Study On Hyperspectral Remote Sensing Identification Method Of Dominant Mangrove Species

Mangrove is a wetland woody plant community growing in intertidal zone of tropical and subtropical coast.The dominant species of mangrove are diverse and often mixed.The identification of dominant species of mangrove can provide a basis for the integrity assessment of coastal wetland ecosystem.At present,hyperspectral remote sensing satellite data has the advantages of high spectral resolution,large number of bands and narrow width,and can capture local subtle information between similar mangrove species,which is an ideal means for fine classification of vegetation.Based on Zhuhai-1 hyperspectral remote sensing image,this paper studies mangrove classification in typical areas of Qi ’ao Island and Gao Qiao Town from three aspects: mangrove growth area extraction,spectral transformation and hyperspectral data reduction and classification methods.The main work of this paper is as follows:(1)Extraction of mangrove growing area.The land in the study area is divided into four categories: mangrove growing area,land vegetation,water body and others.Firstly,the classification features of land objects in the study area are extracted,including features after dimension reduction by Minimum Noise Fraction(MNF),texture features of five indexes and gray level co-occurrence matrix,and feature sets are formed.Then,the object-oriented classification method is used for classification,and post-processing is carried out to obtain the accurate mangrove growth range.(2)Hyperspectral analysis and dimension reduction of mangrove.The ground objects in mangrove growing area are divided into five mangrove vegetation,reeds,Spartina alterniflora and tidal flats which are easy to mix with mangrove.The original spectral characteristics of eight ground objects and the spectral characteristics after four spectral transformations are analyzed,and the hyperspectral data are dimensionalized based on three data dimensionality reduction methods.(3)Classification of mangrove species by hyperspectral remote sensing.Eight land objects in mangrove growing area are classified based on two traditional machine learning algorithms.By comparing the classification accuracy of the original spectrum and the data after four spectral transformations,it is found that the four spectral transformations have the effect of increasing the spectral difference of land objects,among which the data after continuous system removal transformation has the highest classification accuracy,and the classification accuracy of Random Forest(RF)algorithm is higher than that of Support Vector Machine(SVM).Comparing the classification accuracy of three hyperspectral data dimensionality reduction methods,it is found that the Optimal Clustering Framework(OCF)band selection method has the highest classification accuracy.Eight land objects are classified based on the improved hybrid feature learning network model,and compared with the original hybrid feature learning model and two dimensional-convolutional neural network,Comparing the algorithms of 2DCNN and 3D-CNN,the results show that the improved hybrid feature learning model has the highest accuracy and the overall classification accuracy is 96.10%.The best method and process of mangrove classification in Qi ’ao Island was applied to Gao Qiao Town,and the results of mangrove classification in Gao Qiao Town were obtained,with an overall classification accuracy of 98.28%.