Leveraging High-Throughput Phenotyping and Artificial Intelligence for Advancing Sorghum Improvement

Authors: BAKSHI, ALIVA - Kansas State University; PRAMANIK, SWARAJ - Kansas State University; WANG, CHAOXIN - Kansas State University; ARANJO, JOSEF - Kansas State University; BANGARI, MAYANK - Texas Tech University; KUMAR, RITESH - Texas Tech University; SOMAYANDA, IMPA - Texas Tech University; SAINI, D - Texas Tech University; CARAGEA, DOINA - Texas Tech University; Bean, Scott; Hayes, Chad; Emendack, Yves; JAGADISH, KRISHNA - Texas Tech University

Submitted to: Designing Crops For Added Value
Publication Type: Book / Chapter
Publication Acceptance Date: 4/4/2025
Publication Date: 9/30/2025
Citation: Bakshi, A., Pramanik, S., Wang, C., Aranjo, J., Bangari, M., Kumar, R., Somayanda, I., Saini, D., Caragea, D., Bean, S.R., Hayes, C.M., Emendack, Y., Jagadish, K. 2025. Leveraging High-Throughput Phenotyping and Artificial Intelligence for Advancing Sorghum Improvement. Designing Crops For Added Value. https://doi.org/10.1201/9781003623717.
DOI: https://doi.org/10.1201/9781003623717

Interpretive Summary:

Technical Abstract: Sorghum is a versatile crop which is essential for food security, biofuel production, and animal feed. Despite advances in sorghum genotyping, improving sorghum yield, as well as its adaptability to changing environmental conditions is still a challenge. High-throughput plant phenotyping (HTPP) integrated with artificial intelligence (AI) could revolutionize sorghum breeding and advance its improvement. AI and specifically deep learning models can process and analyze large amounts of data to extract feature representations and identify useful predictive patterns. Such properties of deep learning have been leveraged to design robust models for high- throughput plant phenotyping of various crops, including sorghum, for higher yield, weather tolerance and disease resistance. The trait assessment performed with deep learning models can enable sorghum breeders to accelerate the selection of superior genotypes and establish practices that ensure better productivity. In this context, this chapter focuses on HTPP for sorghum breeding purposes using deep learning object detection and segmentation models trained and evaluated for RGB images. We review several object detection and segmentation models that have been commonly used for assessing selection traits in sorghum, describe the generic workflow for training and evaluating such models, and discuss the use of deep learning models for three important selection traits for sorghum breeding, panicle density and grain density (key yield-related traits), and stomata density (relevant in terms of abiotic stress response). We conclude the chapter with a discussion of limitations of current HTTP approaches and ideas for future work. Such ideas will further contribute to the integration of HTPP and AI, integration which offer a promising way ahead for boosting sorghum productivity in light of climate changes and for promoting global food security initiatives.