GPU-Based Parallel Processing Templates for Remote Sensing Images

This report delves into the challenges of remote sensing image processing, characterized by massive data, intensive computation, and complex algorithms, which hinder rapid processing. It explores the application of general-purpose graphic processing units (GPGPU) to overcome these limitations. The study focuses on designing reusable GPU-based remote sensing image parallel processing models and establishing parallel programming templates to simplify the development of parallel algorithms. The research examines related work in GPU computing, programming libraries, and the advantages of GPUs over clusters. It then categorizes remote sensing image processing algorithms into point, local, and global operations, providing a systematic perspective. The core of the paper presents the design and implementation of parallel programming templates for GPU architectures, validated through experiments and performance discussions. The templates aim to provide a more straightforward and effective method for programming parallel remote sensing image processing algorithms, addressing the current limitations of GPU utilization in this field. The research emphasizes the importance of efficient memory management and data transfer between CPU and GPU to fully exploit the parallel organization of GPU architectures.