次分類神經網路在森林覆蓋類型預測的應用

Using Sub-Category Neural Networks for Predicting Forest Cover Types

葉怡成(I-Cheng Yeh) ； 林育仕(Yu-Shih Lin)

《中華林學季刊》 41卷1期 (2008/03) Pp. 59-75

https://doi.org/10.30064/QJCF.200803.0005

摘要

雖然倒傳遞網路可以建構準確的分類模型，但無法發掘隱藏在分類中的「次分類」。為了改善倒傳遞網路不能產生次分類的缺點，在此提出次分類神經網路(Sub-Category Neural Network, SCNN)。為證明此一架構優於傳統的倒傳遞類神經網路(Back-Propagation Network, BPN)，本文以二個人為的分類例題及一個真實的森林地表覆蓋類型（樹種）分類問題進行比較。由實驗結果歸納出下列結論：(1) SCNN的準確度與BPN相近。(2) SCNN可以將部份分類的「次分類」發掘出來。

關鍵字

倒傳遞神經網路 ； 森林地表覆蓋 ； 分類 ； 次分類

並列摘要

Although the back-propagation algorithm can build accurate classification models, it can not discover the sub-categories in the data set. To solve this problem, this study proposed a novel neural network model, Sub-Category Neural Network (SCNN). To prove the performance of SCNN, two artificial classification problems as well as one actual forest cover classification problem were employed to test and compare with back-propagation network (BPN). The results proved that the accuracy of SCNN is about the same as BPN, while it can discover the sub-categories in the data set.

並列關鍵字

Back-propagation network ； Forest cover ； Classification ； Sub-category