A Novel Vector Field Data Mining Approach: Extraction of Front Based on Physical Features of Target

The explosive growing of earth observing data needs to have relative efficient data processing methods. This paper aims at the processing and analysis of large volume of vector field data acquiring from satellite derived, or model assimilation, an approach of fronts extraction from vector field data was proposed. The study is based on the assumption that the distribution of feature vectors for front and non-front are significantly different. A front is represented as a multi-dimensional feature space composed by a set of physical features acts as feature vectors. On the basis of the characteristics analysis of vector field data and the physical feature of fronts, a five-step front extraction process was illuminated in detail, including physical feature abstraction of target, physical features spatialization and multi-dimensional feature space construction, feature space segmentation, identification of possible front and postprocessing. By expert knowledge, speed, vorticity, and the direction variation of vector field were chosen as feature vectors, Fuzzy Clustering method is used to segment the feature space into several regions, and get the possible front. For the filtering of the candidate regions and the identification of possible front, principal components analysis and the domain knowledge were used, followed by a hierarchical threshold technique and other post processing techniques for the removing of false regions. To illuminate the application of the approach, the extraction of ocean front from ocean current field data was taken as an example. Experimental results show that the extracted fronts are in good agreement with the ones identified by Sea Surface Temperature (SST) image. Furthermore, the approach is universal for all kinds of front extraction from vector field data, including the extraction of air front system from wind field. *Corresponding author