结合粗糙集约简的纹理特征影像分类

Image Classification of Texture Feature Combined with Rough and Intensive Reduction

基于光谱特征的影像分类精度过低,不能满足生产的需要,所以研究利用其他辅助手段来提高遥感影像的分类成为未来发展的一个重要方向。使用灰度共生矩阵对研究影像进行纹理特征提取,得到8种纹理特征,然后利用粗糙集约简中的遗传算法对8种纹理特征进行纹理特征选择,最后得到一组最优组合;利用这一最优组合与原始影像融合产生新的影像,对新影像进行分类。通过实验对比分析证明,约简后的纹理特征辅助光谱特征分类能够提高遥感影像分类的准确性和精度。

Abstract：Owing to the fact that low image classification accuracy of spectral characteristics can＇t meet the needs of production, it becomes an important direction to adopt other auxiliary means for improving classification of re- mote sensing image. This study used the gray level co-occurrence matrix to extract texture features of the re- searched images, and then obtained 8 kinds of texture features. Taking advantage of the genetic algorithm of rough and intensive reduction, it proposed a set of optimal combination. New images were created and classified by uniting the optimal combination and the original images. The experiment comparison and analysis showed that the classification of spectral characteristics with the reduction of texture features could improve the accuracy and precision of remote sensing image classification.