Generating Spectral Signature Library for Patterned Object in Hyperspectral Images

Small objects cover within a few pixels or even cover subpixels due to the low spatial resolution of hyperspectral cameras. This factor significantly reduces the detection performance for multi-pattern objects. For this reason, it is important to obtain signatures that can model the target in best way while creating spectral library signatures. In this study, a representative signature generation method for the patterned objects is proposed in order to determine the target in hyperspectral aerial images. The target detection performance, which was calculated by using the pattern signatures of the object is compared with the target detection performance using the Mean Signatures obtained by the average of the signatures from a selected area. In addition, the target detection performance of the average signatures obtained from randomly selected areas are also compared. According to the results, it is observed that the target detection performance of the Mean Signatures is higher than that of the the pattern signatures.