样本自适应多特征加权的高分辨率遥感图像分类

Sample-specific Multiple Features Weighting-based High-resolution Remote Sensing Image Classification

高分辨率遥感影像能够提供丰富的地物细节,但各种地物空间分布复杂,同类目标呈现出较大的光谱异质性,给传统模式识别分类器带来极大的挑战。提出了一种样本自适应多特征加权的遥感图像分类方法。常见的多特征组合分类器未能充分利用各种特征之间的局部相关性,提出通过分析测试样本局部特征相关性,探究各个特征在不同样本的分类中所占权重的不同,据此对不同分类器进行自适应加权。在一个大型遥感图像数据库上的实验结果表明,不同特征在遥感图像中对不同样本的分类作用是不同的,样本自适应特征加权法将平均分类精度从78.3%提高到90%。

High-resolution remote sensing image can provide rich feature details. However, a variety of terrain has com- plex spatial distribution, and spectral heterogeneity of similar landcovers appears largely, which bring great challenge to traditional pattern recognition classifier. For this purpose, this paper put forward a novel multi-classifier combination method for remote sensing image classification based on adaptive weights adjustment for different query samples. Previ- ous multiple features combination classifiers fail to make full use of local correlation among them, with a unifying weight for all the samples. This paper explored different weights of each feature in classification on different test samples, ac- cording to different local distributions. The experimental results on a large remote sensing image database show that dif- ferent features in remote sensing image classification of different samples have different effects, and the sample-specific multiple features weighting-based method presented in this paper enhances the average classification accuracy from 78.3% to 90%.