遥感影像像元不确定性对SVM分类结果可靠性影响研究

Influence of Remote Sensing Image Pixel Uncertainty on SVM Classification Reliability

在遥感影像获取过程中,由于地表环境复杂,传感器参数等因素影响,影像存在不确定性。通过分析不确定性产生的原因及特征,在像元尺度上构建像元光谱多义性、灰度差异性、分类信任度3个指标来度量影像不确定性。在此基础上,利用线性回归模型分析像元不确定性对支持向量机分类结果可靠性的影响。结果表明:像元不确定性对分类可靠性有一定影响,更具体地,像元不确定性越大,影像分类的总体正确性和一致性越低,基本呈现线性递减关系;像元不确定性对分类完整度和可靠度的影响在类别和影像间存在差异,其中,耕地和林地的分类完整度和可靠度与像元不确定性的线性关系较强。

Due to the complex surface environment,sensor parameters and other factors,there are uncertainties in remote sensing images. Based on the analysis of the causes and the characteristics of the uncertainty,the pixel spectral ambiguity,the grayscale difference and the classification credibility of pixels are constructed to measure the uncertainty of image.On this basis,the linear regression model is used to analyze the influence of pixel uncertainty on the reliability of classification results obtained by support vector machine. The results show that the pixel uncertainty does have a certain impact on classification reliability. More specifically,the greater the pixel uncertainty,the lower the overall accuracy and consistency of image classification,and a linearly decreasing relationship appears between them. The influence of pixel uncertainty on classification integrity and reliability varies between categories and between images. There is a strong linear relationship between the classification integrity and reliability of cultivated land and forest land and pixel uncertainty.