AUTHOR=Zhu Li , Tang Long , Ren Shan 

TITLE=Deep learning-based text generation for plant phenotyping and precision agriculture

JOURNAL=Frontiers in Plant Science

VOLUME=Volume 16 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/plant-science/articles/10.3389/fpls.2025.1564394

DOI=10.3389/fpls.2025.1564394

ISSN=1664-462X

ABSTRACT=IntroductionPlant phenotyping is a critical area in agricultural research that focuses on assessing plant traits quantitatively to enhance productivity and sustainability. While traditional methods remain important, they are constrained by the complexity of plant structures, variability in environmental conditions, and the need for high-throughput analysis. Recent advances in imaging technologies and machine learning offer new possibilities, but current methods still face challenges such as noise, occlusion, and limited interpretability.MethodsIn response to these challenges, we propose a novel computational framework that combines deep learning-based text generation with domain-specific knowledge for plant phenotyping. Our approach incorporates three key elements. A hybrid generative model is used to capture complex spatial and temporal phenotypic patterns. A biologically-constrained optimization strategy is employed to improve both prediction accuracy and interpretability. An environment-aware module is included to address environmental variability.ResultsThe generative model uses advanced deep learning techniques to process high-dimensional imaging data, effectively capturing complex plant traits while overcoming issues like occlusion and variability. The biologically-constrained optimization strategy incorporates prior biological knowledge into the computational process, ensuring predictions are biologically realistic and enhancing trait correlations and structural consistency. The environment-aware module adapts dynamically to environmental factors, ensuring reliable predictions across a variety of agricultural settings.DiscussionExperimental results show that the framework delivers scalable, interpretable, and accurate phenotyping solutions, setting a new standard for precision agriculture applications.