Analysis And Improvement Of Spatial-Temporal Fusion Method Of Remote Sensing Image Based On Weight Filtering

Remote sensing satellite sensors are limited by their own hardware performance,making it difficult for a single sensor to obtain remote sensing images with high spatial details and high time information attributes at the same time,which greatly limits the application range of remote sensing images to the ground.In order to solve the problem of time and space contradiction of sensors,the Spatial-Temporal fusion technology of remote sensing image can combine time-space information with low spatial resolution and high temporal resolution image and high spatial resolution low temporal sensing image.Then,The technology construct high-spatial and high-temporal attributes at the same time.The fusion technology can integrate the spatial detail information and the time feature information on the multi-source remote sensing image,and contributes to solving the problem of time and space contradiction of the sensor.In many spatio-temporal fusion models,the model with weight filtering as the core idea is widely used in high-time and high spatial resolution image generation.The spatio-temporal fusion model based on weight filtering has developed more fusion algorithms for different application purposes.In this paper,the research on spatio-temporal fusion model for weight filtering is carried out.Firstly,the current research status of algorithms at home and abroad is analyzed,and the existing problems of the algorithm are further discussed.In the aspect of ESTARFM similar pixel selection,the improved analysis is carried out,and the ESTARFM_NL algorithm is proposed,which can greatly improve the efficiency of the algorithm when the accuracy is consistent with ESTARFM.Finally,the improved ESTARFM_NL algorithm is applied to the Sentinel-2 and PROBA-V images to generate an NDVI(Normalized Difference Vegetation Index)image sequence with high temporal and spatial resolution.(1)The reference phase selection directly affects the model fusion accuracy.In this paper,the fusion time experiment is carried out by using three sets of reference time data with different phase change conditions,and the accuracy of the fusion result is compared and analyzed.The accuracy of the fusion result is related to the time interval between the reference time and the predicted time,and the low-resolution image is combined with the two times.The correlation coefficient between the two constructs the time phase difference parameter,which can characterize the time difference between the reference time and the predicted time.Based on the experimental results and the time difference parameter,the reference time phase selection strategy is designed to optimize the algorithm accuracy.(2)In the similar pixel selection process of ESTARFM algorithm,the number of classifications needs to be artificially set.In order to reduce the influence of human factors on the performance of the algorithm,this paper improves the similar pixel selection process and proposes the ESTARFM_NL algorithm.Through experimental comparison and analysis,ESTARFM_NL can greatly improve the algorithm operation efficiency while maintaining the accuracy and ESTARFM.This provides a feasible fusion scheme for dealing with large-area or long-term sequence remote sensing data.(3)Based on the reference time selection strategy,this paper conducts a spatio-temporal fusion experiment using Sentinel-2 and PROBA-V data as data sources.Using the ESTARFM_NL algorithm to generate high-time-frequency NDVI image products with high spatial detail,verify the applicability of ESTARFM NL for spatio-temporal fusion of other data sources.