Extraction Of Vegetation Types In The Typical Region Of The Yellow River Source Based On Spectral-spatial Joint And Deep Learning

The relationship between vegetation community and ecological environment is very close.Different vegetation communities evolve under specific ecological environment,which is also an important indicator of ecological environment.The vegetation community structure in the source region of the Yellow River is simple,the ecological environment is fragile,and the vegetation type is easily affected by the outside world.Therefore,to obtain accurate spatial distribution of vegetation types in the source region of the Yellow River is of great reference value for ecological environment protection and management in the source region of the Yellow River.Compared with the traditional field survey,hyperspectral remote sensing can quickly and accurately obtain the spatial distribution of vegetation types by using the acquired spectral information and spatial information.However,problems such as "dimension disaster","same matter with different spectrum" and " same spectrum with different matter" will appear when using hyperspectral remote sensing to extract vegetation types.In this paper,the Stacked Sparse AutoEncoder(SSAE)is selected as the classification algorithm of hyperspectral remote sensing images to solve the phenomenon of "dimension disaster" of hyperspectral remote sensing images.In order to solve the phenomenon of "the same object is different spectrum" and "the same object is the same spectrum" in hyperspectral remote sensing images.this paper uses two-dimensional Gabor filter and three-dimensional Gabor filter to extract the characteristic information of hyperspectral remote sensing images,and proposes 2D-Gabor+SSAE and 3D-Gabor+SSAE classification algorithms.In this paper,firstly,Support Vector Machine(SVM),Random Forests(RF),SSAE,2DGabor+SSAE and 3D-Gabor+SSAE five classification algorithm verifies the precision in a public data set and method of optimizing.Then,based on HJ-1A and GF-5 hyperspectral remote sensing images,selected public data set higher classification accuracy SSAE,2D-Gabor+SSAE and 3DGabor+SSAE three methods as classification algorithm.Completed the core region of the Yellow River source area-the eling lake in eastern and southern eling lake vegetation type extraction,and analyzed the eling lake vegetation type space distribution in eastern and southern eling lake vegetation types change characteristics.The results show that:(1)The 3D-Gabor+SSAE space-spectrum combined classification algorithm shows good classification accuracy in the open data set and the practical application of vegetation classification in the eastern and southern parts of Eling Lake.(2)In the eastern region of Eling Lake,there are abundant water resources,numerous wetlands,the most widely distributed high-quality forage grass,and some poisonous weeds.(3)In the southern region of Eling Lake,the quality forage grass decreased,the poisonous weeds increased,the drought-tolerant vegetation increased,and the grassland tended to deteriorate.Therefore,it is suggested to strengthen the grassland management in this region.