| With the rapid development of informatization, the scale of the data becomes bigger and biggerand the form of data towards high-dimensional, multi-source and polymorphism direction. Now theworld is under a state of data explosion, the growth of data has far exceeded than any time of humanhistory, how to effectively storage and analysis these large scale and diversity data have beenbecome a severe challenge.For compress and analysis of the data, this paper studies the multivariate volume datacompression and trajectory data anomaly detection visualization, which involve the machinelearning, hash table, perfect hash function, anomaly detection and visual analysis technology. In thispaper, the main research work is as follows:At first, we propose a nearly lossless compression solution for multivariate volume data in thispaper. The compression process has three steps: First of all, using active learning selectrepresentative elements in the original data and reduce the original data’s dimension with MDS, andthen using the semi-supervised learning reconstruct the volume data, this process called lossycompression. The second, saving the larger error value which deviating from the original data in thelossy compression, then use perfect hash function establish a hash table for the saved error value. Atlast, we combine the result of the lossy compression with hash table of error value implement nearlylossless compression.And then, we design a system which can detect the anomaly trajectory data. Firstly, at the stageof detecting the anomaly trajectory, trajectory visualization analysis use an anomaly detectionmethod which based on kernel density estimation, and take the local characteristics of the anomalytrajectory data in full consideration; then at the stage of drawing trajectory lines, we usedepth-dependent halos algorithm render the dense line data to solving visual confusion phenomenoncaused by large track lines, effectively enhance the effect of visualization analysis. |
