A new target detection algorithm: spectra sort encoding

Authors:
Publisher website
Google Scholar
Abstract
A new target detection algorithm named the spectra sort encoding (SSE) algorithm is presented in this paper. It has two steps. Firstly, relative error (cErr) between the reference and image spectra is calculated. When cErr is greater than the relative error defined by the user (rErr), the current pixel is regarded as none‐target and encoded as zero. Otherwise, the second step is executed to confirm whether or not the current pixel is target. In this second step, the similarity between reference and image spectra is calculated by sorting them respectively. When the similarity is greater than the identification error (the least error limits between reference and image spectra) defined by the user, the current pixel is encoded as one; otherwise, it is encoded as zero. A detailed description is provided of the backgrounds and principles of SSE, and its accuracy is evaluated using multiple categories. Experiments indicated that when the identification error is 15%, the mean accuracy of SSE is 95%, which is 41.9% higher than that of constrained energy minimization (CEM) and 46.9% higher than that of spectrally second‐order derivative (SSD). Results of target detection experiments using Enhanced Thematic Mapper Plus (ETM+) and Hyperion images revealed that it could be used in both multispectral and hyperspectral remote sensing.

This publication has 15 references indexed in Scilit: