论文部分内容阅读
为验证理论训练数量(10~30 p)对参数分类器(如最大似然分类)、非参数分类器(如支撑向量机)的适用性以及样本特征(光谱统计、空间分布特征)对分类器分类精度的影响,选择不同规模的训练样本进行最大似然分类和支撑向量机分类,分析分类精度与样本之间的关系。实验结果表明:随着样本量的增加,最大似然、支撑向量机分类精度均随样本量增多而提高并趋于稳定,最大似然分类精度的增长速度要快于支撑向量机。MLC受样本量的影响较大,在小样本的时候(5个),分类精度不稳定,超过30个样本的时候,分类精度稳定下来;对于SVM分类器,在小样本的时候(5个),分类精度较高且稳定,因此SVM分类适合于小样本分类,不受限于理论样本量的影响。当样本量超过最小理论样本量值(30个)的时候,最大似然分类精度要优于支撑向量机,主要是由于当样本量增加后,最大似然更易于获得有效的信息量样本,而对于支撑向量机边缘信息样本的增加数量不大。研究结果为进一步优化样本进行分类打下前期的实验基础。
In order to verify the applicability of parameter training (such as maximum likelihood classification) and non-parametric training (such as support vector machines) as well as sample characteristics (spectral statistics and spatial distribution) to theoretical training (10 ~ 30 p) Classification accuracy of the impact of the choice of different sizes of training samples for maximum likelihood classification and support vector machine classification, analysis of classification accuracy and the relationship between the samples. Experimental results show that with the increase of sample size, the maximum likelihood and support vector machine classification accuracy increase and tend to be stable with the increase of sample size, and the maximum likelihood classification accuracy grows faster than support vector machine. MLC is greatly influenced by the sample size. The classification accuracy is stable when the sample size is small (5 samples) and the classification accuracy is unstable. For the SVM classifier, when the sample size is small (5 samples) , The classification accuracy is high and stable, so SVM classification is suitable for small sample classification, not limited to the theoretical sample size. When the sample size exceeds the minimum theoretical sample size (30), the maximum likelihood classification accuracy is better than the support vector machine, mainly because the maximum likelihood is more likely to obtain an effective sample of information when the sample size increases, and There is not a large increase in sample information for support vector machine edges. The research results lay the experimental foundation for further classifying the samples.