From Prototype to Inference: A Pipeline to Apply Deep Learning in Sorghum Panicle Detection

Area of Science:

• Agricultural Science
• Computer Science
• Machine Learning

Background:

• Panicle density is crucial for crop yield estimation in species like sorghum and wheat.
• Manual counting of panicle density is labor-intensive and inefficient for large-scale crop assessment.
• Existing machine learning approaches often lack generalizable protocols for practical deployment.

Purpose of the Study:

• To establish a comprehensive pipeline for deep-learning-assisted panicle yield estimation in sorghum.
• To provide a generalized protocol for data collection, model training, validation, and deployment.
• To enable accurate, automated head density mapping for agronomic variability diagnosis.

Main Methods:

• Development of a complete pipeline from data acquisition to model deployment for deep learning-based panicle counting.
• Focus on robust model training to address domain shift issues common in natural field environments.
• Demonstration of the pipeline's application in sorghum fields, with potential for generalization to other grain crops.

Main Results:

• Successful implementation of a deep-learning pipeline for automated panicle density estimation.
• Generation of high-resolution head density maps for diagnosing within-field agronomic variability.
• The pipeline operates without reliance on commercial software, promoting accessibility.

Conclusions:

• The developed pipeline offers an efficient and accurate alternative to manual panicle counting for crop yield estimation.
• The methodology is robust and adaptable for deployment in commercial agricultural settings.
• This approach provides a valuable tool for precision agriculture and crop improvement programs.