リモートセンシングによる徳島市中心部の土地利用状況解析

概要

論文の詳細を見る
In this paper, we propose a pattern classification method for remote sensing data using two neural networks; one (NN1) is trained by a back-propagation method and another (NN2) by self-organized feature mapping and knowledge-based processing. The NN1 has the ability to recognize complex patterns and classify them. However, it has two disadvantages : it may misclassify the patterns, and it is difficult to choose a training set. On the other hand, the NN 2 doesn't need the training set, and a knowledge - based system which uses human geographical knowledge improves the Classification results, compared with the conventional statistical method. We propose a pattern classification method that integrates advantages of both the neural networks and the knowledge-based system. The proposed system is divided into three sub-systems which consist of a preprocessing component, a recognition component, and an error correction component. We use the NN2 for choosing the training set as a preprocessor of the NN1, the NN1 for classification, and the knowledge-based system for correcting mis-classification created by the NN1. Experimental results illustrate the performance of the proposed system.
四国大学の論文
1995-12-05