Study Of Rice Identification And Monitoring Based On Spatial And Temporal Fusion Model In Hilly Area Of South China

Rice is one of the largest and most important crops in the South.How to use effective means to develop crop distribution based on the fragmentation of geographical patches of land in the South Hill Basin is an important research.The rapid and precise characteristics of remote sensing technology take great advantages in extracting crop distribution.However,it is difficult to achieve both temporal resolution and spatial resolution of remote sensing data at the present stage,and it is impossible to effectively monitor rice distribution.Based on this background,the study used two spatial-temporal fusion models to fuse the MODIS and GF-1WFV data into a spatial-temporal data set,and used the multiphase images after fused to realize the monitoring of the change of rice planting area in the hilly area of the South.This article takes Hengyang County of Hunan Province as the research area,using MODIS and GF-1WFV data of 2015 and 2016,and constructs two different spatial-temporal data sets based on STARFM and ESTARFM.Then it compares the extracted data set with the actual value to obtain the optimal value.The solution is to use the fitting method to extract the NDVI fitting curves and phenology parameters,then extract the rice planting area for two years based on the decision tree,analyze the change area and get the following conclusions:(1)The fusion of MODIS and GF-1WFV data can be achieved based on both STARFM and ESTARFM methods,but the results of the two are quite different.The correlation between the fusion results of 2015?2016 and the true value was analyzed.The correlation between the two-year fusion result and the true value extracted by the ESTARFM method was higher than that of the STARFM method.Mainly in STARFM accuracy of the method is about 60%,while ESTARFM fusion accuracy can reach 80%.However,the method of ESTARFM is not as efficient as STARFM in terms of processing efficiency.The former takes 1h to extract a scene image,and the latter requires 10min.(2)This study discusses the existing methods of D-L,A-G fitting and S-G filtering,all of them can smooth the NDVI time-series data,but the results of them are quite different.Compared with the other two method,A-G method can better reflect the crop characteristics of the study area,and the fitting result is also better than the other two methods.(3)Extracting rice planting in the study area in 2015 and 2016 based on the decision tree method.The overall precision in the two years was above 85%,and the Kappa coefficient was also above 75%.The classification effects are good.(4)Comparing the results of paddy distribution in two years,the change of paddy area was extracted.The results showed that the paddy area in 2016 increased by 22.18km~2compared with the paddy area in 2015.The result is close to the actual statistical data of9.8km~2 in Hengyang County.