An Artificial Immune Algorithm for Spectrum Recognition of Hyperspectral Data

An intelligent approach based on artificial immune systems (AIS) is proposed in this paper to perform the task of spectrum recognition in hyperspectral data analysis. Although traditional spectral matching techniques have provided some confirmatory information to aid the interpretation of hyperspectral data, the improvement is yet to be made because of the complexity of the spectra. The immunological algorithm for spectral reactions is described in which a population of memory cells for each of the possible laboratory-derived spectral is evolved using artificial immune operators, such as, clone, mutation, and selection. In specially, the clonal and the mutation operators are two foremost processes. The clonal process can draw the evolutionary process closer to the goal. It raises the average affinity value and gives the following steps a good change to further move towards the solution, i.e. the known spectra. The mutation step generates random changes of single features to the individual solutions and helps the proposed algorithm to avoid local optimal value. By the above training process, a small well-trained specialist library is established for testing their pattern recognition ability. The recognition in the proposed algorithm is the automatic process to find all possible spectral responsible for the observed spectrum, analogous to the antibody’s recognizing antigen in the natural immune system. Whenever a spectrum is recognized for the first time, a copy of it is reserved as a new memory cell for the spectrum. Therefore, when it appears a second time, it can be easily recognized by the antibodies created during its first appearance. Consequently, the proposed method provides a learning methodology for pattern recognition. The proposed algorithm is compared with two well known spectral matching algorithms: binary coding and spectral angle mapper algorithms using simulated and real hyperspectral data. Experimental results demonstrate that the proposed approach can better recognize the unknown spectra than traditional algorithms based on a wellestablished specialist library obtained by different immune operators, and hence provide an effective option for spectrum recognition of hyperspectral data. * Corresponding author.