A NOVEL HYPERSPECTRAL TARGET DETECTION ALGORITHM FOR REAL-TIME APPLICATIONS WITH PUSH-BROOM SCANNERS

In this paper, we propose a novel method for the detection of desired targets in hyperspectral images which has been specially designed for being easily parallelizable in hardware devices for applications under real-time constraints, specially those which involve push-broom scanners. Concretely, the proposed methodology independently processes blocks of pixels as soon as they are sensed, reducing the amount of data to be stored and processed on-board and hence, speeding-up the executed process and fulfilling the causality requirements imposed by line-by-line applications. Based on a modified version of the well-known Gram-Schmidt orthogonalization method, our proposal selects a set of undesired signatures in order to later identify those spectral wavelengths which better distinguish the desired target to be detected from other entities present in the scene. With it, we achieve to perform a dimensional reduction strategy which compresses the acquired image onto a lower-dimensional feature space that maximizes the difference between both desired and undesired targets. We have applied the proposed methodology to two real hyperspectral images. The results obtained support the benefits of our proposal, being able to detect targets regardless of the amount of area covered by them.