Research On Hyperspectral Image Classification Method Based On Improved LSTSVM

With the development of technology,the resolution of the sensor is gradually improved,The obtained hyperspectral image can provide more detailed information about the feature category,and its classification technology has become an important part of the research in the field of remote sensing.People have been optimized and improved the classifiers and achieved good results.However,high-dimensional nonlinearity of hyperspectral data,high correlation between bands,and fewer labeled samples pose great challenges for the classification of hyperspectral images.Therefore,how to quickly and accurately classify hyperspectral images is becoming a research hotspot when there are few labeled samples.The semi-supervised algorithm not only utilizes the information of the labeled samples,but also uses the information of the unlabeled samples to further improve the classification accuracy of the hyperspectral image.In view of the above problems,based on the improved classifier,this paper deeply studies the semi-supervised algorithm and proposes a new semi-supervised learning algorithm.The details are as follows:1.A semi-supervised classification algorithm combined with band selection is proposed.Firstly,the method selects the band selection method to remove the redundant information in the hyperspectral image,thereby reducing the complexity and improving the generalization ability.Then,the differential evolution algorithm cross-mutates the unlabeled samples,and selects the samples with high confidence to expand into the labeled sample group.To improve classification accuracy.The experimental results show that the method can effectively improve the classification accuracy and classification speed of the classifier in the case of small samples.2.The use of physical features in the fixed area has the characteristics of spatial coherence,and at the same time integrates spatial information into the classification algorithm.The training set is extended by the selection of neighborhood samples;the classifier with spatial information is used;the classification result is combined with the neighborhood information to improve the classification accuracy.The experimental results show that the method can further improve the classification accuracy and classification speed of the classifier under small sample conditions.