Design And Implementation Of Spatial Interpolation Method And System For Scattered Data Based On Attention Mechanism

Spatial interpolation is a statistical method for estimating the values of unknown points using the values of known points.In many applications of spatial interpolation,such as environmental monitoring,the location distribution of measurement sites is usually irregular and uneven,that is,the known point data is a kind of scattered data.The irregularity and spatial complexity of scattered data greatly challenge traditional spatial interpolation methods.To this end,this paper designs a spatial interpolation method for scattered data based on the attention mechanism and implements a spatial interpolation system.The main work is as follows:1.A spatial interpolation method for scattered data based on the attention mechanism is designed.Firstly,according to the irregular distribution and disorder of scattered data,a preprocessing method based on a masking mechanism is proposed,which transforms the scattered data into a regularly ordered matrix by gridding the data plane.Then,an encoder-decoder two-stage network structure based on the attention mechanism is designed.The encoder uses the attention mechanism to extract features from the available data to reduce the interference caused by sparse data.The decoder uses the hybrid structure of a convolutional network and attention mechanism to extract local and global features,and complete the spatial interpolation generation task in the global scope.2.A spatial interpolation system for scattered data based on the spatial interpolation method proposed in this paper is designed and implemented.The system includes data,processing,and user interaction layers.The data layer provides the data warehouse for the system.The scattered data interpolation method based on the attention mechanism was embedded in the processing layer.The user interaction layer offers an interactive platform for easy operation.The system realizes the functions of spatial interpolation model training,model loading and spatial interpolation of scattered data.Experiments on the GDEM show that,compared with the best benchmark model,the mean square error(MSE)of the prediction results is reduced by 0.12%,and the goodness of fit(R2 score)of the model is increased by 2.2%.