本研究比較統計資訊彙集法(Statistical Meta-analysis)、Dempster-Shafer理 論、類神經網路(neural network)法等三種資訊融合技術在自動物件偵測系統中的應用。 實驗結果顯示此三種方法均能有效地從自然景觀影像中偵測出人造物。若以統計性第一類 與第二類錯誤為評估指標，統計資訊彙集法和類神經網路法具有接近的偵測效果。此外， 統計資訊彙集法擁有其他資訊融合技術所沒有的特性，即能利用上述兩種統計性錯誤來預 測新加入資訊源對整個融合系統的影響。此種特性很適合應用於評估多資訊源之偵測系統 。

　　　　　This paper compares three information fusion techniques in automatic object detection systems based on the statistical meta-analysis. Dempster-Shafer theory, and a neural network-based method. Experimental results demonstrate that all of the three approaches can effectively classify the man-made objects from natural scene images. It also shows that the statistical meta-analysis and the back propagation networks have almost the same performance according to the statistical type Ⅰand type Ⅱ errors of the results. On the other hand, the statistical meta-analysis can provide fusion criteria to foretell the effect of adding a new information source in terms of statistical type Ⅰand type Ⅱ errors before the source is actually combined. This property is not found in other information fusion techniques. While the statistical meta-analysis can be applied in information combination tasks, the neural network-based method will be played a powerful alternative approach for combining information in computer vision systems.