High Resolution Satellite Raster Image Processing with Different Geometrical Data
Abstract
I have developed an innovative knowledge graph architecture that automatically integrates and queries extensive multidimensional (spatial, temporal, and spectral) raster images from satellites with vector geometrical data in real-time, according to the end user's query. This solution liberates the user from the necessity of understanding complex data structures of many heterogeneous data sources (raster and vector), together with associated metadata and domain knowledge. The framework accepts a user question as input, then selecting the necessary data to construct a knowledge network that addresses the inquiry. I am now conducting research on benchmarking systems that handle extensive raster data and various shapes to assess efficiency. Integrating and querying extensive multidimensional raster data involves expensive data transfers from array-based DBMSs, and the integration with various geometrical data according to user queries adversely impacts system performance, rendering it inefficient for execution on local machines (resulting in Out of Memory Errors). I have containerised the complete data integration and query pipeline for various benchmark scenarios that must execute for all potential polygonal and multipolygonal geometries, comprising over 100,000 points with dense raster images containing millions of pixels, utilising resources allocated by Berzelius. This will assist me in uncovering latent patterns and obtaining vital insights to assess the efficacy of my approach.
