Research On Larch Forest Recognition Based On Multi-source Multi-temporal Data

Larch is one of the most common trees in north China.It has many advantages such as excellent wood material and anti-compression and anti-corrosion,and has high social value and economic benefit.In recent years,larch,as one of the key tree species in the construction of forestry key projects in China,has been increasing the afforestation area year by year.At present,the forestry department pays more and more attention to the investigation of the resources of Larch.In this study,multi-temporal Landsa8 and GF-1 images were used as the main data sources,and on the basis of analyzing the seasonal and spectral characteristics of Larch,the key period and parameters of Larch identification were determined.At the same time,a variety of feature information was extracted,and through the selection of different classifiers,the optimal feature factor and the classification strategy of multi-feature optimization were explored to extract Larch,so as to provide a technical reference for quickly obtaining the spatial location distribution of Larch.Specific research contents and results are as follows:(1)In order to explore the key phenological period of extraction of Larch,Mengjiagang forest farm in Heilongjiang province was taken as the key research area in the experiment.The Maximum Likelihood method,Random Forest method and Support Vector Machine method were used to classify forest types of Landsat8 images of different phases,and the optimal period and method for remote sensing extraction of Lwere selected.The results Larch show that the Maximum Likelihood method has the best comprehensive effect,followed by the Random Forest method.Combined with the NDVI time series curve of Larch,it was found that there were significant phenological differences between Larch and other vegetation in the growth and deciduous period,and the classification effect was relatively ideal,with the classification accuracy higher than 80%.Among them,the image classification effect on November 2 was the best,with the highest accuracy of 85.08%,Kappa=0.81.This conclusion was further verified in the experiment of Dagujia Forest Farm.(2)Compared with the single phase,long phase image contain more abundant features information,the experiment on the basis of above-mentioned single phase image classification,screening classification accuracy is higher than 80%of the images,these images to the permutation and combination,built 69 combined scene for remote sensing image,finally,the Maximum Likelihood method is employed to classify forest type.The results show that:?Compared with the results of single phase classification,multi-phase combination can effectively improve the classification accuracy and is more conducive to the extraction of Larch.?In the multi-temporal combination classification,the image classification effect of two time phase combinations is the best.?The image combination of Larch growth period and deciduous period can reflect the unique phenological law of Larch,and is the best time combination for larch identification,which is conducive to the high-precision extraction of larch.Among the results of multi-temporal combination classification,the images of larch growing period and deciduous period had better classification effect and higher accuracy than 85%.Among them,the images of March 22(growing period)and October 29(deciduous period)had the best classification effect,and the highest classification accuracy was 87.46%,Kappa=0.84.The above conclusions were further verified in the experiment of Dagujia forest farm.(3)GF-1 and Landsat8 images with similar time were selected as data sources in the experiment,and time series features of different vegetation indexes,texture features under different Windows and different terrain factors were extracted by DEM data.The Maximum Likelihood method and Random Forest method were used for classification.The results show that:? The classification accuracy is improved after the addition of feature factors,indicating that the feature factors are helpful to the extraction of Larch.?With the addition of vegetation index timing features,the classification effect is the most ideal and the classification accuracy is significantly improved.It shows that the time series feature of vegetation index can describe the phenological rhythm of well,reflect the seasonal phase feature of Larch,and is the best feature factor to extract Larch.(4)In the experiment,the state forest farm in da hinggan mountains of Heilongjiang province was taken as the whole research area,and the forest type classification was carried out on the MODIS-NDVI time series data set and MODIS-EVI time series data set by using Maximum Likelihood method,Random Forest method and Support Vector Machine method.The results show that the Maximum Likelihood method and the Random Forest method have better classification effect,and the classification accuracy is more than 70%and the area accuracy is more than 85%.Among them,the Random Forest method has the best classification effect and the highest accuracy.The classification accuracy of MODIS-NDVI time series data set is 79.91%,Kappa=0.77,and the classification accuracy of MODIS-EVI time series data set is 78.27%,Kappa=0.76.It indicates that the MODIS time series data set can reflect the seasonal characteristics of Larch and is suitable for the extraction of Larch forests in large areas.