| Hyperspectral images have many bands and high information redundancy.If all bands are directly involved in classification as features,it will lead to a large amount of calculation and low efficiency.If some of these bands are selected as features,some spectral information will be lost,resulting in unsatisfactory classification results.Therefore,how to effectively and fully use the spectral information to improve the classification accuracy has become the focus and difficulty of hyperspectral image classification.With the continuous development of nonlinear theory,multi-scale entropy algorithm has been widely used in various fields,but it has not been applied to the field of remote sensing.In order to make full use of spectral information in hyperspectral image classification,this paper introduces the multi-scale approximate entropy algorithm and the multi-scale sample entropy algorithm into hyperspectral image classification,and designs an algorithm to select the optimal multi-scale entropy feature.The specific research contents of this paper are as follows:(1)Firstly,the multi-scale approximate entropy feature calculation method is described,and the fluctuation of approximate entropy caused by the change of multi-scale approximate entropy parameters is analyzed.The reasonable embedding dimension and tolerance range are selected to analyze the change law of multi-scale approximate entropy curve of different ground object pixels,and the optimal multi-scale approximate entropy feature selection method is designed.Secondly,it describes the calculation method of multi-scale sample entropy characteristics,analyzes the entropy fluctuation caused by the change of multi-scale sample entropy parameters,selects a reasonable embedding dimension and tolerance range,analyzes the change law of multi-scale sample entropy curve of different ground object pixels,and designs the optimal multi-scale sample entropy feature selection method.Finally,the two selected optimal multi-scale entropy features are respectively substituted into the support vector machine(SVM)classifier for hyperspectral image classification.(2)In order to verify the feasibility and effectiveness of this algorithm,classification experiments are carried out on Salinas image,paviau image and Indian pines image respectively,and the results are evaluated quantitatively and qualitatively,and compared with classical hyperspectral image classification algorithm,approximate entropy algorithm and sample entropy algorithm under single scale.The experimental results show that the two algorithms in this paper not only effectively eliminate the redundant information contained in the image,but also retain all the spectral information in the image to the greatest extent,effectively improve the classification accuracy of hyperspectral images,and provide a new idea and choice for hyperspectral image classification.There are 29 figures,10 tables and 67 references in this paper. |
