A Unified Transformer Approach for Granular Crop and Nutrient Prescription Using ViT & LSTM

View at publisher Summarise with VIRAT
Authors: Ezhumalai. P, Jayashree. K, Keerthana M, Keerthiga Devi S D, Sharmila V
Cited by: 0
References: 14

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Progress in high-precision agriculture is challenged less by data volume and more by the inability to unify data collected from multiple types of sensors. Spatial and temporal dimensions can often cause misalignment in data streams, including soil composition, changing climatic factors, and satellite-derived vegetative indices. This pervasive data heterogeneity makes the delivery of precise, effective prescriptions to the farmer nearly impossible. A powerful new approach is introduced, which is the Spatio-Temporal Fusion Network. This deep learning architecture is engineered to instantly harmonize all these competing data sources to deliver clear, highly granular recommendations for nutrient prescriptions and optimal crop selection. The model operates by breaking down the data in parallel using a Vision Transformer (ViT), which analyzes static spatial patterns from satellite images, while a Long Short-Term Memory (LSTM) network tracks the dynamic temporal history of nutrient and climate data. These two distinct perspectives are then instantly merged by a Transformer Fusion Engine using crossmodal attention, a mechanism that ensures the spatial and temporal information are coherently integrated. This unified approach significantly outperforms older, single-source methods. Such accuracy in nutrient targeting and determination of suitability is the much-needed reliable intelligence to reduce waste, save the environment, and control fields effectively.

Fields of Study & Subjects

Smart Agriculture and AI Remote Sensing in Agriculture Agricultural Science and Fertilization

Publication details

Journal / source
International Conference on Communication, Computing and Electronics Systems (ICCCES)
Pages
1322–1328
Publication date
21 January 2026
Language
EN