The Classification of remotely sensed data is the main theme of Remote Sensing Image Understanding and Analysis

while the artificial neural networks （ANN） is one of the latest and most important techniques developed recently in the area of connective artificial intelligence. In this paper

after we made a through study on the structure of the multi

l

ayer perceptron （MLP） and deeply analyzed it’s back

p

ropagation （BP） training algorithm

the framework of how to integrate Geo

k

nowledge with ANN and apply to RS classification is put forward. Firstly

the suggestions of improving the efficiency of BP algorithm

including network architecture selection

use of optimization on learning rate

and assistance with additional data and expert knowledge etc.

are presented. Then

after the general approach of ANN based RS image classification is reviewed

the model of integrating Geo

k

nowledge with ANN for RS image classification is developed with specific experiment of RULE based MLP. Experimental result shows significant improvement in comparison with statistical and traditional ANN classifiers.