PARSUC: A Parallel Subsampling-Based Method for Clustering Remote Sensing Big Data.

Area of Science:

• Geospatial analysis
• Data mining
• Computational science

Background:

• Remote sensing big data (RSBD) presents significant computational challenges due to its volume, diversity, and high dimensionality.
• Conventional clustering algorithms struggle with the scale and complexity of RSBD, leading to inefficiency and reduced accuracy.
• Existing parallel clustering algorithms often suffer from repeated calculations, creating performance bottlenecks.

Purpose of the Study:

• To develop a novel parallel clustering method, PARSUC, to address the computational challenges of RSBD analysis.
• To improve both the efficiency and accuracy of clustering large-scale remote sensing imagery.
• To overcome the performance limitations of existing parallel clustering techniques.

Main Methods:

• Proposed a parallel subsampling-based clustering (PARSUC) method utilizing a novel subsampling-based data partitioning (SubDP) technique.
• Implemented a three-step parallel clustering approach to minimize redundant computations.
• Introduced a centroid filtering algorithm (CFA) to mitigate subsampling errors and ensure result accuracy.
• Leveraged the Hadoop platform and MapReduce parallel model for implementation.

Main Results:

• PARSUC demonstrated significantly higher accuracy compared to conventional algorithms on large remote sensing datasets.
• The method exhibited notable scalability, with performance improving as more computing nodes were added.
• PARSUC achieved substantial time savings compared to existing parallel clustering algorithms when processing RSBD.

Conclusions:

• PARSUC offers an effective and efficient solution for clustering remote sensing big data.
• The proposed method overcomes key performance bottlenecks in parallel clustering for large-scale geospatial imagery.
• PARSUC provides a scalable and accurate approach for extracting valuable information from RSBD.