Remote-sensed Monitoring And Analysis Of Invasive Alien Species Spartina Alterniflora In Shandong Province Based On Deep Learning Classification Method

Spartina alterniflora,produced in the coastal areas of the North American Atlantic and Gulf of Mexico,is one kind of grass plants of Chloris virgate Sw.,For the purpose of beach protection,S.alterniflora was first introduced in 1979 in China.S.alterniflora break out and becomes the most influential invasive species in coastal wetlands of China at present,attracting widespread attention from management departments and scholars.Accurate monitoring of S.alterniflora is the premise for its management and hazard assessment.Satellite remote sensing,as the preferred method for space-time continuum and large-scale observation,can effectively support the longterm sequence monitoring of S.alterniflora in units of provinces.The biological characteristics of S.alterniflora determine that its growing area must have periodic tidal effects.Meanwhile,the local ecosystem adjacent to or mixed with S.alterniflora is similar in the canopy structure and distribution of patches.These characteristics jointly determine the monitoring of S.alterniflora depends on optical remote sensing.However,the optical remote-sensed monitoring of S.alterniflora still has the following problems:(1)Multispectral remote sensing is widely used,but difficult to obtain high accuracy with common classification methods;(2)Hyperspectral remote sensing has natural advantages,but currently few studies aiming at classification for S.alterniflora,and facing the problem of low classification efficiency caused by high spectral dimensions;(3)The invasion of S.alterniflora has distinct regional features,but Shandong Province lacks targeted monitoring and analysis for features and process of S.alterniflora.Deep learning classification model has the advantages of automatically mining deep features and improving classification accuracy,solving above problems with potential.Therefore,this paper attempts to design classification methods of S.alterniflora based on deep learning,in order to solve low accuracy of multispectral image classification and low efficiency caused by high dimensions of hyperspectral images,and analyze the invasion process of S.alterniflora in Shandong Province based on the development methods.The work includes:(1)For the problem that common multispectral classification methods are shallow learning and difficult to improve accuracy,constructing deep learning S.alterniflora classification models that fused with shallow features;(2)For the problem that hyperspectral image has high bands redundancy,researching a deep classification model of S.alterniflora based on features reduction;(3)Carrying out dynamic monitoring and landscape analysis based on the development methods since 1989,and analyzing the driving factors and ecological effects of S.alterniflora in typical areas.This paper develops deep learning classification models and analyzes of S.alterniflora based on its distribution features using a long-term series of mediumresolution Landsat and GF-1 WFV multispectral images and GF-5 hyperspectral images,and provides support for marine resources and environmental management.The conclusions include:1.Constructing two deep learning classification models fused with shallow features,V-DCNN and V-RPNet.Results show that after integrating a total of ninedimensional spectral features of vegetation index and tasseled cap transformation components,the overall accuracy of V-DCNN and V-RPNet are improved by 0.34% and 1.19% respectively,the accuracy of S.alterniflora are improved by 3.25% and 1.86%,which effectively improved its monitoring ability.Taking the accuracy and training time into account,V-RPNet method is more suitable for analysis of S.alterniflora.2.Developing the RPNet deep classification method based on features dimensional reduction of GF-5 hyperspectral image.The results of S.alterniflora in the Shandong Yellow River Delta National Nature Reserve(referred to the Nature Reserve)show that the proposed S-FCBF bands selection algorithm has higher classification accuracy than S-Recorre and S-m RMR algorithm,and also better than FCBF.The highest overall accuracy and accuracy of S.alterniflora are 95.22% and 96.99% at 18 bands,and higher than full-band by 0.56% and 2.41% when integrating shallow features.Simultaneously,the training time can be shortened by up to 4 times after features dimensional reduction.3.The monitoring results of S.alterniflora in Shandong Province indicate:The area of S.alterniflora conformed to a polynomial growth since 1989,which can be divided into three periods: slow latent stage from 1989 to 1999,gradually increasing phase from 2000 to 2009,and a full-scale outbreak stage from 2010 to 2019.(1)The total area of S.alterniflora in Shandong Province reached 6431.41hm~2 in 2019 and distributed in all coastal cities.S.alterniflora in the Nature Reserve has the highest proportion,which is 64.49%;Laizhou Bay takes the second place.The remaining regions in descending order are Jiaozhou Bay,Dingzi Bay,Taoer River to the Old course of Yellow River,Rushan Bay,the coast of Rizhao City,and Wuhao Zhuang beach,areas are 515.29 hm~2,389.2 hm~2,244.28 hm~2,237.25 hm~2,41.18 hm~2 and 34.85 hm~2,respectively.(2)S.alterniflora outbreak came at varied time in different zones.S.alterniflora was few in 1998 in Wuhao Zhuang intertidal zone,its area reached maximum in 2009 and then decreased;In Laizhou Bay and Jiaozhou Bay,the introduction time were earlier than 1989 and the growth rates were slow before 2009,S.alterniflora in Jiaozhou Bay has been growing rapidly since 2013;S.alterniflora was first detected in the Nature Reserve in 2008,and it broke out in 2011 with the fastest spreading rate;S.alterniflora was detected in Dingzi Bay and Rushan Bay in 2009 with an outbreak time in 2013;S.alterniflora was detected in coastal areas of Rizhao in 2011 and near Chao River in 2013 in the east of Taoer River,keeping linear growth.(3)In general,the patch centroid of S.alterniflora moved to the northwest from 1989 to 2019,mainly due to it large-scale spread in Laizhou Bay and the Nature Reserve.With the increase of invasion ages,the shape heterogeneity and the patch fragmentation degree of S.alterniflora in Shandong Province has increased.(4)In addition to strong adaptability,natural factors such as wide silty tidal flats,fresh water replenishment,and seagrass beds are driving forces for fastest spreading of S.alterniflora in the Nature Reserve.Besides,water quality factors and artificial structures also affect its growth area.S.alterniflora plays a role in sustaining the coast,while its blocking of the tide causes degradation of the upper and middle intertidal ecosystems.