Abstract:

Spectral band selection serves as an important part in hyperspectral image classification. In this paper, an improved binary cuckoo search algorithm for band selection in hyperspectral image is proposed. Binary cuckoo search algorithm is improved by these two ways, one of which is that we update the nests of offspring by using a binary encoding algorithm. Another one is that the found nests are updated based on the crossover mode of genetic algorithm. The improved binary cuckoo search algorithm achieves the goal of dimensionality reduction of hyperspectral image by finding the bands with low correlation and the vital function in the image. The improved binary cuckoo algorithm is applied to PaviaU datasets and AVIRIS datasets, compared with binary cuckoo algorithm, binary particle swarm algorithm, minimum redundancy maximum correlation algorithm, relief algorithm. The results show that the improved binary cuckoo search algorithm is more efficient in the band selection, and the selected bands are more representative and can improve the precision of the image classification.

Key words: binary cuckoo search algorithm, hyperspectral image, dimensionality reduction, band selection