Research On The Granularity Effect Of Mangrove Species Classification And Landscape Pattern Based On Remote Sensing Images

Mangroves grow in the intertidal zone,which can purify water bodies,resist wind and soil fixation,reduce coastline erosion caused by tidal waves,storm surges and tsunamis,and effectively resist wind and wave attacks.However,with the deterioration of the environment and improper development,the global mangrove area has declined sharply,and it is imperative to protect mangroves.The sustainable protection of mangroves requires better monitoring of the dynamic information and structure of mangrove succession and distribution.The development of remote sensing technology provides the possibility to meet this demand.This paper takes mangroves as the center,and takes the mangrove natural reserves in Shenzhen Bay area(including Futian Nature Reserve and Mipu Nature Reserve)as the research area,and introduces the research background,significance,status and red mangrove classification.The relevant situation of the forest and its function and value.The OVS-1,Landsat-8,Sentinel-2 and World View-2 remote sensing datas were selected as data sources,and the applicability of hyperspectral data in the inter-species classification of mangroves in the region was investigated.The spatial granularity effect of mangrove classification in this area and the effect of different spatial granularity effects on the landscape pattern index of the study area.Through research and analysis,compared with artificial neural networks and support vector machines,the classification results of stochastic forest classification are the best,and the overall accuracy obtained is 73.77%.The comparison with Landsat-8 and Sentinel-2 image data shows that higher spatial resolution is conducive to improving the accuracy of mangrove classification results,and the short-wave infrared band has a very positive effect on improving the classification accuracy.In addition,adding texture features helps to improve the accuracy of the classification results.The final classification accuracy is increased to 77.87%,and the Kappa coefficient is 0.7036.Therefore,as far as the region is concerned,domestic hyperspectral data has certain potential in the field of mangrove species classification,but the mutual influence of spectral resolution and spatial resolution and the short coverage range of wavelength range have become the limiting factors to improve the classification accuracy of this data.The choice of granularity has a very important impact on the classification results.During the process of gradually increasing the granularity from 2 m to 32 m,the inter-species classification precision and KAPPA coefficient of the mangrove forest in the study area both showed an increase and then a decrease,and both When the particle size is 4 m,the best result with an overall accuracy of 78.67% and a KAPPA coefficient of 0.7420 is obtained.The scale effect produced by the landscape pattern index is closely related to the granularity of the category itself.According to the analysis of the classification results,4 m is the optimal granularity for the inter-species classification of mangroves in the region.Moreover,the change trend of landscape pattern index with scale can suggest important turning points in landscape classification.