面向土地利用分类的HJ-1 CCD影像最佳分形波段选择

环境一号卫星（HJ-1）CCD影像光谱波段较少，地物之间的准确分类识别有一定困难。采用分形纹理辅助地物分类识别是一种有效方法，而波段选择是提高分类识别精度的关键。本文以江西赣州定南县土地利用分类为例，采用双毯覆盖模型对HJ卫星CCD影像6类典型地物的波谱分形特征进行了分析，利用不同地物在不同波段上的分形区分度差异构建了最佳分形波段选择模型，并利用该模型挑选出最佳分形波段来辅助土地利用分类，最后对分类结果进行检验。结果表明：最佳分形波段选择模型能够综合权衡不同地物在不同波段上的分形区分度差异，利用挑选出来的最佳分形波段来辅助分类，其分类总体精度相对于原始影像分类提高了11.77%，相对于第1主成分分形辅助下的分类提高了1.56%。

Optimal fractal band selection on HJ-1 CCD image for land use classification

Classifying and identifying different objects from the Chinese environmental satellite of HJ-1 CCD images, is difficult because only a few spectrum channels can be used for the used data. A selection of spectrum band is the key to improve classification accuracy when using the fractal texture method on features classification and identification from image data. This paper presents a case study focused on land use classification in Dingnan county, Ganzhou city, Jiangxi province. A double blanket coverage model was used to analyze the spectral characters of the six typical objects from HJ-1 CCD images. An optimal fractal spectrum selection model was adopted to pick up the optimal fractal bands in land use classification. This model was developed based on the diversity of the fractal differentiation of different objects in various spectrums bands. The classification results were validated and assessed. Results show that the proposed optimal fractal band selection model can comprehensively balance the diversity of the fractal differentiation of different objects in various spectrums bands comprehensively. Unlike in the original image classification, the use of the optimal fractal spectrum band to assist classification led to the increase in the overall classification accuracy and in the first principal component fractal by 11. 77% and 1. 56%, respectively.