Scholarlink

Context: The intersection of Internet of Things (IoT) and Artificial Intelligence (AI) has been responsible for AIoT systems, transforming precision agriculture. However, unimodal approaches to medicinal plant farming lack the non-linear interaction between abiotic stresses and the synthesis of bioactive metabolites, leading to low-quality forecasts and the failure of field scalability. Objectives: This research will develop a strong multimodal AIoT platform to facilitate improved prediction and estimation of the growth stage of bioactive compounds in medicinal plants under dynamic environmental conditions, and will be able to be deployed in smallholder contexts. Methods: A new architecture was introduced that integrates hyperspectral aerial images, IoT sensor-based environmental monitoring, and genomic and metabolic information. Early and late fusion were combined within a Dynamic Weighted Fusion (DWF) framework supplemented with wavelet noise removal and edge inference modules incorporating attention. Federated learning provided privacy-preserving distributed model training. Results: The proposed system in the paper recorded a prediction accuracy of 92. 9%, 0.058 RMSE, 98 ms latency in inference, and a reduction of 40. 1% error rate and dominated unimodal CNN and LSTM-Sensor Fusion baselines for all four metrics. Conclusions: The AIoT system is highly resilient, accurate, and scalable for medicated agriculture, integrating bioinformatics with climate-resilient agriculture through edge-enabled, privacy-preserving deployment techniques.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Multimodal AIoT Framework with Dynamic Fusion and Federated Learning for Smart Medicinal Plant Farming

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • R. Roja,
  • P. K. Udayaprasad,
  • P. Amulya,
  • S. Pramila,
  • M. Usha Rani,
  • Shridhar B. Devamane

摘要

Context: The intersection of Internet of Things (IoT) and Artificial Intelligence (AI) has been responsible for AIoT systems, transforming precision agriculture. However, unimodal approaches to medicinal plant farming lack the non-linear interaction between abiotic stresses and the synthesis of bioactive metabolites, leading to low-quality forecasts and the failure of field scalability. Objectives: This research will develop a strong multimodal AIoT platform to facilitate improved prediction and estimation of the growth stage of bioactive compounds in medicinal plants under dynamic environmental conditions, and will be able to be deployed in smallholder contexts. Methods: A new architecture was introduced that integrates hyperspectral aerial images, IoT sensor-based environmental monitoring, and genomic and metabolic information. Early and late fusion were combined within a Dynamic Weighted Fusion (DWF) framework supplemented with wavelet noise removal and edge inference modules incorporating attention. Federated learning provided privacy-preserving distributed model training. Results: The proposed system in the paper recorded a prediction accuracy of 92. 9%, 0.058 RMSE, 98 ms latency in inference, and a reduction of 40. 1% error rate and dominated unimodal CNN and LSTM-Sensor Fusion baselines for all four metrics. Conclusions: The AIoT system is highly resilient, accurate, and scalable for medicated agriculture, integrating bioinformatics with climate-resilient agriculture through edge-enabled, privacy-preserving deployment techniques.