论文部分内容阅读
Fisher判别分类的好坏关键在于训练样本集选取的精度和在降维过程中样本特征信息的损失程度,基于此问题,本文根据不同时相同一地区的遥感影像的差值影像中各像素本身的灰度值及其邻域平均灰度值特征获得其一维和二维直方图,针对差值影像无噪和带噪两种情况,根据直方图信息选取Fisher判别分析所需的训练样本,同时为了尽可能降低判别分析过程中有用信息的损失,将所得到的原训练样本集进行非线性变换,使其映射到高维空间中,利用映射后的训练样本求得Fisher判别规则。实验结果表明:与基于原训练样本的Fisher判别分类和基于寻找更多的样本特征的Fisher判别分类方法生成结果相比,在差值影像无噪和带噪情况下,本文提出的方法具有更好的变化检测精度和抗噪性。
The key of Fisher’s discriminant classification is the accuracy of training sample set selection and the degree of loss of sample characteristic information in the process of dimensionality reduction. Based on this problem, based on the difference between each pixel in remote sensing images Gray value and its neighborhood average gray value feature to obtain its one-dimensional and two-dimensional histogram for the difference between the image noise and noisy two cases, according to the histogram information to select the Fisher discriminant analysis of the required training samples, Reduce the loss of useful information in the discriminant analysis process as much as possible, transform the obtained original training sample set nonlinearly, map it into the high-dimensional space, and use the mapped training samples to obtain the Fisher discriminant rule. The experimental results show that compared with the Fisher discriminant classification based on the original training samples and the Fisher discriminant classification method based on the finding of more sample features, the proposed method is better in the noisy and noisy images The change detection accuracy and noise immunity.