Research On Band Selection For Hyperspectral Image Based On Optimal Subset Criterion

Hyperspectral remote sensing contains rich spectral information while massive data greatly increases the space-time complexity of data processing and affect the image processing results.An effective band selection method can greatly improve both the speed and effect of hyperspectral images process.Therefore,the research of band selection technology of hyperspectral image is of great significance.In this paper,we mainly study the framework of the hyperspectral image band selection algorithm based on the optimal subset select criterion and the process of the specific algorithm.The main contents of this paper are as follows:(1)The first part introduces the background and significance of hyperspectral image band selection.Summarize and analyze the research status at home and abroad.In addition,we illustrate the organizational structure of this paper.(2)The second chapter systematically introduces the relevant basic knowledge,including the principle of particle swarm optimization algorithm and parameter settings,and the common clustering algorithm,which provide an important theoretical basis for the following several band selection algorithms.(3)Introduce the optimal subset select criteria and the common search strategies for hyperspectral image band selection,and lead to the criteria and search strategy are used in this article.The algorithm framework of direct optimization based on optimal subset criterion is constructed.On the basis of this framework,a band selection algorithm based on minimum noise band selection(MNBS)criterion is proposed,the algorithm uses the continuous forward selection(SFS)to search a set of better band subsets to initialize the initial particle position list of the particle swarm algorithm,which reduces the impact of the initial sensitive problem.The framework of band selection algorithm based on clustering and optimal subset optimization is constructed.Based on this framework,a band selection algorithm based on MNBS criterion for direct optimization of particle swarm optimization is proposed,which uses spectral clustering algorithm to cluster the bands firstly.Finally,the framework of band selection algorithm based on critical band selection and optimal subset optimization is designed.The clustering performance index based on spectral angular distance measurement is proposed,combining with visual assessment of cluster tendency(VAT)to confirm the cluster number.Then a band selection algorithm is proposed based on the critical band extraction combined with MNBS criterion of particle swarm optimization.(4)Introduce the RX anomaly detection,k nearest neighbor classification and support vectormachine classification methods,which are used in the experiment.Through the experimental results of anomaly detection and classification,the performance of various band selection methods is compared and analyzed.(5)The last chapter summaries the whole work of the paper and put forward the future work on band selection based on optimal subset.