化產生，利用不規則分佈的樣本點來產生網格數 值的內插計算，是常被採用的方法，這種方法是地理資訊系統理論和實務的重要 課題。有關內插方法的比較，目前已經有許多相關研究成果，然而這些研究大都 是以傳統的數值分析方法來進行。類神經網路是人工智慧發展的一個新領域，具 有經驗學習與模式修正的功能，在數值統計和分析方面具有廣泛的應用發展潛 力。因此，類神經網路在內插作業上的應用也值得加以研究開發。本研究利用地 形資料為例，分別以兩種不同的地形資料，利用誤差後傳的類神經網路模式進行 內插。研究結果顯示這種模式是可以達成內插計算的效果，也證實其在內插計算 上的可行性，至於實際的品質和最佳化結果則有待進一步研究來君以探討。

　　　　　Under certain circumstances, there is a need to extract DTM data from topographic contour maps. The most common method is to interpolate the contours on the map into grid DTM data. Previous researches (reference) have demonstrated that the data quality of the DTM raster image is affected by the interpolation method employed, the quality of the original contour map, and the characteristics of the terrain. In terms of the interpolation method, existing GIS software provide traditional mathematical or statistical functions like IDW and kriging. In recent years, researchers (reference) have proposed artificial neural networks as an alternative approach to implement the spatial interpolation process. Artificial neural networks have been used in a broad range of applications, including pattern classification, pattern completion, function approximation, optimization, prediction, and automatic control. For spatial mapping roles, a function approximation or optimization technique may be applied, and researches have been investigating the use of these techniques with various kinds of neural network models, each focussing on a different aspect. Artificial neural networks are data-driven and do not require a priori knowledge of the study area. This paper investigates the use of neural networks for the generation of interpolated terrain data, and validates the results of the back error propagation model with reference to the real world.