A Useful Technique for Target Detection in Hyperspectral Images

Hyperspectral Images are worthwhile data for many processing algorithms (e.g. Dimensionality Reduction, Target Detection, Change Detection, Classification and Unmixing). Target detection is a key issue in processing hyperspectral images. Spectral-identification-based algorithms are sensitive to spectral variability and noise in acquisition. In most cases, both the target spatial distributions and the spectral signatures are unknown, so each pixel is separately tested and appears as a target when it significantly differs from the background. On the other hand, there are many algorithms (e.g. Modified Spectral Angle Similarity (MSAS) as a Deterministic and Covariance-based Matched Filter Measure (CMFM) as sub-pixel approach) for target detection. As a new algorithm, Support Vector Machine (SVM) is a useful technique for Target Detection. In this paper, first we propose a theoretical discussion aimed at understanding and assessing the potentialities of MSAS, CMFM and SVM algorithms in hyper-dimensional feature spaces. Then, we assess the effectiveness of SVM with respect to conventional. To sustain such an analysis, the performance of SVM is compared with those of two other Target Detection algorithms, one-against-all, the one-against-one. Finally, Different performance indicators have been used to support our experimental studies in a detailed and accurate way (i.e., Target Detection accuracy, the computational time, the stability to parameter setting). The results obtained on a real Visible/Infrared Imaging Spectroradiometer hyperspectral dataset (CASI) allow concluding that, SVM is a valid and effective alternative to conventional Target Detection algorithms of hyperspectral remote sensing data.

Keywords:

Hyperspectral image , ROC curve , SVM , Target Detection.