Research On Spatial-temporal Fusion Algorithm Of Remote Sensing Image Based On Multi-source Sensor

As a method of acquiring spatial data,remote sensing is an important part of modern high-tech fields.With the development of sensor technology,we can also obtain more and more types of remote sensing images.Different sensors have their own unique advantages and disadvantages.In order to integrate the advantages of multiple sensors,we need to fuse remote sensing images from different sensors.The spatiotemporal fusion for remote sensing images is one of the research hotspots.The key to spatiotemporal fusion is to obtain sufficient spatial information from high spatial resolution images while obtaining sufficient temporal information from high temporal resolution images.Existing algorithms face severe challenges when the land cover changes due to the limitations of the model itself.To this end,this paper has carried out the research of spatiotemporal fusion algorithm based on multi-source sensors.Based on previous research,this paper proposes two new spatiotemporal fusion algorithms by summarizing the advantages and disadvantages of existing models.The main work of this paper is organized as follows.First of all,in view of the existing methods where only a pair of coarse-fine images are used as input and the prediction accuracy of land cover changes is too low,the first content of this paper proposes a spatiotemporal fusion algorithm based on linear model.The algorithm uses a linear relationship to represent the temporal model between images obtain from different date,analyzes the objective law of temporal change,and constrains the model from both local and global aspects according to the characteristics of temporal change,so that the solved model can represent the temporal change more accurately.In addition,a multi-temporal similar pixel search strategy is introduced,which makes the algorithm more flexible in searching for similar pixels,which can eliminate the module effect of previous spatio-temporal fusion algorithm,thereby improving the accuracy of prediction results.Next,the second research content of this paper is a spatiotemporal fusion algorithm based on deep transfer learning.This method uses deep neural network,and more uses the statistical characteristics to predict the target fine image.This method uses neural network to represent the mapping relationship between images with different spatial resolutions,while estimating the blur kernel between images obtained by different sensors to construct feature space of the prediction task,and transfer completed model to the feature space to improve accuracy of model in predicting the target fine image.Finally,this paper conducted experiments on multiple sensor data sets,and verified the effectiveness of this algorithm through a large number of visual comparisons and quantitative comparisons.