Temporal super-resolution visualization method for time-varying data based on deep learning model

Abstract

The invention relates to the technical field of data processing and modeling. The present invention proposes a temporal super-resolution visualization method for time-varying data based on a deep learning model. First, key voxels are extracted based on a gradient histogram, and the data scale is reduced while retaining the spatial characteristics of the original data. Next, train a multi-scale variational autoencoder to obtain an encoder with feature extraction function and a decoder with volume data generation function, so that the generation problem of time-varying data sequence can be converted into the generation of hidden variable sequence Therefore, volume data can be processed and generated in a low-dimensional feature space that is more concise and can express the potential information of volume data. In the latent space, two ideas are proposed to fit the temporal relationship between the latent variables of each time step data: one method is to directly project the data into two dimensions, and interpolate the latent variables of the new data on the basis of the overall temporal development trend of the data. variable, and further obtain the corresponding volume data sequence.

Description

technical field [0001] The invention belongs to the technical field of data processing and modeling, and in particular relates to a time-varying data time super-resolution visualization method based on a deep learning model. Background technique [0002] Collective simulation data is large-scale time-varying multivariate volume data generated by running scientific models with different parameter combinations. By studying the multivariate data sequences of different collection members, scientists can explore and discover the development laws of various scientific phenomena, as well as special events in the development of data time series. In the analysis of aggregated simulation data, the more time-varying data that can be used for analysis, the more accurate and richer the information of the dynamic spatiotemporal characteristics of the data can be obtained, which will have a very high demand for data storage space. However, ensemble simulation data has the characteristics ...

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to provide a time-varying volumetric data time super-resolution visualization method based on a deep learning model, which is used to solve the technical problems in the above-mentioned prior art, such as: the realization of time-varying multivariate volumetric data in research at the present stage The limitation of the time super-resolution method is that the development of the data over time is regarded as a linear change, which causes a large distortion of the results generated during the period of severe data changes; and the linear interpolation method only focuses on the local time-varying data. information, ignoring the overall time-varying trend of the data

Images