Scholarlink

<p>Accurate crop yield prediction is essential for sustainable agricultural planning, but traditional models often struggle to capture complex interactions among climate, soil, and crop factors. Conventional approaches such as Linear Regression, Random Forest, and standard Convolutional Neural Networks (CNNs) are limited by static assumptions and small datasets, restricting predictive performance. This study proposes a deep learning framework that integrates multispectral satellite imagery with environmental variables to improve crop yield prediction and support data-driven decision-making. Conditional Generative Adversarial Networks (cGANs) are used to generate synthetic training data, enhancing model robustness, while Differentiable Architecture Search (DARTS) optimizes the neural network structure. Recursive Feature Elimination (RFE) identifies the most informative variables, and vegetation indices such as NDVI are extracted to ensure high-quality inputs. The model was evaluated on historical crop yield datasets spanning multiple seasons and regions, using Linear Regression, Random Forest, and standard CNNs as baselines. Results show that the proposed model reduces Mean Absolute Error (MAE) by 23%, Root Mean Squared Error (RMSE) by 19%, and increases R² by 17% relative to these baselines. Simulation-based assessments, grounded in representative crop management scenarios, indicate that AI-informed strategies could increase yields by approximately 27% (± 3.4%) and reduce resource costs by 22% (± 2.7%) compared to conventional practices, demonstrating the practical benefits of predictive analytics. This framework is integrated into a decision support system providing actionable recommendations for irrigation scheduling, fertilization, and resource allocation. Future work will incorporate climate forecasts, soil health metrics, additional crop types, and real-time automation to further enhance predictive accuracy and operational efficiency.</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

Deep Learning and Remote Sensing for Crop Yield Prediction and Decision Support

  • 期刊论文

  • 发表日期：

    2025年12月26日

查看全文
  • Shriya Sahu,
  • Priyank Jain

摘要

Accurate crop yield prediction is essential for sustainable agricultural planning, but traditional models often struggle to capture complex interactions among climate, soil, and crop factors. Conventional approaches such as Linear Regression, Random Forest, and standard Convolutional Neural Networks (CNNs) are limited by static assumptions and small datasets, restricting predictive performance. This study proposes a deep learning framework that integrates multispectral satellite imagery with environmental variables to improve crop yield prediction and support data-driven decision-making. Conditional Generative Adversarial Networks (cGANs) are used to generate synthetic training data, enhancing model robustness, while Differentiable Architecture Search (DARTS) optimizes the neural network structure. Recursive Feature Elimination (RFE) identifies the most informative variables, and vegetation indices such as NDVI are extracted to ensure high-quality inputs. The model was evaluated on historical crop yield datasets spanning multiple seasons and regions, using Linear Regression, Random Forest, and standard CNNs as baselines. Results show that the proposed model reduces Mean Absolute Error (MAE) by 23%, Root Mean Squared Error (RMSE) by 19%, and increases R² by 17% relative to these baselines. Simulation-based assessments, grounded in representative crop management scenarios, indicate that AI-informed strategies could increase yields by approximately 27% (± 3.4%) and reduce resource costs by 22% (± 2.7%) compared to conventional practices, demonstrating the practical benefits of predictive analytics. This framework is integrated into a decision support system providing actionable recommendations for irrigation scheduling, fertilization, and resource allocation. Future work will incorporate climate forecasts, soil health metrics, additional crop types, and real-time automation to further enhance predictive accuracy and operational efficiency.