Research On High Dimensional Spatio Temporal Data Visualization Based On WebGL

With the de-velopment of information technology,the degree of digitization in various fields is gradually increasing,and the amount of data is getting larger.Among various types of data,high dimensional spatio-temporal data contains both time and space information as well as high dimensional data.This kind of data is complicated and contains a lot of valuable information.Data mining and visualization are important methods of data analysis.However,applying these two methods to high dimensional spatio-temporal data can not fully reveal the connections between data points.This paper studies the visualization of high dimensional spatio temporal data.This paper proposes and implements the visualization scheme by combining data mining and data visualization.The main work of this paper includes three aspects:First,in order to present the complete relationship between high dimensional data in three-dimensional space,this paper studies t-SNE algorithm in-depth,and improves the process of calculating the slimilarity of sample points in high-dimensional space.The original t-SNE algorithm directly uses Euclidean distance,the first-order proximity of sample points,to measure the similarity of sample points in high dimensional space.However,the Euclidean distance can not faithfully reflecte similarity on nonlinear manifolds.In this paper,we propose to use the neighborhood structure which lis the second-orderproximity of sample points to measure the similarity.Based on second-order proximity,this paper proposes the Second Order t-SNE,ST-SNE.ST-SNE and t-SNE is compared on multiple data sets such as MNIST,USPS and COIL-20.The results show that ST-SNE can effectively improve the KNN classification accuracy and visualization effect of dimensionality reduction.Secondly,this paper designs and implements a high-dimensional spatio-temporal data visualization system based on WebGL.WebGL directly calls the GPU to draw the graphics which could ensure the system is smooth and reliable in the case of dealing with large data size.The multiple views system,combining with ST-SNE algorithm,could clearly present the relationship between time,space and high-dimensional data.The system is designed with rich interactive functions.Users can manipulate data in various views,and the system could show related information in the remaining views to help users explore the hidden relationships and information more intuitively.The three-dimensional space view introduces the concept of progressive visualization.User could observe the whole process of dimensionality reduction,and could pause the dimension reduction to explore the data at any time.This visualization scheme solves the problem of a long,uncontrollable process of dimensionality reduction visualization.Finally,this paper applies the high dimensional spatio-temporal data visualization system to the air pollution data of each city in the United States from 2000 to 2016.The case study is used to illustrate and demonstrate the effectiveness of the system.