高分一号影像地物识别精度分析

Accuracy Analysis of Object Recognition for GF-1 Images

高分影像具有更加丰富的地物信息,能够更加清晰地反映地表信息。采用监督分类和非监督分类对高分一号影像进行分类,并从面积精度和像元精度两个角度对分类结果进行精度验证,对分类精度进行比较和分析。结果表明,面积精度无法真实反映分类精度,而像元精度能够很好地反映分类精度;监督分类的分类精度整体上优于非监督分类,监督分类中最大似然法和支持向量机法的分类精度较高;基于地表真实感兴趣区的精度验证与基于目视解译的精度验证的结果就有较好的一致性,可以代表真实地类结果进行精度验证,但是采用基于地表真实感兴趣区进行精度验证时,总体精度值比实际高出11%左右,在精度验证时应当考虑此现象;采用传统分类方法对高分一号影像进行分类的真实精度总体上并不高,精度最高的是最大似然法,总体精度值为80.17%。

High-resolution remote sensing images have richer object information,and can clearly reflect surface information.We supervised the classification of GF-1 images by supervised classification and unsupervised classification at first.And then,we verified the accuracy of the classification results from the two aspects of area precision and pixel precision.Finally,we compared and analyzed the classification accuracy.The results show that the area accuracy can not reflect the classification accuracy,and the pixel precision can reflect the classification accuracy well.The classification accuracy of supervised classification is better than unsupervised classification.The methods of maximum likelihood and support vector machine in supervision classification are better than other methods.The accuracy verification of the surface real ROI has good consistency with the result of visual interpretation,and the real classification accuracy can be verified by use the surface real ROI.However the total accuracy value is about 11%higher than the real value when the accuracy verification of the surface real ROI is used,so the phenomenon should be considered in the accuracy verification.The real classification accuracy of GF-1 images by traditional method is not high.It is the maximum likelihood method with an overall precision value of 80.17%.