| In recent years,the free availability of satellite data has provided conditions for the development of time-series surface coverage products,and time-series surface coverage products have been rapidly developed.However,the traditional classification strategy rarely considers temporal continuity and spatial consistency,resulting in pseudo-variation among multi-period products in addition to real changes in the ground surface,making unreasonable variations between different periods of the same product common.Accurate and consistent feature information is the key to the availability of surface coverage products,and the existence of this phenomenon has affected the practical application of time-series surface coverage products to a certain extent.The reasons for the existence of pseudo-variation can be divided into two types:classification error and offset between temporal ground coverage products.To address the above reasons,this paper carries out the research of consistency improvement model of temporal ground cover products,determines two commonly used consistency adjustment models,spatio-temporal window filtering model and hidden Markov model(HMM),and implements the models,among which the HMM model(hidden Markov model)has the problems of relying on expert The HMM model(hidden Markov model)has the problems of relying on expert experience and lack of spatial relations.To address the two problems of HMM model,this paper optimizes the model and proposes HMM＿LCT(Land Cover Transition,LCT)model.The main research contents are as follows:1.To address the problem of relying on expert experience in the HMM model,the HMM＿LCT model is solved by introducing the land cover transfer probability matrix(LCT)method.The land cover transfer probability matrix is calculated by the two phases of Globe Land30 products to obtain the natural law of feature transfer in the study area and apply it to the state transfer probability matrix in the HMM model.Unlike traditional methods that rely on expert knowledge,this paper uses existing products for statistics,which are more relevant to the actual situation in the study area,while reducing the influence of subjective factors.2.For the problem of lack of spatial relationship in the HMM model,the HMM＿LCT model introduces a spatial factor.In this paper,based on the idea of Local Binary Pattern(LBP)algorithm,the occurrence probability of each category is calculated separately in a 3×3 window,and combined with the observation probability to form a new observation probability matrix to increase the spatial aggregation of classification results.3.The validity and stability tests of the HMM＿LCT model were conducted.In this paper,the three models were used to adjust the time-series land surface classification results in Wuhan Huangpi District of Phase III Sentinel 2 as the study area,and the results were compared and analyzed.Firstly,accuracy verification and image local comparison are performed,and the results show that the HMM＿LCT model performs better than the remaining two models;Then the effect of the HMM＿LCT model was analyzed in terms of land cover trajectory(stability of feature class change)and feature offset(spatial consistency)adjustment,and the results showed that the HMM＿LCT model effectively improved the consistency of time-series land cover products;The HMM＿LCT model proposed in this paper can improve the overall accuracy of the time-series ground cover product by about 2-4%;meanwhile,the land cover trajectory is adjusted to better express the change trend of features;in addition,it also has the effect of improving the feature offset phenomenon,effectively enhancing the consistency of the timeseries ground cover product in terms of time and space,and enhancing the usability of the product.Finally,the robustness of the HMM＿LCT model is tested by selecting the remote sensing images of Beijing Changping District in the third phase of HMS-1,and the results show that the model has good stability. |
PDF Full Text Request