改善TM卫星遥感图像自动识别精度的研究和应用

A Study on Improvig the Classification Accuracy on TM Remote Satellite Image

介绍利用计算机处理ＴＭ遥感图像数据产生重庆市江北县典型地区土地利用分类图技术。为克服单个像元由于光谱混杂和其它噪声引起的分类不确定性，对图像数据实施了两次分类处理。首先用最大似然监督分类法对像元点做一次分类，并且在修改训练样本和先验概率，分层处理等方面采取了灵活的对策。继而采用局部领域覆盖法进行二次分类，利用空间信息进一步改善分类精度。在重庆江北县地区所做的实验研究结果表明，上述做法有效地提高了分类精度，是合理的，算法简明实用。

Classified landuse map in the typical area of Jiang Bei County, Chong-Qing has been developed. The Landsat TM data was to product this map by digital comptiter. The key problem in the classification is how to determine many uncertan pixels that was effected by the mixing of spectral or other noises. To resolve this question,a two steps procedures was proposed. First,the maxmi-likelihood method was tised to carefully classify each orig-inal data. Then the local area was processed by a land cover classifed method according to the previous results. The based ideal of this method is to untilize the priorknowledge and space information to improve the classified results, In fact, a satisfactory results has been obtained.This method is specifically used to classify the land in the multi-vegetable and broken area.