研究目的
To present a MapReduce based approach for parallelizing classification algorithms of remote sensing images on the cloud computing platform, specifically using the K-Means clustering algorithm to demonstrate the approach's applicability and effectiveness.
研究成果
The approach is effective, and the more computing intensity of the algorithm has, the better performance the approach has. Future work will study and take more classification algorithms of big remote sensing data as examples to improve the approach.
研究不足
The paper only tries the simple K-Means clustering algorithm, so the applicability of the approach still needs to be tested and enhanced with more classification algorithms of big remote sensing data.
1：Experimental Design and Method Selection:
The paper presents a MapReduce based approach for parallelizing classification algorithms of remote sensing images on the cloud computing platform. The iterative processing is transformed into iterative Map and Reduce tasks that can be executed in parallel.
2：Sample Selection and Data Sources:
The testing remote sensing image is a 7 bands TM data with 5244 rows and 5205 columns.
3：List of Experimental Equipment and Materials:
The experimental environment includes OpenStack Newton, Centos7 X86_64, Spark2.1.0, a cluster with 1 Master and 4 Workers, each with 16 Cores and 32GB ROM.
4：0, a cluster with 1 Master and 4 Workers, each with 16 Cores and 32GB ROM.
Experimental Procedures and Operational Workflow:
4. Experimental Procedures and Operational Workflow: The paralleled K-Means model is built and implemented with the SPARK cluster deployed on the OpenStack cloud computing platform. The classifier is broadcast to each computing node, Map tasks assign each pixel in the data partition to the nearest center point in parallel, and Reduce tasks collect all clustering results of Map tasks to update the center points in an iteration step.
5：Data Analysis Methods:
The processing times of the normal K-Means model and the paralleled K-Means model are recorded and compared with different parameters.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
