Study On Hyperspectral Remote Sensing Classification Based On Ant Colony Algorithm

The hyperspectral remote sensing imagery with hundreds of narrow continuous bands has great information redundancy. It makes the hyperspectral classification face many challenges.In order to alleviate the Hughes phenomenon that often happens in the hyperspectral remote sensing classification, we employed the ant colony algorithm(ACA) to select the optimal bands to reduce the spectrum feature dimensionality in this paper. Then, we proposed the hyperspectral remote sensing imagery classification rule extraction algorithm based on ant colony algorithm(HRSC-ACA), hyperspectral remote sensing classification algorithm based on the hybrid algorithm of ACA and CSA(HRSC-HA) and hyperspectral remote sensing imagery classification algorithm based on SVM optimized by ACA(HRSC-SA) for the supervised classification of the optimal band subset.(1) We designed the band selection of hyperspectral remote sensing imagery based on polymorphic ant colony algorithm(PACA-BS). The PACA-BS significantly decreased the searching space and time complexity. Thus, PACA-BS has more fast convergence speed than hyperspectral remote sensing imagery based on ant colony algorithm(ACA-BS). The performance of PACA-BS is more superior to ACA-BS from the following aspects: the lower computing time of PACA-BS, the better separability, larger information amount and higher overall classification accuracy of the band subset derived from Hyperion or AVIRIS by PACA-BS.(2) In this paper, we proposed the HRSC-ACA, HRSC-HA and HRSC- SA for the classification of hyperspectral remote sensing. Experiments show that the HRSC-HA algorithm can effectively improves the image classification accuracy. However, between the intervals of the image gray attribute exists serious overlapping phenomenon that lead to serious decrease of image classification accuracy. We discussed the application of ACA in the parameters optimization for kernel function of SVM and put forward HRSC-SA. These algorithms is used to the classification of Hyperion and AVIRIS images band subsets. The classification accuracy of is improved greatly through HRSC-SA.