采用标准的多分类器结合方法进行遥感图像的分类研究。首先介绍了标准的多分类器结合的算法,然后以Landsat-TM多光谱遥感数据的土地覆被分类为例,分别给出了抽象级上相同训练特征的多分类器结合、抽象级上不同训练特征的多分类器结合和测量级上的多分类器结合进行土地覆被分类的方法,并进行了实例研究。参与分类器结合的单个分类器包括最大似然分类器,最小距离分类器,马氏距离分类器,K-NN分类器,多层感知器神经网络分类器。分类器的分类精度用总体精度、用户精度、生产者精度、kappa系数和条件kappa系数评价。结果表明,每一种多分类器结合的分类方法都能够比较显著地提高总体分类精度。文章最后对不同多分类器结合方式的优缺点进行了分析。

Deriving thematic maps by classifying remotely sensed data was a major application fields of remote sensing techniques. The most often used classifiers in classification process of remotely sensed data include various statistical classifiers and artificial neural networks. Comparisons among these classifiers found no classifier as "panacea". While most efforts were made to develop new classifiers for more accurate classification results, to fully exploit the potentials of the existing classifiers by combining multiple existing classifiers is an effective way in many fields of pattern recognition applications. In this paper, the standard multiple classifier combination method was used for land cover mapping using remotely sensed data. Landsat TM data in Lanier Lake was used as an experimental data. Land cover maps were derived by combining classifiers at abstract level with same training features, combining classifiers at abstract level with different training features and by combining classifiers at measurement level respectively. Classification accuracies of these maps were compared with those of classifiers combined. Results showed that for all classifiers combination methods, the classification accuracies were improved. Advantages and drawbacks of every method of classifiers combination were analyzed and further study in combining multiple classifiers for remotely sensed data classification was suggested.