Class-specific artificial immune recognition method for hyperspectral image classification

Artificial immune recognition system (AIRS), as an efficient and successful computational intelligence method, has been widely used for classification. However, this method is seldom used for hyperspectral image classification due to its complexity. To address this problem, a class-specific model based on AIRS, named as Single Class Learning Network AIRS (SCLN-AIRS), is proposed in this paper to improve the classification accuracy for hyperspectral images compared with AIRS based method. For SCLN-AIRS, the outliers of training samples from irrelevant classes are ignored first. Then, a novel MC evolution strategy is proposed to prevent memory cells being affected by other ones from different classes. In the novel model, the calculation complexity is guaranteed by the fact that the class is expressed only by few memory cells while classification result is improved. Experimental results on AVIRIS dataset demonstrate the effectiveness of the proposed method for hyperspectral image classification.