多光谱遥感数据直接分类变化检测的神经网络方法研究

Research on Direct Classification Change Detection for Temporal and Multi-spectral Remote Sensing Data Using Neural Network

变化检测是近年发展起来的一种遥感时序数据处理方法,用于识别遥感数据在不同时间所记录的地表变化信息。采用传统的基于统计学的分类算法检测两个时期多波段遥感数据变化信息时,如果采取直接分类变化检测的方法会出现统计数据结构的奇异性问题,表现在同一位置上出现不同的光谱特征值。因此,该文提出和实验了使用基于样本和数据权重的自组织特征映射神经网络(SOFM)直接分类检测变化信息的方法。结果表明,SOFM直接分类变化检测法与两个时期最大似然方法分类后相减的结果相比,检测精度有显著提高。

Change detection is a newly developed method for identifying land surface changes recorded in temporal re-mote sensing data.When traditional statistic classification algorithms are normally used in change detection for temporal multi-spectral remote sensing data,a singularity structure problem occurred when direct classification change detection approach performed,which caused by different spectrum come from the same training area.In order to solve this prob-lem a Self-Organizing Feature Map neural network is developed,which is based on the weight of samplings and data.The result indicates direct classification change detection is better than post classification comparison based on maxi-mum likelihood classification algorithm.