Scholarlink

<p>Effective crop yield prediction is critical in an effort of maximizing the use of irrigation, resource distribution and sustainable agricultural planning. Nonetheless, traditional methods tend to lack the ability to reflect temporal variations, agro-environmental complex interdependencies, and dynamism to the field conditions. In order to overcome these issues, this paper develops a novel Reinforcement Learning-Enhanced Clustering and Contrastive Decision Optimization in Agriculture framework (RLCCDOA). This is aimed at enhancing the idea of prediction of crop yield, the efficiency of irrigation, and the use of water resources in agricultural IoT settings. Three important components may be found in the proposed approach. To start with, Context-Aware Temporal-Adaptive Clustering Generative Adversarial Networks (CATACGAN) learn to organize multi-source agricultural data into time-sensitive clusters, which are useful to capture crop-related and environmental dynamics. Second, Contrastive Adaptive Multi-Channel Graph Attention Networks (CAMC-GAT) model interactions among the soil, climate, and crop factors through contrastive learning to allow a better discriminatory ability of optimal and suboptimal yield conditions. Third, Hierarchical Action Space Proximal Policy Optimization (HAS-PPO) allows real-time multi-level decision-making, so that the irrigation schedule and resource allocation can be optimized using adaptive policies depending on real-time sensor measurements. It is shown in experimental results that predictive performance is improved, with 99.4% accuracy, 99.3% precision, 99.8% recall and 99.5% F1-score on Dataset I, and 98.7% accuracy, 99.1% precision, 98.4% recall, and 98.75% F1-score on Dataset II, which is much better than available methods. Ablation and statistical tests also confirm the efficiency, strength, and convergence efficiency of the proposed strategy. The RLCCDOA model can be broadly discussed as a scalable and flexible approach to precision agriculture, and the model has plenty of implications on the optimization of irrigation, the better management of water resources, and promoting sustainable agriculture in dynamic environmental environments.</p>

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

Optimizing Crop Yield Prediction and Suitability using Reinforcement Learning and Context Aware Multi-Channel Networks in IoT-Enabled Smart Agriculture

  • 期刊论文

  • 发表日期：

    2026年4月24日

查看全文
  • J. Jencewin,
  • M. R. Geetha

摘要

Effective crop yield prediction is critical in an effort of maximizing the use of irrigation, resource distribution and sustainable agricultural planning. Nonetheless, traditional methods tend to lack the ability to reflect temporal variations, agro-environmental complex interdependencies, and dynamism to the field conditions. In order to overcome these issues, this paper develops a novel Reinforcement Learning-Enhanced Clustering and Contrastive Decision Optimization in Agriculture framework (RLCCDOA). This is aimed at enhancing the idea of prediction of crop yield, the efficiency of irrigation, and the use of water resources in agricultural IoT settings. Three important components may be found in the proposed approach. To start with, Context-Aware Temporal-Adaptive Clustering Generative Adversarial Networks (CATACGAN) learn to organize multi-source agricultural data into time-sensitive clusters, which are useful to capture crop-related and environmental dynamics. Second, Contrastive Adaptive Multi-Channel Graph Attention Networks (CAMC-GAT) model interactions among the soil, climate, and crop factors through contrastive learning to allow a better discriminatory ability of optimal and suboptimal yield conditions. Third, Hierarchical Action Space Proximal Policy Optimization (HAS-PPO) allows real-time multi-level decision-making, so that the irrigation schedule and resource allocation can be optimized using adaptive policies depending on real-time sensor measurements. It is shown in experimental results that predictive performance is improved, with 99.4% accuracy, 99.3% precision, 99.8% recall and 99.5% F1-score on Dataset I, and 98.7% accuracy, 99.1% precision, 98.4% recall, and 98.75% F1-score on Dataset II, which is much better than available methods. Ablation and statistical tests also confirm the efficiency, strength, and convergence efficiency of the proposed strategy. The RLCCDOA model can be broadly discussed as a scalable and flexible approach to precision agriculture, and the model has plenty of implications on the optimization of irrigation, the better management of water resources, and promoting sustainable agriculture in dynamic environmental environments.