A data-driven framework for analyzing science–technology interaction in agricultural AI: A case study of plant image recognition

• A dual publication–patent framework was developed to examine agricultural AI plant image recognition. • BERTopic and TLI were integrated to quantify topic evolution and science–technology temporal alignment. • Publication and patent networks showed different structural patterns of cohesion and fragmentation. • Scenario simulations identified how data sharing, transformers, and sensing technologies may reshape innovation trajectories. • The findings reveal a shift from delayed technology transfer toward more coupled and anticipatory innovation dynamics. Technological innovation is often associated with prior scientific development, but the temporal and structural relationships between these domains remain difficult to characterize in data-intensive fields such as agricultural artificial intelligence. Plant image recognition provides a representative context in which scientific research and technological application evolve in parallel. This study develops an integrated framework to examine the interaction between scientific publications and patent activity through semantic alignment and structural analysis. A Translational Lag Index is introduced to quantify the temporal relationship between research emergence and technological activity, enabling comparison across topics and over time. Analysis of 4,063 publications and 308 patents suggests that translational lag has decreased over time and, in some cases, approaches or slightly reverses, indicating closer temporal alignment between scientific and technological development. At the same time, scientific knowledge systems appear more densely interconnected, whereas technological systems tend to exhibit more modular structures, consistent with observed network characteristics. Further results indicate that different drivers of change may operate unevenly across domains. Expansion of scientific output is associated with broader changes in the research landscape, while advances in algorithms and sensing technologies are more strongly reflected in technological activity. In contrast, variations in knowledge transfer efficiency appear to influence the timing of innovation more than its structural configuration. Overall, the findings suggest that innovation in agricultural artificial intelligence is shaped by overlapping temporal dynamics, differentiated structural organization, and domain-specific influences. The framework enables systematic analysis of science–technology interaction and supports an empirically grounded understanding of knowledge translation.