| Forest category is one of the important basic data of forestry resource management and monitoring,the important basis of forestry construction of various decision,forest category information accurately and rapidly has been the research focus of forestry workers.Time series images can reflect the vegetation phenology information,it can help to increase vegetation clustering accuracy,especially for single phase images on the growth characteristics of similar species effect is particularly obvious,using time phase or in the time series of remote sensing image reflects the vegetation classification,the characteristics of vegetation time is always the important research direction of vegetation remote sensing classification and research hot spot.Time series data of ground objects using remote sensing image classification is the key to the two sequence similarity analysis,Dynamic Time neat algorithm(DTW,Dynamic Time Warping)to solve the matching problem of long Time series,and can resist noise such as cloud caused by outliers in Time series,the similar feature matching so as to achieve better effect.This article is based on six image data time series GF-1 respectively by using DTW and Euclidean distance(ED,Euclidean Distance)for time series similarity measure,respectively use DTW distance and distance ED to K-Means algorithm of image in the stand of clustering,preliminary forest types of different clustering results are obtained,the final preliminary clustering results are wrong Salt and pepper Pixels,identification,and then using the method of weighted K neighbor to forest types,complete the image classification reprocessing,finally verifies the accuracy of the result of the post-processing of the classification and comparison.Research results have shown:1)Eight GF-1 image of tree species have distinguish differences on the blue,green,red and near-infrared band and NDVI time series,the visible light band(blue green red)is responsible for the classification of tree species mainly in the winter to the early stages of the growth season(form February to May).And near infrared band on the contrary,the time is the main contribution of the tree species classification in the growing season(from June to September),winter basic cannot serve as classification characteristics,description of remote sensing image GF-1 time series for the feasibility of forest types classification2)Based on CD-K Means clustering the time series of remote sensing image tree species as a result,the overall classification accuracy was 90.84%,the Kappa coefficient is 0.88.Based on DTW-k Means distance clustering the time series of remote sensing image tree species as a result,the overall classification accuracy was 92.21%,the Kappa coefficient of 0.90.DTW-K Means method can achieve good results for tree species classification of time series remote sensing images,and is superior to CD-K Means method. |
