Development of a Low-Cost Distributed Computing Pipeline for High-Throughput Cotton Phenotyping

Area of Science:

• Agricultural Science
• Computer Science
• Data Science

Background:

• High-throughput phenotyping is crucial for crop improvement.
• Existing methods can be costly and computationally intensive.
• Distributed computing offers a scalable solution for agricultural data analysis.

Purpose of the Study:

• To develop a low-cost, distributed computing pipeline for cotton plant phenotyping.
• To leverage Raspberry Pi, Hadoop, and deep learning for efficient image analysis.
• To enable high-throughput phenotyping in field-based agriculture.

Main Methods:

• Utilized a primary-replica distributed architecture with Raspberry Pis and Apache Hadoop.
• Employed a pre-trained Tiny-YOLOv4 model for cotton bloom detection.
• Implemented distributed file system for robust data access and parallel processing.
• Evaluated cluster performance with varying node configurations.

Main Results:

• Successfully developed and implemented a functional distributed computing pipeline.
• Demonstrated parallel processing of cotton image data from pre-processing to bloom detection.
• Provided performance comparisons of four-node vs. centralized and smaller clusters.
• Achieved efficient spatio-temporal map creation for cotton phenotyping.

Conclusions:

• The developed pipeline offers a cost-effective solution for high-throughput cotton phenotyping.
• This work pioneers distributed computing applications in field-based agricultural research.
• The system facilitates robust and scalable analysis of agricultural imagery.