Research On Forest Dynamic Change Detection Method Based On Sentinel-2

Forest is an indispensable part of the earth's ecology.Humans obtain the material basis for survival and development directly or indirectly from the forest.Forests are of great significance for global climate regulation,biodiversity,water conservation,and carbon sequestration.However,at present,with the rapid development of urban expansion and industrialization leading to the overexploitation and deforestation of forest resources,large-scale forest fires are becoming more frequent,and the forest cover is constantly changing.Studying the forest cover and forest dynamic changes is of great significance to reveal the changes of the forest ecosystem environment and master the law of vegetation restoration.Therefore,it is necessary to monitor the dynamic changes of forest cover.Traditional forestry surveys are time-consuming,have high labor costs,low costs,short periods,and convenient access to remote sensing data,making it easier to detect changes in forest cover.However,there are many sources of remote sensing data,there are various types of vegetation index and change detection methods,and the detection results are uneven.Based on Sentinel-2 remote sensing image,the system systematically analyzes the response status of different vegetation indexes to changes in forest cover changes of different intensities,and analyzes the difference in response sensitivity between the vegetation index with red edge band and the normal vegetation index.For different levels of forest cover changes,the types of change are divided into heavy logging,moderate logging,weak logging,unchanged,slight recovery,moderate recovery and complete recovery,selecting 57 training samples and 42 inspection samples.Select two remote sensing images on April 28,2017 and April 3,2018 to calculate 5 types of non-red edge vegetation indexes NDVI,NBR,GNDVI,SR and DVI,16 types of red edge indexes NDVIre1,NDVIre1 n,NDVIre2,NDVIre2 n,NDVIre3,NDVIre3 n,PSRI,CIre,NDre1,NDre1 m,NDre2,NDre2,SRre1,SRre2,MSRre and MSRren.Spearman correlation analysis was performed on the difference between the forest cover change sample and its vegetation index to calculate the correlation coefficient.The NDVIre1 n is selected as the vegetation index that best responds to changes in forest cover.The selected NDVIre1 n difference,change vector analysis method and principal component analysis method are used to construct bitemporal images,using support vector machine,Mahalanobis distance,neural network,Random forest,maximum likelihood and minimum distance to classify and verify accuracy.The five non-red edge vegetation indices are positively correlated with changes in forest cover.The correlation coefficients of NDVI,NBR,GNDVI,SR,and DVI are 0.886,0.843,0.859,0.841,and 0.855 in sequence,with NDVI having the best response to forest cover changes and SR being the worst.The correlation between the 16 kinds of red edge vegetation index and the change of forest cover is quite different.Among them,the distribution of NDre1 m and NDre2 m in the vegetation calculation is more discrete,and there are more abnormal values,which is not suitable for forest cover changes.Among the red-edged vegetation indexes,NDVIre1 n,NDre1,NDre2,MSRren and NDVIre1 responded better to forest cover changes,and the correlation coefficients were 0.916,0.914,0.913,0.908 and 0.899,respectively.Among the 6 classification methods,the overall classification result of random forest is the best.The overall accuracy based on the vegetation index is 93.8094%,the Kappa coefficient is 0.9209;the overall accuracy based on the change vector analysis is 90.1754%,the Kappa coefficient is 0.8748;It is 94.6757% and the Kappa coefficient is 0.9320.The 10 m resolution image of the Prince's Mountain area has 753,395 pixels.Based on the principal component analysis of random forest classification results,there are 19,784 heavy-cutting type pixels,29,365 moderate-cutting type pixels,216,197 weak-cutting type pixels,357,498 unchangeable class pixels,and 115,197 pixels.Light recovery type cells,13,390 medium recovery type cells,and 1964 full recovery type cells.The red edge vegetation index and the non-red edge vegetation index are different in the study of forest dynamic change detection.Among the 21 plantation indexes calculated in the study,the five species with the best response to forest cover changes are the red edge vegetation index.Among the six methods used in the study,random forest has the best effect,can effectively extract the change information of the forest,and can meet the precision requirements of forestry production and research.