| Remote sensing technology can quickly obtain the spatial distribution of forest and make rational planning and application of forest resources.Accurate identification of forest type information is the basis of many forestry work.Using remote sensing technology to identify forest type is the research hotspot of forestry remote sensing.Environment 1 satellite(HJ-1A/1B)has high time resolution and good macro characteristics.By using time series images,it can effectively obtain the growth characteristics of forest in different periods,and there are few phenomena of "same object with different spectrum" and "foreign object with the same spectrum".In this paper,different classification methods are adopted based on the single phase and time series images of The Environment 1 satellite.The identification and classification of major Robinia pseudoacacia L.,Platycladus orientalis(L.)Franco and Quercus variabilis Bl.forests in Nanshan Forest farm of Jiyuan Were carried out.It is expected to provide some reference for the application of HJ-1 satellite in forest resource identification.The main research contents and conclusions are as follows :(1)In the classification of single time phase data,four methods of Maximum Likelihood(MLC),Neural Network(NNC),Support Vector Machine(SVM)and Random Forest(RF)were used to identify the main forest types in Nanshan forest farm of Jiyuan.Among the classification results using only spectral data,the Maximum Likelihood method has the highest classification accuracy of 71.35%,Neural Network method has the highest classification accuracy of 78.38%,Support Vector Machine method has the highest classification accuracy of 75.80%,and Random Forest method has the highest classification accuracy of 75.18%.Among the six time phases,Robinia pseudoacacia forest and Quercus variabilis forest were classified by four methods.The user accuracy of the classification results reached more than 90% in April,and the identification results of forest types in Nanshan forest farm of Jiyuan were better in April.(2)Based on the results of single time phase spectral value classification,three time phases with better classification results were selected.NNC,SVM and RF were used to identify the main forest types of the forest farm by combining the spectral values of the three time phases images with NDVI data.In the classification results of single-phase spectral values combined with NDVI data,the classification accuracy of Neural Network method is 81.73%,Support Vector machine method is 76.88%,and Random Forest method is 74.82%.The results show that the single-phase spectral value combined with NDVI feature data can improve the accuracy of single-phase image classification.(3)In the classification of time series data,three time series are constructed: multi-temporal spectral value time series,NDVI time series and the time series combining spectral value and NDVI.SVM,RF and Dynamic Time Warping(DTW)were used to identify the main forest types.Among the three time series,the overall classification accuracy of Support Vector Machine is 77.08%.The highest classification accuracy was 84.62% by Random Forest method and 88.21% by DTW algorithm.The classification accuracy of forest type identification with time series data is greater than that of spectral value of single time phase image and NDVI combined data.The results show that the classification result of time series data is better than that of single phase data.Therefore,using time series images to identify forest types can effectively avoid the error caused by the problem of "same object and different spectrum",so as to obtain the spatial distribution of forest types more accurately. |
