LAUSR

Abstract With the development of remote sensing technology, more and more multi-temporal multispectral imagery becomes easily available, thus the research of target detection for this type of data will become indispensable. However, the traditional technology of target detection is generally designed for the single-temporal data. In this paper, we introduce the multilinear function as a mathematical tool to deal with the multi-temporal target detection problem for the first time. For an M time phases multispectral data set, we design an M th -order tensor filter, which corresponds to an M-linear function, to minimize the filter output energy while keeping the target output value invariant, named filter tensor analysis (FTA). Experiments using Landsat time series with two temporally changed targets (i.e. farmland and airport) and two temporally constant targets (i.e. roof and reservoir) all show the effectiveness of FTA for multi-temporal target detection under several commonly used evaluation indices.