Triticale field phenotyping using RGB camera for ear counting and yield estimation.

Area of Science:

• Agronomy
• Plant Science
• Genetics

Background:

• Triticale (X Triticosecale Wittmack) is a wheat-rye hybrid with high adaptability to environmental stresses.
• Its cultivation is promising for agriculture in a changing climate.
• Efficient phenotyping is crucial for triticale breeding and crop improvement.

Purpose of the Study:

• To describe RGB on-ground phenotyping of triticale cultivars.
• To evaluate the accuracy of an object detection algorithm for counting triticale ears.
• To explore the correlation between ear count and yield in triticale.

Main Methods:

• Field-based RGB imaging of eighteen triticale cultivars over two growing seasons (2018-2019, 2019-2020).
• Application of the YOLOv4 object detection algorithm with ensemble modeling for ear counting.
• Analysis of image data captured under naturally varying light conditions.

Main Results:

• The YOLOv4 algorithm achieved 95% accuracy and a mean average precision (mAP) of 0.71 in counting triticale ears.
• A statistically significant correlation (p=0.16) was found between the number of ears and yield in the 2019 data.
• The study demonstrates the feasibility of automated phenotyping for triticale.

Conclusions:

• Advanced object detection algorithms like YOLOv4 can effectively phenotype triticale in field conditions.
• Accurate ear counting via phenotyping can contribute to understanding yield determinants in triticale.
• This approach addresses bottlenecks in modern plant breeding and phenotyping.