Research On Quadtree Grid Based Fast Hierarchical Clustering Heatmap Visualization

Maps, as the most important presentation of the information that carried by data, have been widely used in many areas. The heat map is a new-type information visualization method that goes beyond the traditional graphics. In the modern visualization, heat maps are more widely used. However, with the advent of the era of big data, not only did the multi-element and multi-dimensional properties of data increase sharply, but also the data scale. This made the efficiency of current methods of heat map visualization for large-scale data greatly reduced, even exceeding the machine workload, causing the collapse of programs and failure of implementing data visualization. To solve this problem,we proposed a fast hierarchical clustering heatmap visualization method based on quadtree grid in this study. This method required to clustering-preprocess the big data before using visualization technique to present the data by heat map.Firstly, a fast hierarchical clustering heatmap visualization method based on quadtree grid was proposed. An appropriate clustering method was needed to preprocess the large-scale data to reduce the data length by order of magnitude, then analyzing the current clustering methods and integrating the advantages and disadvantages of these methods. This algorithm is suitable for the clustering of large-scale data, avoiding data loss caused by multiple clustering. The clustered data has a high quality and the properties of the original data remain to the maximum extent.Secondly, visualization technique was applied to calculate the heat value for the heat map. Taking the preprocessed data center as the heatpoint, Gauss function was used to fit the effect-value that each heatpoint acted on its surrounding areas, then accumulating all the effect values to get the ultimate heat value. This heat value was also taken as the volume data to three-dimensional heat map with visualization technique.Thirdly, according to the features of the three-dimensional volume data, a simplified ray casting method was proposed. This method made use of GPU parallel processing capability and the programmable feature of the shader, directly calculate the longitude and latitude heat values in the shader. Then converting the heat values to height values proportionally, and assign color values to draw three-dimensional heat maps. Comparedwith conventional ray casting algorithm, this method reduced the calculation of intersection test and repeated sampling, and increased the frame rate of drawing heat maps.Finally, the visual application framework was designed and implemented, and the proposed methods were validated and analyzed by experiments. The results demonstrated the feasibility and effectiveness of the proposed methods.