激光与光电子学进展, 2020, 57 (23): 231202, 网络出版: 2020-12-08
基于中位数绝对偏差的异常训练样本探测方法 下载: 728次
Abnormal Training Samples Detection Method Based on Median Absolute Deviation
测量 遥感图像 光谱信息 监督分类 中位数绝对偏差 异常训练样本探测 measurement remote sensing image spectral information supervised classification median absolute deviation abnormal training samples detection
摘要
遥感图像的监督分类技术在信息提取和变化检测领域中具有广泛的应用,其中训练样本的选择至关重要,训练样本的好坏直接决定分类精度的高低。然而,受到条件的限制和人为的错误均可能导致一些不纯或错选的异常训练样本被选取,从而造成分类精度的降低。为了解决这个问题,采用中位数绝对偏差法,根据图像的光谱信息探测和剔除遥感图像监督分类任务中不纯和错选的训练样本。选取由Landsat-8获取南昌市部分地区的光学遥感图像数据,使用支持向量机对含有异常训练样本和剔除异常训练样本的两种情况进行监督分类,并对分类结果进行比较。实验结果表明,剔除异常训练样本的分类精度明显优于含异常训练样本。
Abstract
The supervised classification technology of remote sensing images is widely used in the field of information extraction and change detection, in which the selection of training samples is very important, and the quality of training samples directly determines the accuracy of classification. However, due to the limitation of conditions and human error, some impure or wrong training samples may be selected, resulting in a decrease in classification accuracy. In order to solve this problem, the median absolute deviation method is used to detect and eliminate impure and wrong training samples in the supervised classification of remote sensing images based on the spectral information of the image. The optical remote sensing image data obtained from Landsat-8 in some areas of Nanchang city is selected, the support vector machine is used to supervise and classify the two situations that contain abnormal training samples and eliminate abnormal training samples, and compare the classification results. Experimental results show that the classification accuracy of removing abnormal training samples is significantly better than that of abnormal training samples.
龚循强, 张方泽, 鲁铁定, 陈志高. 基于中位数绝对偏差的异常训练样本探测方法[J]. 激光与光电子学进展, 2020, 57(23): 231202. Xunqiang Gong, Fangze Zhang, Tieding Lu, Zhigao Chen. Abnormal Training Samples Detection Method Based on Median Absolute Deviation[J]. Laser & Optoelectronics Progress, 2020, 57(23): 231202.